🎙️Listen 🎙️

Maybe I’m being a stickler on this but I don’t like when companies say they’re AI-native and they’ve been around for 100+ years. I feel their ambition is to be AI-first - or engineering first!. AI native should be reserved for companies born in the AI era or indeed new products/capabilities built for this era.

Here Travis Hedge Co-Foudner and CEO of Vouch Insurance talks about AI Native being about the culture - which I totally get and agree with. Before we do, we have to think! Links to the Allie K. Miller post below too - mindset matters!

Here he joins Kyle Nakatsuji from Dearborn Labs Context Window Podcast to talk though more - from how AI Should Industrialize Expertise, Not Replace Humans, how and why vouch was formed (mitigating critical blind spots that stall fundraising, contract execution, and scaling) through to how speed to service drives value! Full podcast here

A massive percentage (around 40%) of what clients truly value in tech-forward insurance is raw speed and seamless service delivery. Combining deep industry domain knowledge with modern digital execution allows platforms to give early-stage startups enterprise-grade guidance long before they are massive companies.

This from an old friend Gil Brodnitz who shared the Ezra Klein show - The A.I.s Are Already Out of Control, joined by Helen Toner, Executive Director for Georgetown’s Center for Security and Emerging Technology and former OpenAI board member (remember when Sam was removed).

The technology is evolving faster than our ability to govern or reliably predict its behavior.

Hacking events have gone from science-fiction to science-fact! with more than a couple of recent examples. These aren’t going to slow down any time soon

The headline for this week is “Speed Kills.” Matt Connolly And I were just talking about this - what I mean specifically is that push back on Data Centers. Public Sentiment Has Turned Radically and Rapidly. This is no longer a technology debate, but very much a political one too - with recent switches in allegiance from pro to anti data center in exchange for votes.

The Backlash is Rooted in Big Tech Mistrust and Poor Messaging:

More the latter, I feel, and the speed at which the AI world has moved and is moving is causing half the challenge. Maybe more. The polls say the same - public sentiment has dramatically shifted when it comes to AI.

• 75% of Americans opposed to having a data center built near their home (up from 51% earlier in the year). I recall the same upheaval and debate in my local community around both a solar farm and a data centre that have since been rejected and approved, respectively

• 61% expressing strong opposition (surging from 24% over a 12-month period).

• 78% believing data centers place too much strain on local power grids, and 75% believing they consume excessive water supplies.

I think the The AI Daily Brief has done a really nice job summarising all the things that are causing challenge, but also all the positive news stories from cities and states benefiting from data centres over the last 20+ years. We just don’t see enough of them - lowering property taxes or building new sports facilities isn’t as newsworthy as a water or energy costs changing. Equally, the big tech leaders who used to message that AI was taking 50% of white-collar jobs, may perhaps have also contributed to this?

In every revolution, be it the industrial or otherwise, there is always going to be an element of backlash as we go through that period of change The era of “move fast and break things” might just be impacting people’s trust at this moment in time while we sort out the misinformation and benefits.

📺 Watch 📺

20,000+ people turned up to tune in with Allie K. Miller and Mark Cuban for her AI Agent Workshop. I don’t want to spoil it - but well worth watching the replay here. Her key point, winners are the ones that take action! while the difference is mindset, not the tools (See Travis Hedge‘s point above too).

Honestly, I need to re-watch it again!. The feedback I’ve seen online from folks that have tuned in, watched and re-watched this is also just brilliant. Thank you Allie K. Miller.

Probably one of the biggest questions there is right now for every enterprise and organisation. Here Andy McCall, General Partner and Joe Schmidt IV Partner at Andreessen Horowitz sit down to talk about the two very different ways AI startups can go to market: the lighthouse and the landgrab.

• The Lighthouse playbook focuses on high-stakes, regulated industries (like legal or insurance) where big-name logos provide critical social proof and mitigate buyer exposure.

• The Land Grab playbook targets the mid-market with solutions that replace existing workflows, relying on clear math and immediate utility rather than heavy educational missionary work

Ultimately as they point out, you do eventually end up doing both . Above all, (and no surprise), execution matters most. What’s the old saying - good is better than perfect? Perfect is the enemy of enemy of good enough. Good, solid advice all round! Full article here

I enjoyed Jon Gray‘s take on last weeks announcement that I covered last week to - NVIDIA plus many of the financial institution’s enabling chips to be their own asset class for the first time.

And finally, what does the sale of the LA Lakers have to do with Insurance?. Watch an explainer video below - and full article here. Just wild indeed to see how all of this is linked.

It’s TWG Global, Mark Walter’s holding company where he keeps his favorite assets — ranging from cool stuff like sports teams to boring stuff like a handful of insurance companies whose money Guggenheim manages.

📚 Read 📚

Further potential fallout from TWG Global - Cadillac F1 owner denies sale talk as the Insurance Probe widens. My 2c, I don’t think we have seen the end of this yet, or the wider implications for the market.

Formula 1 team through TWG Motorsports - officially shut down rumors of a sale, asserting that neither the Cadillac F1 team nor any of its other motorsports interests are on the market.

Beyond all this, as Gia Snape shared - it raises further questions and a wider industry debate around insurers’ growing use of private credit and the potential conflicts created when an insurer invests in assets connected to companies under common ownership.

Quite a headline from the Financial Times last week.

AI hasn’t gone rogue. It’s worse than that!

On the back of recent events (hacking etc), Madhumita Murgia highlights that these aren’t going rogue - more just reflect the technology is operating precisely as it was trained to do under given conditions! (Capture the flag etc)

The offensive capabilities we have reached today, we have reached deliberately,” says Boyan Milanov, senior research scientist at the independent New York-based AI Now Institute, who studies the security risks associated with AI agents.

We saw what happened when the Crowdstrike incident happened a while back and the time it took to play catch up on this across the world. The same for the NHS systems post a cyber incident.

Our bigger issue now is more of a practical problem - real-world guardrails and security protocols are falling short against current deployments. Quite simply - Are current guardrails and government capabilities simply inadequate given the speed at which all this is moving ? Some say so!

While still somewhat skeptical about the number of people at least using this in the early stages using this end to end - I do believe it’s coming, it is also good to see it continue to expand - here Insurify has updated their ChatGPT plugin to do more. distribution may have shifted, but I suspect this is more of a threat to the comparison sites than the insurers themselves.

Eligible shoppers can answer a few questions about themselves, their vehicle, and the coverage they want, then compare quotes from leading insurance companies directly in the chat. They can also ask insurance questions along the way and take the next step toward purchasing coverage with Insurify.

There is no doubt, there are huge questions looming ahead for humanity, especially when it comes to AI. Daniel Hulme, Global Chief AI Officer over at WPP highlights a great piece from The Economist - Could AI become Conscious? - with a book on the subject coming in September from Conscium . Elsewhere, Stripe have said we are reached singularity in January, Demis Hassabis has said we were at the foothills of AGI back at Google IO. In Daniels words:

I have come to think the more tractable concern is what moral risk we run by assuming a system is not conscious.

Back to my old consulting world for a minute, I found this interesting - but not surprising one bit. I actually think it’s a great time to be in consulting right now, given the amount of both business change and technology change going on. Organisations across every industry are going to need help framing, training, and deploying this responsibly across their customer base

Bernhard Huber compared them all originally here. 10 Consulting firms, 10 framings. as with any consulting engagement, it’s choosing the right firm that’s going to challenge you appropriately for the task at hand, whether that’s technology execution, business change management, or operating model. of course, what’s different here is the start point, context, inputs and then finally the ambition.

This headline is as true for RPA, ERP, Core System Modernization as it is AI. Create a repeatable framework and get it to market as effectively as possible, update regularly. We shouldn’t be surprised. The good news is, of course, they see a lot more implementations, both good and bad, than many people have in a single organisation. The goal: harness and learn from that as quickly as you can

👀 Quick bites 👀

1. Instagram and Facebook could change forever if Meta loses child privacy trial - as a parent, continuing to watch these play out and the record fines being paid.

2. Governor Hochul Announces WithCoverage to Expand in New York City, Creating 205 New Jobs - WithCoverage - congrats Max Brenner

3. An AirTag in a rare book led to an Amazon book scanning warehouse in Las Vegas

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! Grok, Data Centers and Climate Risk! find me an episode where we’re not laughing and smiling. Always good fun!

Unfiltered with Matt & Nige

Grok, Data Centres, and Climate Risk

Unfiltered with Matt & Nige | Episode 20 🎙️…

Listen now

14 days ago · Matt Connolly and Nigel Walsh

And that’s It! See you next week! 🚀

🎙️Listen 🎙️

This one shocked me, and to be fair many others - Airtable sold for $1.3bn, previously valued at $11bn! back in 2021 - purchased by Bending Spoons as part of their continuous roll up. Equally of note was their comment of potential founder fatigue setting in - after a decade of growth, pivoting and now the shift needed with the advent of AI. Airtable also always felt more technical than something like Notion for example - now of course much of which folks have more control to build for themselves rather than use a low-code platform.

This week, Nikesh Arora, Chairman and CEO of Palo Alto Networks joins Harry Stebbings, Rory O’Driscoll and Jason M. Lemkin for a fun conversation. Beyond Airtable, they discuss the Importance of Organizational Context over Raw Models: As foundational models mature, the differentiator for enterprises won’t just be the model itself, but the domain-specific context, data ingestion, and proprietary guardrails companies build around it. I think we have talked about this for a while.

At what point does frontier model capability become commoditized? and the value shifts to the edge?

📺 Watch 📺

This week, NVIDIA partnered with six major global asset managers — BlackRock, Blackstone, Apollo Global Management, Inc., Brookfield, KKR and Goldman Sachs to establish dedicated capital pools totaling over half a trillion dollars. The goal - lowering the massive upfront capital barrier for companies and sovereign nations looking to purchase Nvidia’s high-cost GPU infrastructure.

Jensen Huang and analysts frame compute infrastructure not merely as technology hardware, but as a foundational utility—akin to electricity or water

Given the depreciation of traditional hardware, this has always been a challenge - at least up until now. With improvements in software, those legacy chips are also now lasting longer/going further. This week’s announcement between NVIDIA and the financial giant’s (or should I say memorandum of understanding) brings a whole new view on this Below Ed Elson chats with Jay Goldberg form Seaport Global Securities LLC, Research Division for his take.

It of course raises questions again about the whole circular economy and how each of these is funding each other. That said the whole goal of organisations should be to unlock barriers to access their services - and here Nvidia are doing just that. In the same way that Uber provided loans for people to buy cars so they can get more Uber drivers on the road.

I saw Memorandum of Understanding as recently, Nvida signed something similar with OpenAI for $250bn, and less than a month later have scaled it back - more here from the The Wall Street Journal

What do I mean by circular economy, well Ed Elson is back at it here with all the recent departures at Open Al - the most recent, Denise Holland Dresser, CRO after just 8 months.

Anyone who knows me, know’s I love a good Insurance Ad. This from Jonathan Kerr (JK) and team at Budget Direct, part of Auto & General Australia is good fun and humorous. Part of the Aussie pschye if you ask me! Across the world, everything from Meercats to Lizards to Emu’s - Insurance has it all! See JK’s original post here. This now in its eighth year.

Maybe I should add a feature here to add my favourite insurance ad of the week, or indeed yours! Comment below with your favourites, or indeed if you want this included more regularly! They’ve always fascinated me by how they’re so different around the world in a segment that craves and needs a positive link to customers.

📚 Read 📚

The 2nd August 2026 was a big day in Europe - especially when it comes to AI models. for me, the EU represents one of the leading regulators, and often you see other jurisdictions follow suit

EU AI Act rules on AI models become enforceable today, making Brussels the world’s top AI regulator.

Importantly, The European Commission’s AI Office and national authorities now possess official regulatory teeth to actively oversee, investigate, and penalize developers of general-purpose AI (GPAI) models.

Companies that fail to comply face severe financial penalties, including fines up to €15 million or 3% of their global annual turnover.

Thanks to Paige Bailey over at Google for highlighting this one earlier this week here. What a cool use! And also supports how we’re going to get more and more use out of the chips per G2 love. Will talk about that last week. This directly from Paige

They borrowed Per-Layer Embeddings from our Google DeepMind Gemma models: instead of loading the whole network into memory, ~25M of the parameters live in flash as a lookup table, and only the ~450 bytes needed for the current token get pulled in at inference time. The heavy compute stays in SRAM/PSRAM. Full model: 14.9MB after 4-bit quantization. Output: ~9.5 tokens/sec straight to a display, no cloud in the loop.

Here’s an interesting take from Praveen Neppalli Naga, CTO at Uber on X. Traditional Endpoint Detection & Response (EDR) sees the file write, but not the prompt that triggered it. It sees the network call, but not the agent’s reasoning. The intent, the thing that separates malicious from benign, is invisible.

Today, our agents run 50,000+ sessions per day across thousands of endpoints. And this isn’t just engineering anymore. Employees across the company use agents that read code, run commands, call internal tools, analyze data, and act on real systems.

He also calls out the issue with guardrail’s at an individual isolated level, the importance to me of a wholistic end to end control tower, regardless of where or who is operating the agent - be it direct or a 3rd party.

Security cannot focus on individual tool calls because attacks often hide within causally linked chains of actions that appear benign when viewed in isolation.

Interesting given recent debates around this - Uber has also open-sourced their ADR system (the sensor, detection framework, and a benchmark called ADR-Bench) - having already seen significant success internally, eg blocking 97.2% of credential leaks before they leave employee laptops.

Many of you will have seen my love of Wispr Flow, well 10x that when you combine it with Ironman. Thanks Eden Britt for highlighting this one - Wispr Flor AI Lab is doing just that and bringing JARVIS to life! Hiring former Amazon and Meta‘s Ariya Rastrow as their Chief Scientist. Wispr Flow have been adding a tonne of features recently which is starting to make thinks - lots of these services will start to converge, from Granola, Plaud, Zoom and many more.

This has been talked about a lot, I recall the startup acquired by WIX with just 3 employees, or the guy who created OpenClaw and now works for Open AI. New tools and capabilities are giving everyday folks (Not coders) the ability to vibe code new applications and businesses, while outsourcing other functions and more.

While not impossible, I do wonder how these differ in the B2C and B2B worlds. Would it pass the risk register of a regulated entity offering services to customers or the bank/insurer? Does it have the right security, governance, controls and back up plans/failover in place?

Don’t get me wrong - I love the idea and opportunity of an AI cofouder. I already believe we are entering a world of personal software - I have some great personal examples which I’ve built for myself for Financial Advisor/Insights and Health Advisors - bringing together oodles of data from multiple sources and more. However you look at it - the barrier to entry continue to fall - exciting times all round!

Ah, this did make me laugh. All of a sudden everyone is an AI expert - reminds me of the folks that have been doing FinTech for 20 years. Here we can now look back to see how Job Title, Responsibilities and much more have changed - after folks have left the role. In the same way many of the grifters on the blockchain bandwagon moved to pushing AI, whatever next... Quantum experts?

Instead, they’ve used cached data to monitor how LinkedIn users “retroactively edit the title or description of a job they have already left”.

LinkedIn can still be weird! For the record, my first job was always - PJ Shoes in Luton as a Saturday Job.

Great to see this from Allstate - C3 (Connected Customer Cloud).

With our orchestration layer, we don’t need to get rid of old technology.

More and more insurers have created a thoughtful curated approach in scaling AI across the enterprise, see examples from Travelers, Allianz, Brown & Brown and many more. The era of allowing 1,000 flowers bloom has passed and now the focus is on creating the roads and rails for scale, adoption and safety.

“We have a technology-driven strategy, not a strategy supported by technology,” CEO Tom Wilson said.

“The architecture has eight integrated components that would enable agent-to-agent processing,” Wilson said. “For example, one component will handle all customer interactions. Each component is comprised of multiple agentic agents, which are built to be reused across the enterprise.”

This is becoming a bigger and more prominent risk with folks like Aon increasing capacity here to $5bn for Data Centers Recently. What it does highlight is the complexity of the value chain, from the above news from Jensen around GPU’s as an investable asset class to then managing them long term.

• Severity vs. Frequency: fire accounts for well over 50% of loss value (analyzed across roughly €700M / ~$800M in total claims), making it the single largest driver of catastrophic loss severity.

• Water & Interruption: By volume, water damage is the most frequent cause of claims (21%), followed by willful/cyber acts (19%) and fire (14%). Across all lines, Business Interruption (BI) remains the primary driver of financial impact due to the severe cost of operational downtime.

Loads more insights here too, not least the trust issue - more and more states are rallying against new Data Centers with 30 states now banning them. Worth a read.

As my kids look to the workforce in a few years times Oscar Walsh!, it’s interesting to see that evolution of predictable careers. In fact I’ve believed the ‘job for life’ has long gone - university teaches you about the ability to learn and adapt and re-learn. Here Lynda Gratton writes for Harvard Business Review on how this Career Arc is over. Lynda has a number of key themes here. I’ve grouped them here, but key is prepare before you need to and commit to what lasts (ie compound knowledge).

• Portable skills: Core technical fluency, problem-solving, and adaptability.

• Relational capital: High-trust networks across industries, not just within a single firm.

• Vitality & curiosity: Physical and mental stamina to sustain multi-stage career pivots.

Thanks to Bryan Falchuk, PACE-P for this (and Lee Brooke-Pearce for highlighting!) I actually remember one of the very first iTC’s - Brian Duperreault was on stage and the question was put - in this InsurTech age, will there be agents/brokers in the future. I remember his answer to this day, 10 years on - he wasn’t wrong.

Over the past 10 years, the share of U.S. property/casualty premium written through intermediaries has held almost constant at 57-58%

Fast forward a decade and replace InsurTech with AI and the same question has arisen. I think like all of these - all roles evolve - as they did when we went from typewriter to PC. We are still in the most part extensibly doing the same tasks, some are automated, enriched and some still need/benefit from that human touch.

On this, I want to build on Bryans point - the UK market for example is very different with the advent of Price Comparisons Websites (clue is in the title) - I’d still argue people don’t know what they are really buying, or have compared the wording. It’s different in the USA, but the common theme here, these are not daily or frequent interactions - so we need that expert help and guidance. Where and how we get it, might indeed change! Bryan Falchuk, PACE-P sum’s it up perfectly. Worth a read!

The agent channel will still be here. Many of today’s agents, and the way they work, will not be—and we will not need a decade to see it.

👀 Quick bites 👀

1. Anthropic CEO Says His Employees Are a Bunch of Untrustworthy Rats

2. Insurer tells driver to remove ‘Jesus Loves You’ stickers or lose cover, weeks after approving them

3. Are AI recruitment tools affecting mid-life women’s careers?

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! From voice agents to data centres and AI exclusions! It’s funny, Matt and I were recording this week’s show earlier and talking about wildfire risk changing - given some of the extreme heat events in the UK . The next minute, we see more new news coverage on this - either we have a crystal ball, or we know what we’re doing.

Now on week 20 of this, it’s a genuine highlight of my week, I get to hang out with a great friend and check in on what we are seeing in the market together. See Matt’s summary here too. Latest episode below.

Unfiltered with Matt & Nige

Voice Agents, Data Centres, and AI Exclusions

Unfiltered with Matt & Nige | Episode 19 🎙️…

Listen now

21 days ago · Matt Connolly and Nigel Walsh

And that’s It! See you next week! 🚀

🎙️Listen 🎙️

This week, Mark Geoghegan, The Voice of Insurance along with Aon Clyde Bernstein - Placement Technology & Trading Analytics Leader with everything from DPX, the broker’s new Digital Placement Exchange placing system, designed to automate the placing of following lines. Similar to what Ki pioneered a few years back, everyone expected to fail, but now everyone wants one. Cancel that, needs one.

Importantly, whilst Clyde leads placement technology, he’s a broker at heart, so he’s been there and seen what the business needs, rather than a pure tech version of the story. Empathy for the role and tasks involved is critical.

I’ve shared before I’m a Hannah Fry fan, I’m not the only one. Here Helen Tupper, co-founder and CEO of Squiggly Careers sits down for a chat on her career and all things AI. Everything from the importance of Critical Thinking (You will have heard me talk about creativity, curiosity, and storytelling for too long now) to Developing skills that will guard against de-skilling - I think more important for our generation, then the next.

I always remember my dad talking about new technology and how we adapted it quicker than he did with his line, “You grew up with it. It’s easier for you.” I now see the same in my teenage kids - often wondering how they did something so quickly. I don’t quite need them yet to sort the printer out. Mind you, I doubt they know what that is.

Helen’s summary here, full episode below. The last bit was probably my favourite. - I handy remind it to myself too!

Knowing when to disconnect is just as important as knowing how to use technology. True creativity, empathy, and meaningful connections come from stepping away from screens and engaging with the real world

📺 Watch 📺

OK, this will be covered in a number of places this week, 24 year old Leo Aschenbrenner - margin calls. Harry Stebbings sum’s it up in a quick YouTube short.

Despite growing a $225 million fund into over $20 billion in just two and a half years, Aschenbrenner’s fund lost 35% in a single month during a broader AI stock selloff

Allowing, Citadel and Ken Griffin to swoop in and buy the lot. don’t worry, a significant number of private assets survived. I was more concerned about all this happened the week he was to get married ( to none other than Anthropic’s CEO, Chief of Staff) - as if getting married isn’t stressful enough! The bigger question of course is:

is this the start of an AI bubble or just a blip along the way!

Deirdre Bosa, CNBC is joined by Jeetu Patel, President and CEO of Cisco, Lin Qiao, CEO and Co-Founder of Fireworks AI and Evan Beard, Co-Founder and CEO of Standard Bots.

Jeetu closes his part with a really good challenge (one for us Alexis Bliek Anna Mazzone)

as AI agents become more autonomous, security and observability must fuse together. It is becoming increasingly difficult to distinguish between:

They also touch of the current raging debate of Open Source, and whether by not allowing it in the US - the cost of Intelligence in the USA is higher than elsewhere in the world, namely China! where many of the new models are emerging, including the most recent Moonshot - Kimmi K3.

Equally, from a robotics perspective, China deployed 10x more robots than the US did last year, 300,000 vs 30,000. I’m always amazed by their scale and pace and love the stat that, in the same time it’s taken us to debate a third runway for Heathrow, they’ve built 30 international airports.

Speaking of control problems and the need for governance and controls - bill oliver a member of the Legislative Assembly (MLA) of New Brunswick, Canada. As Fast Company called out here

The sudden digression was clearly an AI-generated instruction that was inadvertently left in. The politician did not seem to notice, continuing to read aloud without hesitating or stopping.

If nothing more, this calls for humans in the loop still and he and his team checking in advance! That said, I’m also on the other side of the coin - I’m pleased our politicians around the world are using this. It also goes to show you we’re all learning together. as an aside, what is it about Canada? First Air Canada, then a BMW dealership, now this? Maybe they are just leading edge

📚 Read 📚

The story just gets more and more wild. And many of the tech podcast and outlets focussed on this, more with regards to the impending bubble, if any, and the leverage that he used to get the fund to grow so much in the last couple of years. This is a good read from Berber J. and the team at the The Wall Street Journal.

In a new report from OpenAI, the the theme of expanding more capability with the same number of people is backed up with data from their customers. In a report titled, ‘How AI is expanding what people do at work’. Three key takeaways from Sarah Friar recent post here. Full report below.

• Financial calculation and technology troubleshooting were among the top three crossover tasks in every other occupation group studied.

• Marketing work extended across five other fields, especially design.

• AI Acts as a Generalist Force, Especially in Smaller Businesses

In a challenge to embedded insurance, the Financial Conduct Authority in the UK has written to insurers this week around conflicts of interest concerns for vertically integrated propositions. .

Its new director of insurance, Chris Knight, Director of Insurance, said that “one item that has crossed my desk is vertically integrated business models” and the conflicts they can create.

Importantly, he does say - “having a conflict of interest doesn’t automatically make a business model unacceptable, but you need to take these risks seriously”. The Insurance Times reports below and official statement here.

The FCA emphasized that simply disclosing conflicts of interest to policyholders does not satisfy regulatory requirements. Firms must actively manage, govern, and evidence that their structures deliver good outcomes for customers.

Speaking of which, will NatWest be in hot water following their recent partnership with Uinsure. The aim, “let customers buy home insurance in under 60 seconds” through the bank’s app, website, phone line or branches, with a panel of insurers competing on price for the same Defaqto five-star rated cover. Sounds very familiar to what Amazon tried a few years back and subsequently shut down. Responding to Insurance Business MAgazine, Toni Powling, head of insurance at NatWest said:

The partnership was designed to tackle customer inertia around home insurance rather than replace advice.

Bringing two of my passions together: financial services and AI - I love the idea of this. I think everything will be conversational in the future. It’s just the next interface that people already have - then we’ve got to work at how we actually engage, whether it’s a keyboard or your voice. I’m feeling more and more voice is my path forward with Wispr Flow, growing stronger every day (closing in on 200,000 words now!)

In the same way I can now chat with my WHOOP data using their OpenAI integration. Note for the record that the very first one was totally rubbish and felt pointless but I persevered because I had the 2,000 day data streak and the advice. The new AI, I have to say, is getting much much better! As an interesting aside, I wonder if having ChatGPT or similar attached to your financial services actually drives stickiness and loyalty to your financial institution because they have all your data?

I expect the same with my money, my finances, my pension, and so much more. The bit most of these have ever seen, of course, is the connective tissue between all of them. What I built with my health was taking my WHOOP, my FitnessPal, my weighing scales, my food, the supplements I take, and so much more and then using Claude to join all the dots and find the gaps. That was brilliant. We need the same for financial services! I don’t expect it far away.

Security will never be far away - especially as AI rolls out to millions and millions of users around the world. We’ve seen this with other models in the past. This time, Anthropic‘s Claude in the spotlight. First highlighted by Allie K. Miller here, Madeleine Varner at WIRED have shared a piece on how many links shared with public access are accessible via Google and Bing searches. Another reminder to us all!

One downside of AI increasing is the velocity of things people can produce from newsletters (mine’s all manual, including all my own smelling mistakes! - although I am trying out AI agent to summarise the newsletter for my 3 things you missed on Wednesday post), Video’s LinkedIn articles, Substack posts and much more.

This week LinkedIn also added a ‘report AI slop’ button - more here from TechCrunch. Hari Srinivasan, Chief Product Officer shared a post on how they are tackling it here. Snap Inc. Substack and many others joined in.

Snap, the parent company of Snapchat, said on Friday that the platform would stop recommending “wholly AI-generated videos” in its popular Spotlight feed in favour of “authentic, human-made content.”

I have to say, I actually don’t mind some of them. I would rather they just be called out specifically if they are AI-generated, AI-edited, or authentic. I do mind lazy journalism, or many of the LinkedIn posts with Claude assets, saying I’ve found the trillion-dollar insurance industry, and then did a very lazy job of: a) understanding the industry and b) reporting on it and add in critical comment.

Finally, it feels like no one’s safe, this week PwC was the latest of the Big Four to be highlighted for their use of AI in reports, TNW reports here.

PwC published four Middle East reports stuffed with fake footnotes and made-up sources, the telltale signs of AI hallucination, and one footnote URL literally ended in “chatgpt.com.”

👀 Quick bites 👀

1. PwC published reports on AI marred by AI hallucinations

2. These Shoes Are Everywhere. Can On Topple Nike? - innovation from On

3. Lyft and Baidu enter London’s robotaxi battleground as testing begins - they are coming Matteo Carbone

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! 18 episodes in and we’re still laughing!

Unfiltered with Matt & Nige

Beyond the Pilot

Unfiltered with Matt & Nige | Episode 18 🎙️…

Listen now

a month ago · Matt Connolly and Nigel Walsh

And that’s It! See you in two weeks! 🚀

🎙️Listen 🎙️

First up, a good discussion with Thomas Kuhnt, in a surprise to few rolling AI across the enterprise, he shares that AI is clearly a business initiative, not a technology one and to that, Business leaders must own this, not technology. I’d probably update to say it’s a partnership between the two. We saw last year how Moderna merged HR and IT to enable people as well.

We’ve got to get used to a brand new operating model to accelerate learning and embed change. His next quote always reminds you about outsourcing that I had from a client, probably 20+ years ago, in that outsourcing a broken process just meant it was a cheaper broken process. The same is true if you apply technology to the same broken process, and which is why I’m upbeat about ServiceNow and AI in the flow of work, not just bolted on as many do.

Applying AI to inefficient or broken workflows simply “accelerates the breaking.”

Equally his point about automating non-core tasks to elevate human expertise is critical. We also share our mantra of “think big, start small, act quickly.” and ultimately, keep iterating!

Next up The AI Daily Brief explores more on how the USA is uniting around Open Source, the big news this week, including Jensen first ever tweet. More on this below, but a good explanation here too (as always), thanks Nathaniel Whittemore. Love the show!

📺 Watch 📺

Here Joe Schmidt IV, partner at Andreessen Horowitz sits down with Aman Gour, CEO and Founder of FurtherAI for a discussion on AI Psychosis, Insurance, and the Most Dangerous Thing in Enterprise AI! If that’s not enough to grab you, they discuss a whole load of topics

The single most dangerous path in enterprise AI is building a thin wrapper—taking a general model, plugging in standard APIs, and adding basic agent orchestration on top

For me, at the speed at which the frontier models are operating, this is eminently dangerous, or borderline delusional. it may give you a short-term bump, but Will see your business taken away in a heartbeat overnight. I’ve often talked about the combination of LLMs and SLMs, which many folks are now doing, using SLMs to include their data that is unique to them and gives them the edge, and what makes them unique and different.

Deep domain experience wins every time. Just look at how this is playing out in legal with Harvey and Legora (plus MikeOSS looking to disrupt them with Open Source) and healthcare. insurance has been a little bit different for a while but it is starting to make me wonder!

Next up, Harry Stebbings sits down with Curative CEO and Co-Founder, Fred Turner on the wild pivot from an agricultural DNA startup into a $5B COVID testing operation. Funny, I was watching a Dan Martell video on a boat, where he talked about how Intercom/Fins Pivoted and he had a chance to invest early on, 15+ years back. Often its the right people, wrong thing, or wrong time. But it’s always the right people.

One of the first things that stood out to me, of course, was the vibe coding of a CRM app that replaced the SaaS licence. I go back to my previous comment: This is great if you’re an engineering-first or technology-first organisation. Never mind if you’re AI-native or not. I’d argue many folks don’t know how to ask the right questions to go create this.

Curative canceled a $600,000/year Salesforce contract after “vibe-coding” an internal CRM in just two months that integrates directly with their internal AI agents

There are other great examples in here, including agents that do multi-step tasks that previously took months

Curative replaced a manual 5-person physician credentialing team with an autonomous AI agent. What previously took 2–3 months and cost ~$50 per doctor now takes 12 hours and costs ~$0.20

He also discusses orthogonal supply chains and going beyond the normal paths that we’re buckling under pressure to find new ways to do this, this meant - zero to 206,000 daily tests and 7,000 employees in 9 months by sourcing alternative materials outside the traditional medical supply chain

Well it was kind of in stealth, for the last 8 years, Travis Kalanick (former Uber founder) has emerged with $1.7 billion in funding led by Andreessen Horowitz and having Ben Horowitz join their board. What started (and shared previously on The All In Podcast) as CloudKitchens, is now way more - Atoms

Rather than focusing purely on digital software, ATOMS leverages AI and specialized robotics to modernize physical industries, starting with verticals like Atoms Food, Atoms Transport, and Atoms Mining

And in a final, full circle moment - Uber also joined the investment - to close out Unfinished Business. Excited to see where this goes from one of the defining founders of our generation.

This is so true. And you might be thinking, “What does fitness have to do with innovation in insurance and technology?” Well, this video got me thinking about the depths you have to go to in terms of the art of the possible that stretch your thinking and then bring you back to making a change or a habit inside your own organisation.

If like me, you’re tracking your steps, trying to eat enough protein, getting your workouts in, trying to get good sleep, this is an interesting watch - with, as I say, some intriguing parallels to innovation in organisations. What do you think Brian Solis Brett King John Licata Dave Wright

A final fun one, OK I’m a Jon Gray fan - not because of Blackstone, but because of his running across the world and his movement (anything in common here Greg Reed) and there’s a new level of personality that comes out here, beyond the behemoth that is $1.3trillion Blackstone.

Top 3 takeaways, the New York Knicks, Inc. won, Spain won and Taylor Swifts Wedding! what were you expecting. Life is way more fun when we make it relatable. All that aside, their AI focus is paying off! Specifically, their strategic decision to lean into AI.

Large-scale investments in artificial intelligence infrastructure serve as one of Blackstone’s strongest growth drivers and they continues to scale platforms like the BXDC data center initiative and power generation assets to feed AI’s compute and electricity demand.

📚 Read 📚

McKinsey & Company are back with their latest survey of 1,000 Global Executives, titled - Operational Excellence is the “Hidden Accelerator” of AI Scale. This quote really jumped out to me .

AI is powerful, but it doesn’t replace strong operations, it amplifies them. The greatest impact comes when AI is paired with clear processes, scalable capabilities, and disciplined execution.

To the previous comments, digitizing a broken process or working with crappy data just makes things worse overall, adding AI gets you there faster. Possibly a good thing.

Rather than relying on periodic continuous improvement programs, leading companies treat operational excellence as a dynamic system combining execution discipline, real-time data, and continuous learning.

McKinsey outline the 6 key things to do to avoid these in the first place. Well worth the read!

1. Start with a small number of operational priorities.

2. Design for scale from the first use cases

3. Build the operating backbone alongside the technology.

4. Embed capability where decisions are made

5. Sequence ambition with absorption capacity.

6. Use operational excellence as the stabilizer—and technology as the accelerant.

This is equally a good read from McKinsey and my friend Sid Kamath and team, How AI will reshape the economics of insurance: A CEO’s guide to strategy, focussing on Technology, Talent and Operating Model.

Following the recent OpenAI rogue Agent attack on Hugging Face, Congressman Ted Lieu, and Congressman Nathaniel Moran introduced a bill named the “AI Kill Switch Act” on Thursday.

Hoever, the commitment is voluntary and does not mandate a kill switch. Neither would that mechanism stop a product in testing, like OpenAI’s GPT-5.6 Sol, which was involved in this week’s incident.

In addition to the kill switch act, a bipartisan group of six US lawmakers has proposed legislation that would require developers to submit them for independent security audits. The auditors would be overseen by a new role in the US Department of Commerce, which would also accredit the auditors.

On the flip side to this, the US Department of Business, under Marco Rubio tells diplomats to play down the AI Kill Switch, if nothing more drives further push for Open Source and Sovereign AI models.

Which leads to this announcement which drove a lion share of the news this week, was Jensen Huang took to twitter along with leading organizations, and wrote all about the Open Secure AI Alliance, including ServiceNow.

The world needs both closed and open models. For cybersecurity, open models and open harnesses are essential because they democratize defensive capabilities, increase transparency for defenders, enable cyber defense while protecting data, and complement frontier closed models with customizable, localized controls. Open source enables massively distributed community-driven and self-controlled defense – with no single point of failure.

All this partly rallying against some of the closed models and the the scale they have built so quickly. David O. Sacks shares more here. TL:DR, this is complex, but equally we are here thanks in many cases to Open Source technology and contributions over the last few decades.

I know people don’t have a lot of sympathy for Chinese companies, that’s fine. I’m not defending Chinese companies. I’m defending American developers who need to be able to use everything in the public domain.

At some point, all of this CapEx spend on AI was going to impact the market. For example, while Google Cloud soared (congrats) with an 83% growth and enormous backlog, the number that stood out was AI spend and free cash generation, this quarter a negative $5.9b (compared to $10b positive in Q1)

The cost of insuring debt against default has spiked significantly for the major hyperscalers and AI infrastructure leaders—including Oracle, Nvidia, Alphabet, Amazon, Meta, Broadcom, and SpaceX. Bond market investors are increasingly using CDS contracts to hedge against the financial exposure required to build out global AI infrastructure and data centers.

Did you see I’m hiring - TL;DR, Head of Insurance, GTM - Commerical Insurance Clients (we split our business into commercial and enterprise), USA, East Coast Based. Full details below or here, apply via the link please.

👀 Quick bites 👀

1. OpenAI admits an AI ‘agent’ caused a major cyber breach by itself - I wrote about it here

2. What Humanity Needs To Flourish In The Next Decade - Long Read or listen, thanks James Manyika for sharing.

3. The Wide Open Frontier - another great post from Saanya Ojha at Bain Capital Ventures (BCV)

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! From LLMs to Claims, The Customer Interface Is Moving!

Unfiltered with Matt & Nige

Beyond the Pilot

Unfiltered with Matt & Nige | Episode 18 🎙️…

Listen now

a month ago · Matt Connolly and Nigel Walsh

And that’s It! See you next week! 🚀

🎙️Listen 🎙️

I listened to this on one of my daily walks at the weekend, that then drove me to the post too. It’s really got me thinking about the AI native Company vs the Engineering Native company. I mean lets be honest, AI Native to me would mean it was born that way, but lets not worry about the detail.

What this does highlight to me is the difference between end users of the technology and developers of the technology. It’s easier for a software company to be AI first that often a company that is a consumer of the technologies and platforms. Buy vs build, learning loops vs prompts—all of it applies

As a visual learner, I see an egg timer with the benefits and effort inverted here for manufacturer vs consumer of this, especially when it comes to adoption. While you might be able to ship 3x faster, if the adoption is still a country mile behind, then we still have the problem. Everything circles back to the same problem: adoption and change management. Maybe the next shift will take us from pushing features out to removing features. Jobs said it best: adding is easy, simplicity is hard.

The original post from Amjad Masad, founder at Replit can be found here

Following self driving and future proofing, this is a good listen from Mark Geoghegan, The Voice of Insurance and old friend William Ross, Co-Foudner and CEO of Federato

Lots covered from New consumption models versus traditional premium-based models, typical of Insurance Applications or the growth tax as Will calls out to what future proofing really means in todays day and age vs yesterday. I think this is every enterprise challenge and I don’t doubt we were having the same conversations when building on mainframes all those years back (yes, they are still here). So what will our legacy be? self driving software of some sort.. a return to a component business model of sorts (hello IBM Insurance CBM!) or a hybrid of all of these?

As above, My bet: we need to start removing things, not adding them. Then string them all together in a smarter way, and that doesn’t need to be linear given the new agentic capabilities readily available and emerging. In a positive constant, I guess one thing that hasn’t changed or is super slow moving - the business model of insurance hasn’t yet changed...

📺 Watch 📺

Direct from Insurance Business, this interview highlights the The professional risks market is facing a host of pressures - from Silent Cyber to Silent AI. Everything from underwriting caution to rising investigation costs and reputational risk. Given the environment, no surprise here

Even as broader commercial lines fluctuate, insurers are placing stricter scrutiny on risk controls, policy terms, and capacity limits to mitigate exposure.

And all paths lead to Controls and Governance, whether that’s in and down to your organisation and the risks you write up and out into the market

Advancements in AI tools and shifting corporate governance standards are creating new avenues of liability for firms. Beyond standard claims, clients face escalating expenses related to regulatory investigations, data privacy, and reputational risk management, making legal defense and response costs a growing proportion of total claims.

An interesting watch including Bridge Specialty Group, Jensten, Champion Professional Risks, Talbot Jones Ltd and WTW

Two very quick clips from Talanx and AXA - similar message. No holding back here from Wolfgang Jenewein, Professor at University of St.Gallen siting down with Torsten Leue, CEO of Talanx - that’s one hell of an opening question! on the back of so many previous poor communications of how AI will augment and impact the workforce.

AI will replace people who create no impact

Torsten shares how AI is a game changer for our industry - while highlight 70% of implementation is about culture and people. Short and sweet.

AXA‘s Emmanuel Touzeau, Group Deputy HR Director Agrees. Optimism across the company has increased - having trained 70,000 people - ensuring learning is at the core of the organization.

His point was simple, and I share it: technology alone doesn’t drive transformation. People do.

📚 Read 📚

This week, Satya Nadella, Microsoft CEO’s turn for a Twitter Long form post. Is X making a comeback or did I just leave too early and it never went away? That aside, there seems to be a full on onslaught from many tech providers about giving your Data and IP away.

Lots of insights here from the double standard on licensing costs to paying for AI twice, with the ‘reverse information paradox’ (first time with money, then after that with your know-how and feedback loop). He also highlights the 5C Framework for AI Sovereignty, a topic that keeps coming up time and time again. Control, Capability, Choice, Cost and Compound

This message is not slowing anytime soon, with Bret Taylor, co-founder of Sierra joining the debate on CNBC, as did Marc Benioff with their version of it. Brett highlights

All of it should be yours. And what’s exciting about agents built on platforms like Horizon. Every time they make a sale or accomplish a task, they learn how to do it better the next time. And so this durable asset - the relationships businesses have with their customer and the memories of previous interactions - becomes the moat, and it compounds. It doesn’t go to the model makers, it doesn’t go to competitors.

Speaking at Vibe Coding, Starbucks fired shots at Microsoft and IBM, aiming to cut it’s $400m annual software spend. This does tie back to the very first article about buy vs build, which is coming up all the time. There’s loads of questions in this, from is this where we have skills and experience to where we should be focused. Every enterprise is asking the same question, just ask Keith Pearson or Simon Cox

What is clear is that Business Process Optimization must precede AI

For me, this goes back to a comment Amit Zavery shared a while back, where AI needs to be in the flow of work not just bolted on. Sitting in insurance, we see this all the time - easy to do things in isolation, it gets harder as you need to move the work across the organization.

It also reminds me of Linas Beliūnas‘s post on Vibe Coding vs Software Engineering post here. Worth a look, just for the graphic!

The counter to all this of course is out there - Mismanaged AI makes software more expensive than humans!.

You Just Hired a Million Bad Employees. For the first time in history, humans are cheaper than software. Managing AI is harder than managing people, because AI scales dysfunction instantly.

I like George Sivulka‘s 7 Parallels of Agent Human Workforces on the Andreessen Horowitz blog. He wraps up with a great quote, Humans just became cheaper than software. Someone still has to tell them both what to do.

Another great guest blog from Colin Matthews, now head of education with Lenny Rachitsky. I love the simplicity of this three-ladders framework

Think of this framework as three ladders, each for a different type of leverage that AI can give you.

and importantly, it’s what AI can give you, not what it replaces. you go from being a doer to a director. The examples in here across each of the ladders are just great

The most effective PMs treat AI as a team of semi-autonomous delegates rather than a single chatbot search engine. The highest level of leverage comes from identifying processes that are partly deterministic and partly non-deterministic, building custom harnesses around them, and acting as the final editor or “reviewer-in-the-loop” rather than the worker.

Finally, Colin also points out (which is why I picked this up) that these skills are no longer limited to PMs, they apply across a whole host of skills, including go-to-market, ops, and more!

Speaking of processes, I love what Greg Ingino does here, CTO at Litera in his post - How the Agentic PDLC Accelerated Delivery and Drove AI Adoption. Here he covers a few of the things I’ve discussed already, from becoming an AI-native to engineering-native organisation and ultimately increasing the velocity of things that you can push through the system. But as always, we know that every action has a reaction. In this instance;

The speed of release has become its own challenge. We can now ship faster than the organization can absorb, which has pushed the constraint downstream to go-to-market. Every release carries change that has to be learned and enabled across the company, and customers have to be prepared for the volume that is coming.

Elsewhere, the Sovereign AI or Open Models debate rages on as the long tail of enterprises across the world picks up on the cost of tokens accelerating and looking for alternatives, often after they have tried the primary frontier models. Aaron Levie sums it up well here

Every time we lower the cost of frontier intelligence, the use-cases that enterprises can take on just go up. There’s a tremendous amount of workflows that enterprises would love to deploy that are only gated by the cost of tokens.

Marty Swant and Alex Kantrowitz of Big Technology have a good breakdown of it here too.

Moonshot’s newly announced Kimi K3 — a 2.8-trillion-parameter multimodal model that’s performing on par with Anthropic’s Opus 4.8 and OpenAI’s GPT 5.5 across many benchmarks — is already so popular that Moonshot temporarily paused new subscriptions Sunday.

Finally, Written by me for humans! That’s how I often sum up my newsletter. This study from Pangram Labs which analysed over a million posts of 50+ words or more on multiple platforms suggests that over 40% of long-form content on LinkedIn and 30% of short-form content is AI-generated ... ugh.. Thanks Jeremy Connell-Waite for highlighting.

I can confirm this is all me here, poor grammar, strong opinions loosely held, spelling mistakes and all. Some interesting stats all the same.

👀 Quick bites 👀

1. Mastercard picks UK as test environment for agentic AI sandbox

2. Klaimee raises $5.5m in seed funding - Y Combinator, congrats Ines Boutemadja Julien Catonnet

3. ‘It’s not going away’: The Stanford economist who called the AI entry-level jobs crisis early has the receipt

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I!

Unfiltered with Matt & Nige

From LLMs to Claims, The Customer Interface Is Moving

Another good episode this week…

Listen now

a month ago · Matt Connolly

And that’s It! See you next week! 🚀

🎙️Listen 🎙️

The sovereign debate charges on, as does AI ROI - I think the discussion this week is bang on and builds on last week’s discussion with the All-In Podcast crew. Not only that, the AI token reckoning is coming, specifically when it comes to ROI on AI tokens alone. Chamath Palihapitiya Has some really interesting insights with regards to token increased spend versus benefit attained. The debate then shifts to harnesses and token routers, which can dramatically change the impact, and I suspect this is a massive area to focus on going forward. Oh, and of course, two further trillion-dollar IPO’s, will they or better yet, when will they? This episode actually sums up nicely some of the full circle confusion I’d been noodling on last week’s post.

Finally Brad Gerstner Success with the Trump accounts that went live on the 25th birthday is genuinely a masterclass in my mind. Years in the making. I think we had something similar in the UK a while back with child trust funds, but they were closed to new applicants in 2011.

I read Lenny Rachitsky post on this a while back and listened to the podcast while driving yesterday. It’s super interesting from Noam Segal .

The #1 fear in tech right now is not losing your job to AI—it’s being squeezed to do more work for the same pay.

I particularly like the four archetypes that they described here - and while it may relate to tech people specifically, I think this is true for many people across multiple industries - or could be easily applied. Well worth a listen and read!

• 41% of folks are energizer’s (thats me!).

• 35% are conflicted,

• 12% are Disorientated and

• 12% are Resentful.

This episode of the The AI Daily Brief is great, theres two things that stand out for me. First, Building on the Tech Survey above from Lenny, is the people issue with AI and specifically AI Brain Fry is a real thing! and secondly, the Uber Case Study on AI PODS and deploying engineers to accelerate deployment.

📺 Watch 📺

I missed this a few weeks back, but I saw on Linas Beliūnas feed earlier. It’s Tom Blomfield build on Jack Dorsey’s “From Hierarchy to Intelligence” article I shared a few weeks back, and then his insights and ideas on how to build an AI-native company while at Y Combinator. A true master class in 13 minutes. Really enjoyed this and some of his commentary and ideas on this. Tom is another proponent of recording everything so AI can use it, analyse it, and help you. His example of making introductions, etc., it’s why I use Plaud everywhere I go. Listening to this, I think I can do more with it too

Of course huge congrats too on the news this week in joining Anthropic

A quick take from Scott Galloway on how organisations are staying private longer. The question is whether that’s good news for you. I don’t want to spoil the surprise. but... Our previous era of when Amazon when public at 400m, are long gone.

Private markets are so large, so liquid and so well funded.

In 2000, companies went public in just 6 years, now thats closer to 14 and all the returns are happening behind closed doors.

📚 Read 📚

The MIll’s Review (AI and the future of retail financial services) was released this week and Sheldon Mills, Executive Director at the Financial Conduct Authority says it best himself. (short video below)

The Mills Review marks the first work of its kind initiated by a regulator globally, exploring how AI could reshape the industry by 2030 and beyond.

TL;DR - firm-by-firm regulation is not enough to assess AI risks, especially at scale. I already worry that many people go to LLMs for their advice and insights not just financial but relationship and so much more. It can often be seen as the wild, wild west. Matt Connolly spoken about influences before and how they’ve been charged for misleading consumers, so why should this be treated any differently? the advice versus guidance debate has raged on for years. He also proposes a trusted framework for AI agent participation, specifically in retail financial services.

The big question, of course, the FCA need to examine and potentially adapt its regulatory approach to how LLMs are used within the financial services sector. Is this the beginning of regulation of LLM's?

The full report can be viewed here. There’s also a good one-pager here from Suchitra Nair at Deloitte and a further write up here. Nicely put!

Firms will need to prepare for not only an AI-enabled customer, but also an AI-enabled regulator.

I’ve often read there are two business models - bundling and unbundling. We’ve seen it in every industry, from Telco (mobile, landline, broadband, tv) to Insurance - Bancasurance.

I’ve not seen too many do this though, although they are not the first. Here first direct instead of just switching underwriters, is exiting the market completely. This represents a decisive shift away from the traditional “packaged current account” model - ultimately bringing more friction to consumers. I’m not a fan, maybe ChatGPT (chose your AI model of choice) will do it all for you going forward - which leads me to the age old question of ‘Who owns the customer?’ and my article on Cake! ‘We have all the ingredients, but we just want cake!’.

This caught my eye from Jaya Gupta, Partner at Foundation Capital on value capture problem, not necessarily value creation. Specifically as Jaya starts off with an insurance example citing multiple carriers as an example and how they use models. This builds on the whole debate (shared last week) from Alex Karp at @palantir on giving. your IP to a frontier model company. The flywheel and speed right now that this is evolving is in overdrive.

If the same model learns from every carrier, is your claims judgment still your advantage? The underwriting exception that protected your loss ratio becomes a benchmark. The fraud pattern your team caught early becomes a feature sold back to the market.

But this is more than about training data and signals - its about IP and what you retain and what you share. Having the same core OS or Core Claims platform is one thing, but sharing the same intent, appetite and risk profile is what makes you unique in the first place.

Notice what you keep and what you lose. The insurer still owns the risk, the customer relationship, the regulator, and the loss ratio. The shared model increasingly owns the learning curve. Your mistakes, overrides, and hard-won claims intuition become training signal.

Without including all of it, the last piece Ill share is this.. Have a read!

The deeper asset is institutional context and know-how: the judgment in people’s heads about hard professional work.

Before putting any high-value workflow into Anthropic, OpenAI, or another shared model, ask one question: if every competitor learned how we handle this decision, would we still be better than them?

Bill Devine and our friends over at our new ServiceNow partner Naitiv Partners have mapped out the 3 year journey for Insurance CIO’s. It’s straight to the point and highlights where we are within the industry in terms of exploration and maturity. Of course, not all folks are at the same place. t

1. Escape “Pilot Purgatory” by Targeting Core Workflows, Not Fragmented Copilots

2. Move Governance from an IT Compliance Burden to a Competitive Advantage

3. Anchor Success to Hard Operational KPIs, Not “Vibe” Metrics

Bill breaks down that the gap between an LLM license and an organization genuinely known for AI is not primarily a technology gap. It’s a sequencing gap 📆, a culture gap 🤝, and a partnership gap 🔗.

Speaking of playbooks and sequencing, Manulife aren’t here talking about new POC’s etc, they are talking about an AI operating model shift - from projects to products, common capabilities customized locally, finance-governed value, and workforce adoption baked in from day one—not bolted on.

AI Fluency as Structural Requirement, Not Training Afterthought & Common Platform, Not Scattered Projects

Bolting stuff on in the early days, as you optimise for experience and learning, is great, but after that, if you keep doing this, you’re going to end up in a bigger mess down the line . As Anna Havens shared

“This common foundation allows us to build common capabilities with a common set of standards, including embedded AI governance and guardrails,” Havens says. “This is really going to be the foundation for acceleration and reuse across the organization.”

👀 Quick bites 👀

1. World Cup referee also works as an insurance broker - theres always a side hustle

2. Apple sues OpenAI alleging theft of top-secret information - this looks to be as well integrated as ChatGPT is to the iPhone

3. Allianz confirms hundreds of job cuts as AI reshapes insurance

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! Corgi seems to be getting a lot of press coverage, or Matt and I like little legged pooches!

Unfiltered with Matt & Nige

From Corgi to Alan, The Difference Between Hype and Build

Another good episode this week…

Listen now

2 months ago · Matt Connolly

And that’s It! See you next week! 🚀

🎙️Listen 🎙️

It’s all going on this week - from Anthropic declaring war to Alex Karp Palantir Technologies not throwing shade, Token Maxing is already over, at least for the early adopters. Those companies in early like Uber are peaking in terms of AI budgets, and now Coinbase looking to cut 50% of their AI spend. Harry Stebbings and crew go on to debate the frontier models desire for regulatory capture, specifically after Anthropic’s complaints to the US Government re Chinese Open Source Models. You could argue, the US has heavily taxed Chinese EV imports, why not the same for AI?

While Rory O’Driscoll and Jason M. Lemkin give Harry a hard time re his tweet on what good enough growth looks like in todays money. $1.5 to $5m ARR is not enough!

Elsewhere, Ericson Chan from Zurich Insurance is taking tasks, not jobs. I wholeheartedly agree with this narrative and i feel more broadly the debate has changed. See last weeks data from Ramp and The Financial Times that shared the same, as well as that companies that adopt AI early are growing faster and hiring more people. Whats changed is the hype cycle and pace that we go through it right now, whats not is many regulated industries are still grappling with technical deb through capability debt, often plugged into thousands of applications and green screens still.

True transformation requires a structured portfolio approach (eg Zurich’s enterprise-wide AI360 framework) driven by leaders who are willing to “get their hands dirty.”

He also goes on to talk about future roles (similar to Jack Dorsey from Hierarchy to Intelligence and including coordinators and architects.

📺 Watch 📺

Its not shade, its reporting! A fun 20 mins, and less Alex Karp than Alex Karp (in his own words), but here he calls out partly the points people aren’t saying out loud. Everything how Token-maxing benefits the frontier model companies more than anyone else (yes) - and as a result enterprises are losing trust as a result of massive spending, but little value back.

The answer, NVIDIA partnership with Palantir Technologies to give companies control of their data back. David O. Sacks has a good view on this too on X here

Karp argues that raw models alone aren’t the answer; true enterprise value requires three pillars: an open or agnostic model, a heavy application layer (like Palantir’s “Ontology” to keep data securely ring-fenced), and compute

I think this is one of the bigger points though - if you keep giving the frontend models your data and insights, aren’t you in essence giving away the IP to your business, which is what Alex Karp is playing on. From Claude Code to Claude Legal, what’s next? Claude Insurance?

As Karp exposes, true enterprise safety isn’t trusting that a lab’s future roadmap won’t include your business. It’s retaining the ability to choose — at the model layer — who gets to see and use your alpha.

I love this from Alan, raising €480m as they continue to drive growth for their integrated prevention, now at a $5.5b valuation with 53% YoY growth. Jean-Charles Samuelian-Werve CEO and Cofounder has some really good points on CNBC - specifically

• AI is not making healthcare impersonal, quite the opposite. The existing process is making it so

• A “doctor-in-the-loop” model drives unprecedented engagement (eg, 1 in 3 members opens the Alan app every week, 1 in 10 opens it every single day — to act on their health)

A nice link here to the NHS outlining their plans to reward people for waling a marathon a month. More here. Outside of Alan, he is also a later co-founder of Mistral and noted that Europe shouldn’t just try to keep up with the US and China, but aggressively invest in its own sovereign compute infrastructure and open-source models to win specific enterprise sub-segments. See below for more on this!

One final area I’m passionate about is technology and children. Casey Newton has a great article here, that highlights Candice Odgers TED Talk on why Social Media bans are gaining steam. I’ve not watched a ted talk for what feels like eternity, but this is great and well thought through as is Casey’s article in which he highlights Jonathan Haidt work and the differences to Candice’s, namely -

• The SmartPhone panic doesn’t match the data - it doesn’t stack up

• Smartphone bans do more harm than good

• Pouches don’t address the real underlying issue.

As a parent of teens and technology optimist. My summary is: it’s very confusing, in all honesty, if I had one thing I would undo, it would be giving the kids a smartphone when they went to senior school. Whether it was a thing to do, peer pressure, or just time, I don’t know. At least the requests for TikTok, Snapchat, and Instagram from my daughter are now starting to dissipate, knowing that they’re going to be banned very shortly regardless - so its a win for us.

📚 Read 📚

I’ve been saying this for a while now. Many others are too. The narrative is wrong. Now the data backs it up. A study of 21,559 firms in the U.S. finds that

Companies that adopt AI tend to grow faster following adoption. Firms making the largest AI investments grow employment by roughly 10% following adoption, while low-intensity adopters see no statistically significant change. Entry-level headcount rises 12% for high-intensity adopters.” “Gains emerge gradually and are broad across roles, including engineering, sales, administration, and customer service. The results counter predictions that AI adoption will lead to broad job loss.

I’m sure I can go back to a Capgemini and I study from 15 plus years ago said the same thing about digital transformation - Alan Walker Lee Brooke-Pearce. The study is based on observed AI spending from Ramp card and bill pay data linked to Revelio Labs workforce records. See here for the Ramp study. TLDR, we are scaling up, not shrinking here and doing more with it!

Whilst I couldn’t be there in person for the Financial Times Insurance event last week in London, plenty of friends were, including Simon Hayes who has a good write up here with the key takeaways. Of his list, these three stood out to me.

• Insurers remain focused on assets with stable cash flows and predictable risk characteristics rather than speculative technology bets.

• Digital infrastructure is becoming a major investment opportunity as AI drives unprecedented spending on data centres, connectivity and supporting infrastructure.

• Governance is becoming the critical differentiator for insurers.

See the official key takeaway’s here too.

A full circle moment, perhaps why I think we see the narrative changing from the frontier model companies more broadly. A good read from The Economist

The Not in my backyard (NIMBY) happens all over, folks are all for cleaner energy or new capability, but as soon as it’s in your local village or town, then it’s not here, please. I experienced the same in my village, where we’re looking to build a solar farm which was met with great resistance and ultimately killed this particular project.

Ultimately, we are usually OK with all of this until it actually (may) impact us - be it economic uncertainty (think fuel prices given the war in the middle east) to chip shortages and costs given the rush to build data centers, my Apple Mac and iPhone replacements just went up in cost.

The circle keeps going. That new AI tool that I objected too, that may have taken my job is now being used to help sue their bosses in the first place. In the las few years, there has been a 39% increase in workplace claims. While good for individuals, the sheer volume and complexity of these AI-assisted submissions are pushing an already overloaded tribunal system to its breaking point. .

Alex Mizzi , legal director at the law firm Howard Kennedy LLP, says AI-assisted claims over the past year had gone “from being a novelty . . . to an endemic feature”

The same can be said to be true for many industries, including insurance. As I’ve quoted before, when the cost of arguing gets to zero, then lots of other stuff needs to change to support it

Tech is not losing its shine, but it is revealing boring is beautiful! Forbes shared their recent summary from the latest Y Combinator cohorts.

Compliance and Regulation, which years ago we may have considered a burden are now the accelerator to all of this, as Monica Caldas shared a while back from Liberty Mutual. We see exactly the same with our AI Control Tower and broader capability. CNA Insurance shared their recent success here giving them more control, capacity and ultimately building trust.

And finally, yet related - synthetic testing is now as important. Perhaps boring was always beautiful!

an autonomous AI agent can accidentally trigger live APIs, send incorrect real-world emails, or mess up financial transactions during development, the way developers test software has to completely change.

And finally, a reality check from Lenny Rachitsky and Elena Verna. A simple pause to realise that often we live in a bubble. Whilst exciting living at the leading edge, not everyone is at the same place. nor is what they say or do online equals to what they are actually doing or implementing at scale in their enterprise. I have no doubt it will come

Meanwhile, I’m over here asking ChatGPT to rewrite the same paragraph for the third time because it keeps defaulting me into ‘LinkedIn wisdom post’ mode. GARH.

👀 Quick bites 👀

1. Google smashed with $2bn bill in comparison site case - in Sweeden

2. Weak hands and blurry vision: Are you getting a ‘phone body’?

3. UBS Finds Enterprises Throttling AI Spending - Simon Taylor has the lowdown here

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I!

Unfiltered with Matt & Nige

Unfiltered with Matt & Nige is a straight-talking podcast about insurance, innovation, and everything in between. From tech trends to family life, training, and life on the road.

By Matt Connolly

And thats It! See you next week! 🚀

🎙️Listen 🎙️

Some great insights again from Phil Fersht, sitting down with Balkrishan “BK” Kalra, CEO of Genpact on the $18 trillion reason your AI isn’t working! Some great takeaways in here, including this.

No Artificial Intelligence Without Process Intelligence: Real returns on AI investments cannot be achieved without structural, foundational work on underlying business data and operational processes.

or as our Amit Zavery, AI in the Flow of Work! and that AI Needs a Holistic Business Leadership Agenda: Indeed, more and more of my conversations are about business change adoption and operating model as opposed to just the technology. I also really like his idea of AI builders and AI practitioners, goes back to the Jack Dorsey article of “The Journey to Intelligence” shared recently too.

📺 Watch 📺

Scott Galloway I love you, but I don’t agree with this one bit! insurance is not a scam! The old saying is true, the many pay for the few.

Insurance is a scam, a transfer of wealth from poor to rich…who can absorb risk the poor and middle class can’t. The industry should be regulated—EBITDA margins >20% triggers customer rebate(s).

Not everyone is in a position to self insure nor wants to, and this differs country to country and line of business to business. Our industry has a bad enough reputation as it is. This does not help.

I wouldn’t want to be without healthcare in the USA which reminds me how thankful we are to have the NHS in the UK. it’s not a fair comparison to take Scott Galloway versus the average Joe in the US or UK and say they could do the same thing. It’s also why we have folks uninsurable in the USA given wildfire, flood or hurricane in certain states and state backed Insurer of Last Resort. The comments section here on the LinkedIn post are an interesting read.

A huge thanks to Mark Thomas and Beyond the Desk Podcast where we got to sit down and talk about my own career journey, not something I often or like to talk about, or at least in public.

It was great to do and I love the people stories he has allowed guests to share over the previous episodes so was a fun discussion! Really makes you look back and be grateful for the great companies and more importantly people I’ve had to work with over the last 20 plus years in insurance.

Finally, Google just dropped a free 8-minute lesson on building your first AI agent. This is a super clear explanation from Smitha Kolan.

At the simplest level, an AI agent doesn’t just answer—it decides and takes action. In this video, Smitha goes beyond basic chatbots and demonstrates how to build a fully autonomous, self-correcting multi-agent system from scratch using Google’s Google Agent Development Kit (ADK).

📚 Read 📚

This is genius, many folks have pointed it out this week including Owen Williams over at Ascot Group. Here Bain & Company are vibe coding AI replicas to test a takeover target’s edge, as the rapid advances in AI leave buyers struggling to judge which software groups will endure, especially given the news recently from Thoma Bravo and Medallia

The mock-ups let potential buyers test how difficult the technology would be to reproduce as the cost of building software is rapidly falling, and also understand how the product could evolve.

“It’s kind of the difference between seeing something in 2D versus 3D,” said Rebecca Burack, head of Bain & Company’s global private equity practice. For me this is a rapid digital twin that tests the future

Prediction Markets have had a load of press over the last few years from Kalshi to Polymarket and many more. I’ve shared before how some countries are calling them as they are - gambling! and as a result, have the rules and regulations of gambling. The Wall Street Journal dug in to find out more - do all the social videos really make it as good as it looks? as the old saying goes, “if it looks too good to be true, it usually is.”

The WSJ’s analysis of over 1,100 social media videos revealed that Polymarket paid online content creators to post videos claiming massive financial wins. Creators falsely claimed they collectively won $1.9 million; in reality, many of these bets never existed or would have resulted in heavy losses. In about 10% of the videos analyzed, creators claimed ~$900,000 in winnings when identical parallel bets on the actual site would have resulted in a loss of over $166,000.

In some of these instances, creators use dummy sites to record the videos and then post them online. When some of the creators were contacted, their videos posted were scrubbed and they declined to comment!

From token maxing to token cost. To be fair, the former didn’t last long! The best models used to be American and closed, but that’s been massively challenged by open-source models now. I love this from Saanya Ojha at Bain Capital Ventures (BCV).

But perfection is not the threshold for disruption. GLM 5. 2 only needs to be good enough to make the premium models defend their pricing. Z.ai’s founder is already betting they can reach Fable-class open-source models by the end of the year.

I suspect we will see many firms move to ‘good enough’ as token costs and and adoption rapidly rise.

First we had silent Cyber, now Silent AI. Insurance giants including Berkshire Hathaway , Chubb Travelers AIG and more recently won approval to largely drop AI liability protections from standard commercial policies. We’ve seen this for a while, and I firmly believe this will evolve into a huge new market. Where there’s risk, there’s opportunity. but where there’s a lack of understanding or uncertainty, there are exclusions!

I’m also back on my jobs bandwagon, firmly of the belief it’s creating new and more, not just taking jobs away. Here PwC share analyzed more than a billion job ads and found that AI is reshaping junior roles rather than erasing them. Some fascinating stats too

Junior roles are now seven times more likely to demand skills that traditionally took a decade to develop: high-level judgment, leadership, stakeholder management, and strategic decision-making.

Whilst I love that we’re raising the floor, sometimes we have to enjoy the journey, and experience only comes from... experience. We need the failures along the way to teach us the path. Jumping straight to the answer isn’t always the answer. This is why I believe the adoption question is really hard and a challenge to most large organisations as we scale this.

Of course, the other issue that they talk about here is that whilst the roles have increased in responsibility or ‘seniorization’, the pay hasn’t!

If you’ve not got the impression from my newsletter over the last year or so, it feels like Silicon Valley is firmly back!. My old friend Daniel Garsin over at Elixirr shared his views on why you need to experience this firsthand. It was great to join him on this trip too along with one of our customers and share the experience first hand. Apparently their was Karaoke too!

Finally - too much change? Not quite sure how this is the Ferrari CMO’s challenge - Slightly similar to what happened with JLR, Don’t tell anyone, but the Luce (I cant afford one anyway) is growing on me, after the initial - WTH! moment and I still stand by my view the that new Jaguar will also be a resounding success!

👀 Quick bites 👀

1. Four in five under-16s in Australia using social media despite ban, study shows

2. NatWest launches bank-wide AI ethics accreditation

3. How DXC Is Using Agentic AI to Modernize Insurance Operations with Assure Smart Apps - DXC Technology‘s Ray August talked to Dan Reynolds on their layered, non disruptive architecture! (I’m biased, they are a huge partner of ServiceNow in bringing this to the Global Insurance community)

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! Corgi, Trillionaires and the AI Reality Gap! plus loads more as usual!

Unfiltered with Matt & Nige

Corgi, Trillionaires, and the AI Reality Gap

Unfiltered with Matt & Nige | Episode 14 🎙️…

Listen now

2 months ago · Matt Connolly and Nigel Walsh

And thats It! See you next week! 🚀

🎙️Listen 🎙️

Logistics is dirty work. Not just that, it’s not cheap either.

a medium-to-large physical goods company spends roughly 5% of its total revenue on logistics. An industry average would be 8-12% of total revenue.

Nice to hear a non AI and tech story for a change (don’t worry, plenty more of them below) - although he does talk to how they are leveraging AI and building agents to support 100 core workflows. No industry is safe., and if you go to their www - it opens with AI powered Logistics for the real world. Let’s be honest, this is impacting every industry and Heavy Industries are more prone than any other.

Here Flexport CEO, Ryan Petersen joins Harry Stebbings on a pretty broad ranging conversation - ok they do mention AI and tech briefly, but more around remote work (not shy on his view) including their 5 day office policy, parenting, his meeting with Masa, SoftBank Investment Advisers, the benefit of being a Founders Fund company, while the broader VC industry appears to be a circular rumor mill.

📺 Watch 📺

Starting off with Perplexity this week, Aravind Srinivas argues that the industry is too focussed on racing to build the best model, however these will be commoditized. The next bit is what I’ve been saying for a while and what excites me most. No point having the best model if it cant doing anything. why I often describe ServiceNow as the practical application of AI.

The true value lies in the orchestration layer—the unified system or “harness” that can route tasks efficiently across different tools, files, local compute, and server-side models from multiple competing labs

Lots more in here too with Harry Stebbings including China’s acceleration as a result of the Export Controls (see last week), as well as dismissing an AI bubble. The real challenge goes back to Infrastcture.

Infrastructure is the physical bottleneck restricting token production, hardware and memory supply giants—like Micron Technology —will capture enormous pricing power, potentially making them more valuable than Meta

Not for the first time this week, but Jonathan Haidt Professor at NYU Stern School of Business and author of The Anxious Generation sits down with Anderson Cooper - and talks to how early tech exposure is not an advantage. Not all friction is bad. In fact, its what we need to learn.

While kids in the 1990s learned valuable skills like hardware mechanics and basic programming by tinkering with computers, today’s “attention economy” software is intentionally engineered to be effortlessly intuitive

This ease of use is equally creating a dopamine hit each time. It may feel like magic at the time, but its actually doing more harm than good. Theres also a great framework and audit Jonathan highlights from Kevin Stinehart here. I love his framing and the audit itself.

Before we ask, “What is wrong with kids?” maybe we should ask, “What happened to the ecosystem around them?”

Finally, a great example of how AI is being used to help improve safety - especially when 2 million people converge on one city. A person can only monitor a finite number of screens before fatigue sets in. - the same way that Air Traffic controllers only work for short periods of time.

Instead of just recording events or notifying officials after an incident occurs, this AI system analyzes crowd movement in real time to spot density build-ups before they turn dangerous.

A clear example of how AI is being put to good use and saving lives in simple practical terms.

📚 Read 📚

Statefarm Insurance made headlines this last week post their employee kick off in Las Vegas, what started with a Pink concert, Jimmy Fallon-led sing-along and ended with a bombshell

Jon Farney, a 33-year State Farm veteran, told his sales force he was ripping up their existing contracts. Any agent who wants to stay past 2027 will have to sign up to a new compensation deal and sales targets. “State Farm needs to change,” he said, according to a video of the event reviewed by The Wall Street Journal.

A big driver of the change, AI, which can’t be a surprise to anyone. Every industry across the globe is rapidly evolving and using AI from heavy goods and distribution to of course, financial service’s. In in fact calling it a bombshell is quite something given that most employees are asking now how work out how they leverage it to do their jobs faster and more efficiently. The writing’s been on the wall for a while, and like most situations, doing nothing is never really an option.

• Lost #1 spot to Progressive: Facing competitive pressure, State Farm lost its decades-long spot as the nation’s top auto insurer to Progressive

• The AI Mandate: State Farm is forcing its 19,000 independent agents into a new “Human + Digital” requiring them to use AI tools for instant quotes, policy summaries, and automated claims processing.

“State Farm is under pressure to meet customers’ needs that we cannot and should not ignore,” “We have a finite window to change.”

and whilst hard to read in some instances, it’s no surprise at all that we all have to change how we work as customers evolve how they wish to buy, engage and their expectations. I suspect many others will follow.

Mistral is back in the spotlight, not for what they have done, more for what others have - namely Anthropic and their challenges around Fable 5, Mythos and more. Aaron Levie talked about this too - with the big winners, being open weights models - here.

That aside, with the conversations shifting back to Sovereign AI and the role of governments in regulating AI. Arthur Mensch‘s opening line here says it all

We exist to make sure that everyone gets access to the best AI systems, outside of centralized control exercised by states or corporations that feel the need to control in-fine deployment of AI

Over on X, Satya Nadella shares a short but important blog on ‘A frontier without an ecosystem is not stable’. Here he highlights how over-reliance on the frontier model itself. Importantly he talks to the difference between Human Capital and Token Capital. Not a incidence on his words when he talks about high value Human Capital.

Every company is going to have to build what I think of as human capital and token capital. Human capital comprises the knowledge, judgment, relationships, ingenuity, and pattern recognition of its people, while token capital is the firm’s AI capability it builds and owns.

Beyond this he talks about not worrying about picking the best model - thats the wrong question. Instead think about building a learning loop on top of models where human capital and token capital compound. Focus less on frontier model’s and more on frontier ecosystems. or to my Peloton friends - #togetherwegofar!

This is the ethos I’ve grown up with where platforms enable more value on top than is captured inside, and where every company can continuously innovate and build value of its own.

While we are thinking about all these, and can build lots out - Robert Pick warns of getting too far ahead without the correct governance, specifically the 1 year lag.

AI’s technical capabilities (especially agentic, autonomous tools) are running roughly a year ahead of a company’s actual ability to govern and monitor them safely.

Article below and key points here from Bob while at Insurtech Insights in NY recently. here.

Case in point perhaps, a Canadian BMW Group Dealer had to eat a $7,000 error from Quinn, their AI chatbot. What actually stood out to me was not the error, but how the customer felt when they released (after) they were negotiating with a chatbot. All that said, the expectation is the same. Honor what your tell me. (remember Air Canada a 18 months ago here)

“I feel embarrassed, and I feel angry that I’ve been negotiating with this bot,” Giacomelli told CBC News. “If they’re going to be replacing their employees’ jobs with AI, then they need to be honoring what that AI says.”

Great to see this for Allianz and all our friends across insurance with latest Evident AI index (now in it’s second year) for Insurance which maps out the Top 30 global insurance companies for AI across a number of dimensions including Talent, Innovation, Leaderships & Transparency. Take a look for yourself

Allianz ranks #1 among 30 insurers in the Evident AI Index 2026. With a 28% larger AI talent pool, Allianz sets the benchmark for AI in insurance.

Beyond this, there are some great examples and use cases from insurers across the index on how they are attaining demonstrable benefits from deploying AI across their organization.

Speaking of reports, there seems to be a whole host of new AI and Insurance reports hit the market over the last few week’s. Here are just a few.

• Deloitte - State of AI in the Enterprise - 2026 - Deloitte

• Insurance at an AI Inflection Point - EY, Ireland - EY

• Rethinking insurance distribution in personal lines for the AI-empowered consumer - Deloitte

• Global Insurance Outlook 2026 - NTT DATA

• State of the CIO, 2026 - CIO Magazine®

I rarely use AI to write this report (it’s my thinking and learning time), but in this instance I took all the reports and passed them through My favourite French friend which summarized the surprising findings as follows (3 slides)

Sticking with learning, my former colleague 🤖 Lee Provoost - CTO over at Flagstone shared this summary and I agree, a great resource on the top 25 AI concepts from Tokens to RAG to prompt engineering, all easily explained with infographics and quick summaries.

I really liked this for my friends Caroline Bedford and Hélène Stanway who took a trip to a museum of failure in france. Reminds me of some of the work the X, The Moonshot Factory do in celebrating failure. I loved comments like this -

The intuition was completely right. The technology worked. The world simply wasn’t ready for it.

As the old saying goes, we learn more from failure than we do success. If only more organizations were less hung up on getting it right first time, instead of learning through the journey.

And finally, another old saying - when the service is free - you are the product!. Here an AI company is sending a team of camera-clad cleaners and a private chef to boot, all wired with high-tech recording apparatus show up at your home.

No this is not the latest edition of Love Island oh a strange realityTV show - but actually working out what folks do day in day to help train the next generation of humanoid robots.

But, there is a catch: this AI company is gathering data to train the next generation of cooking and cleaning robots, and every inch of your apartment is now being recorded.

👀 Quick bites 👀

1. Billionaire Ambani wants AI in every call, app, and home

2. Privacy has long been dead. Meta accused of more privacy breaches with ai glasses. She Came Out of the Bathroom Naked, Employee Says

3. Santander goes Open Source - which insurers will follow? - thanks Simon Taylor

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! Shipping is easy, adoption is hard - life in the gap! it seems to be the recurring theme of late.

And thats It! See you next week! 🚀

🎙️Listen 🎙️

I’m not very good at reading books as I said before but this one I couldn’t put down. (This is the answer I should have given you, or Outlive from Peter Attia Mark Thomas) It’s one of my favourites over the last few years: Build by Tony Fadell So we’re super excited to see this interview with Lenny Rachitsky. Just a brilliant listen throughout, From their battle with BlackBerry, seeing if the hardware and software can come together and what’s a data versus opinion decision

The unintended consequences of the iPhone, how often they were on the brink of failure . vibe coding version 2 is easier than vibe coding from zero. E.G. Flighty and Steve Jobs reaction to allowing porn on the iPhone to how he describes the iPhone as a refrigerator, in which you can put bad food in or good food. A throughly enjoyable listen.

Speaking of all these maxing, Citadel‘s report got a lot of attention this week, given the cost of a token is dropping. The AI Daily Brief covered it best on on their recent podcast. This directly from their description.

The scary token chart isn’t measuring demand and it’s not measuring total expenditure.

Wall Street has decided the Silicon Data “token expenditure index” proves the agentic boom is over. It doesn’t. The chart measures the weighted-average price of a million tokens, drawn only from third-party token routers — companies whose entire job is to find cheaper tokens. It says nothing about total demand, volume, or expenditure. The real story is the shift from token subsidy to token scarcity, and when the median firm still spends $11.38 per employee per month on AI, the growth in total tokens consumed will dwarf any drift toward cheaper baskets. That’s not a bubble popping — it’s a market rationalizing.

📺 Watch 📺

In the latest Emily Chang Interview with Anthropic founder Dario Amodei - interesting times ahead from the Bet on the Enterprise, which clearly is playing out in the Open AI vs Anthropic IPO Valuations (and OpenAI’s revised shift to the Enterprise) to Hyper Growth thanks to Claude Code.

Through to the real world clash with the White House, despite winning a huge contract - this of course has evolved even further recently with both the launch of Fable 5, and the subsequent white house order to withdraw it. Or from my friends at Armilla AI. Now you no longer need to wonder about Sovereign AI or Sovereign Cloud conversations.

The US government invoked export-control authority to suspend all access to Anthropic’s Fable 5 and Mythos 5, forcing the company to disable two commercial models for every customer.

Equally, the UK Government just announced a £1.1bn investment into Sovereign AI here

Another of my favourite movies, The Big Short investor - Steve Eisman talks to CNBC on the SpaceX IPO and what it means. TLDR, he’s not a fan!

He points out that according to SpaceX’s S1 prospectus, the company defines its Total Addressable Market (TAM) at a staggering $28.5 trillion Crucially, 85% of that projected market is entirely in AI, not space or satellite internet - so xAI or Grok!

He also highlights the Capex changes from pre AI pivot (42%) to post AI (215%) of Revenue, mainly due to AI infrastructure. Despite deals with Google and Anthropic for renting Colossus 1 with some pretty heft monthly returns.

Despite all this, his final two worries focus around Commoditisation of LLM’s (what differentiates all the major providers) - which I don’t quite agree with - I’d have stuck more to the Sovereign Risks, and his second one is the end of the Subsidies Era for AI. More on that below from Citadel.

Speaking of large rounds, Jeff Bezos is back with Prometheus with a whopping $12bn raise, out of stealth. Launched just 6 months ago with $6.2bn. These numbers or astonishing, but that means that the opportunity must equally be astonishing.

Bezos explicitly clarified that Prometheus is not a robotics or factory-floor automation company. Instead, it is building an AI system designed to act as an “artificial general engineer.”

The software aims to compress the complex, grueling lifecycle of heavy industrial engineering—such as designing jet engines, aerospace systems, microchips, and medical devices. Ultimately, given this is his first CEO (co-ceo) role since Amazon, he’s all in!

Final one to close out, is my old friend Adrian Jones at Acrisure interview at ITI recently on how tech capabilities increasingly delineate successful MGA’s to those that don’t. Adrian point’s out the following, my only comment - this is the industry wide, not just MGA’s now.

A successful MGA can no longer rely purely on traditional relationships and legacy systems. Advanced tech stacks and AI capabilities have shifted from being an “added advantage” to an absolute prerequisite

Importantly, while technology will handle high volume, low to mid value property - the real win is the combination of Human + Machine.

while preserving specialized human underwriting strictly for massive, highly customized risks (up to $10 million).

Finally, given the news of the smartphone ban in the UK, this is both a really clever and really hard watch (as a parent) all in one from the Smartphone Free Childhood group. Jonathan Haidt, Professor at New York University sums it up beautifully by saying.

this perfectly illustrates the predicament we put children in, in the name of “connecting the world.

His full write up and associated credits here

📚 Read 📚

The debate on jobs rages on, but with a twist. Thsi time more and more folks talking about how the original message of mass job loss was overblown. I’ve talked about this a lot over the last few weeks with folks like Aaron Levie from Box talking about new roles being create.

Here David Solomon, CEO of Goldman Sachs wrote a guest essay in The New York Times to which the core of thesis is we don’t just automate a task and bank the savings. With this approach, we would never make progress and grow, it would always be squeezing the same lemon. It’s goes back to one of my fav phrases right now, we are not just raising. the ceiling, but the floor too.

companies don’t just keep the same product and make it cheaper; they raise their standards and offer more.

And while he believes as many of us do, disruption is real - it will likely be only 25% of current work hours, and as a result the headlines of an “Apocalypse” is Overblown. Equally, the new moat will be a human one, as roles change and evolve as well as physical roles. Theres plenty of great campaigns right now talking about how physical constructions sites are manned and built by people. Doesn’t mean to say that AI wasn’t leverage along the way. But this is just natural evolution.

Jeff Bezos joins in too - and shares how AI will bring the Multiple Golden Ages, not Mass Job Losses - here.

My friend Simon Taylor, author of the brilliant weekly Fintech Brainfood shares his view on the AI Operating Model and highlights 4 companies that have moved from chatbot (everyone) to How to actually use AI in production.

A quick and really good read covering Deep dives from Anthropic, Ramp, Allica Bank and BBVA on how they have deployed AI touching on:

• The 5 Levels of AI Operating Model

• What makes a company move up the ladder

• Custom Harnesses, Skill Sharing and Employee upskilling covered

Speaking about deploying AI, this from Bryan Davis, ChFC, CPCU over at VIU by HUB, where he highlights the latest Boston Consulting Group (BCG) on the New Rules of Customer Experience in which bCG call out their 5 new rules.

Consumers now expect every interaction, human or AI, to feel personal and effortless, and that raises the bar for brands across every industry.

Having started out many years ago in CX, i feel the expectations have just been racing up and the ability to delight and wow has got harder and harder, with any missteps creating exponential frustration. Bryan talks about meeting customers in the journeys they are already in - reminds me of the way Amit Zavery talks about ServiceNow AI in the flow of work. It’s more comfortable, natural, relatable. Full report here

The nuance is that “seamless” means very different things to different people, especially for a decision as personal as insurance. Some consumers want a fully digital experience. Others want a live agent. Many want both, on their own timeline.

it must be report season as there are new reports out on AI from EY, Deloitte, McKinsey & Company and others as well this month. More next week

In a world of continuous evolution, we went from GenAI to Agentic over a rapid time which also meant we moved from Search to Prompt Engineering to Context engineering.

Now Boris Cherny, creator of Claude Code says he no longer writes prompts, he writes Loops. Here Dharmesh Shah, Founder and CTO of HubSpot shares how to write your first AI loop

From Tokenmaxing to Outcomemaxing. This from Sierra highlighting interesting lessons from their first year and how Outcomemaxing is fundamentally rewiring their business.

When you charge on outcomes, your incentive is to get the job done: complete the mortgage application, retain the subscriber, process the insurance claim, increase the basket size.

The cost of raw intelligence is going to fall (see last weeks Block post). That’s the clearest trend in technology right now. Software priced on top of that compute will compress with it. What doesn’t compress is outcomes: a customer who did not churn, a home sold, an insurance claim processed has the same value, irrespective of what the model costs.

Finally, our job in insurance is to remove risk. Given the ongoing and hopefully nearly ending conflict in the middle east, this is good to see from Emirates Now offering insurance to tempt passengers back to Dubai. To put this into context;

Transfer traffic at Dubai International Airport (DXB) has fallen to roughly 40,000 passengers daily, down from 100,000 before the conflict, as insurance exclusions deter bookings.

Tim Clark, the airline’s president, said they will fly people home if they are stranded, “on other carriers if necessary” sets a higher bar than most airline disruption policies. No formal launch date or policy terms have been published yet, but the intent and objective is clear.

👀 Quick bites 👀

1. Morgan Stanley will soon open its trillion-dollar wealth management funnel to AI agents

2. Opendoor’s India exit is fueling a bigger conversation about AI and outsourcing - thanks Phil Fersht

3. AI will drive efficiencies, won’t replace brokers: CEOs -

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I!

And thats It! See you next week! 🚀

🎙️Listen 🎙️

The Voice of Insurance himself is back - this week Mark Geoghegan sits down with my old friend Andrew Johnston from AJ Gallagher. I’m not sure if this is an own goal or the beginning of AGI, or quite simply why we don’t let people mark their own homework. The importance of multiple different models, checks and balance and the right governance and guardrails.

AI Physician, Heal ThySelf”—is that while AI is actively creating a brand new class of modern liabilities and risks for businesses, AI is ironically the only technology powerful enough to help insurance underwriters accurately map and solve those same risks

This and no surprise that AI is dominating all investments into InsurTech with 95%+ all going to AI. Honestly, I suspect its very hard to get funding for anything not AI enabled or powered right now, but this wont always be the case.

Finally, when it comes to Risk of Fragility vs Resilience. Ths sums our current state quite nicely:

The pushback emphasizes that simply stacking more software on top of an already unquantified, emerging peril doesn’t actually build resilience for insurance companies—it potentially just creates a fragile, self-referential loop

Sticking with AI, a great take from Harry Stebbings and the CTO of Legora, Jacob Lauritzen, with everything from vibe coding through to token maxing (and why it’s bonkers) and some of the things he wishes he had done differently, e.g. hiring faster. Now using these examples to go to 100x (not 10x) and how they can scale quicker from a brilliant base. Really good pragmatic and honest discussion all around.

Even a mention for Matt Connolly and our new favourites - Wispr Flow.

📺 Watch 📺

I met up with Erik Roen, SVP and CIO Claims at Travelers last week at the Stone Point Capital Symposium in NYC. Eric had recently sat down with Denise Holland Dresser, CRO at OpenAI to share their work together on Transforming FNOL and launched an AI-powered claims assistant nationwide.

To scale from an 8-state pilot to a countrywide rollout in just two months, Travelers built a near-real-time data streaming platform called Mission Control

What’s super interesting to me here is the shift in how software was developed, from a traditional 80:20 - Tech:Business to a fully collaborative, 50:50. The results speak for themselves.

80% to 90% of opting-in customers successfully completing their claims entirely through the AI assistant, with about 35% of folks bypassing AI entirely and opting to speak to a person still.

It’s only just getting started, given the success here, almost 20 other areas are lining up behind Auto Physical Damage example and across the claims lifecycle. Really great to see this example live!

The Unicorn Economy is back. This from the All-In Podcast Equities event, a great summary and insight into the return of the Unicorn economy with Thomas Laffont, cofounder at Coatue Management.

The venture capital and unicorn (startups valued at over $1 billion) ecosystems have been heavily out of balance, consuming far more cash than they return to investors. However, a massive liquidity event is underway.

With 3 imminent IPO’s at the very peak of technology (SpaceX , Anthropic and OpenAI) returning trillions to investors. The speed at which these companies is scaling has never been seen before, but that comes at a cost, more specifically, a smaller number of highly concentrated organizations. This time around, some are even seeing profit on the journey to public markets.

📚 Read 📚

I was in NYC, this last week for a number of events including Stone Point Capital AI Symposium and Insurtech Insights - thanks to Spencer Holtzman Nick Zerbib and Rory Pyke respectively. A few updates from friends around the industry. This from Rory Yates at Synechron as well as updates from Roger Ferrandis at Sixfold here. A great week all round, both at the events and in and around them catching up with old friend’s across the industry.

It feels like the general consensus and original fears and hype of a job-apocalypse were much overhyped and the attention now turning to the roles that are being created as a result. The New York Times just did a piece on some of the 13 new roles that Box have been hiring for, driven by the growth in AI. From Aaron Levie directly here

Internally, we’re bringing on roles that can help us accelerate how we work and serve our customers due to the power of AI agents. We have roles that help us eval new AI models, internal automation engineers that help us leverage AI for our workflows, and professional services roles supporting a forward deployed engineering motion to help our customers adopt and optimize AI.

Speaking of which, this is a great weekend read from Sarah O’Connor at the Financial Times on Should AI steal your job? Three things stood out to me:

1. Technological Change is Not a “Natural Disaster”

2. The Shift From “What AI Can Do” to “What It Ought to Do”

3. The “Future of Work” is Filled by Human Friction

and to me, not all friction is bad! we need it in many places to inject empathy and much more - not everything should be Straight Through Processing.

As organization’s evolve and adapt AI, a new org structure and key roles are evolving. Jack Dorsey (yes, the Twitter one) and now CEO of Block along with Roelof Botha, partner at Sequoia Capital penned this great blog ‘From Hierarchy to Intelligence’. In summary, they point out that The Org Chart Was an Information Routing Protocol, Not a Human Design and AI will Absorb Coordination, Replacing the Need for Traditional Hierarchy. Finally, a shift to ‘The Edge’ will leave three key surviving corporate roles

• Individual Contributors: Deep, specialized practitioners who build and operate the business. They take context directly from the AI system rather than a manager, allowing them to execute instantly without waiting for permission.

• Directly Responsible Individuals (DRIs): Outcome owners who have full authority to pull resources from the system for a defined sprint, solve a problem, and then move on.

• Player-Coaches: Experienced practitioners who continue to do the core work but also invest directly in developing, mentoring, and upskilling the people around them—functioning as true mentors rather than administrators.

We’ve seen some of this already with folks leaving large corporate roles to be Individual contributors at Anthropic.Thanks to Ruediger Adolf from E3 Group (also on stage at Stone Point Capital this week) for highlighting.

Not all AI is at it seems, and this was to be expected I’m afraid - Harriet Scott at the Insurance Times shares more. To be fair, we have seen this before with Greenwashing, the difference then - that came with obligations and fines!

While 99% of insurers have generative AI initiatives in motion and 83% are spending over $5 million annually, only 42% have actually deployed AI into active business functions.

As expected at this stage, there is a massive gap between the hype surrounding AI and its actual operational value. Never mind the latent liability or the AI Gloss, the bigger issues will be downstream for Brokers and Clients and could spill over into any of these areas:

• Professional Indemnity (PI): If the AI gave bad advice to a client.

• Directors and Officers (D&O): For leadership’s failure to govern operational risk.

• Cyber & Regulatory Fines: For misrepresentation or data privacy breaches.

Those not AI Washing are setting their organizations up for success and sharing examples (like Travelers above). Here Steve Rullo, Chief Digital & Technology Officer and Guardian Life shares how they are building AI-Ready Talent for this next wave of transformation across the enterprise.

Scaling AI isn’t just a technology challenge; it’s an organizational one. It requires new skills, new ways of working, and a shift in mindset across the business.

Steve also shares insights on placing scalable bets. We cant simply chase everything (especially at the current pace of change), They fall trypically into 3 areas. Broad Productivity Tools, Targeted Business Transformations with a Golden Rule that true value is only realized when

experimentation is tied to strict data governance, cyber controls, and clear business ownership of the final outcome.

For all parents out there, Allie K. Miller is helping with the never ending questions of what and how much do we let our kids play or engage with AI. There’s loads of great stats here, but these are the ones that scared me. Parents are in the dark about their kids’ AI usage:

• 49% of parents have NEVER spoken to their child about generative AI

• 96% of families don’t know their elementary school’s AI policy

• 44% of parents feel they lack the knowledge of AI to guide their children

I wont list out the 9 things here, but 100% worth a read for us all - its sound advice, including - #4 : You have to be able to function without AI!

👀 Quick bites 👀

1. After recent comments from the CEO of Standard Chartered, the FT asks, Am I lower Value Human Capital? - I look at this differently, what will your role be in 12, 24, 36 months out. enabled by AI and how should you stay ahead.

2. Billionaire tech exile pledges almost £2bn for Britain’s AI data centre blitz

3. In a sea of identical smartphones, Nothing stands out - an interesting take for Nothing phones and Carl Pei

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I!

And thats It! See you next week! 🚀

🎙️Listen 🎙️

Enjoyed this from Robin Merttens InsTech and Mark Twigg FRSA and Robert Agnew from Agentiv-x - Focusing on Neurosymbolic AI. It’s not the first time I’ve come across this, I first met David Ferrucci who was building a company called Elemental Cognition a few years back while at Google. Love the idea there. I can see why this is important specifically in insurance. Two key things for me here are:

1. It’s move away from a black box AI to actually explainable AI

2. The move from underwriters to brokers as the primary focus

It will be interesting to see how this develops, evolves, and is adopted by the market.

I was super excited to see this from Corgi, who’ve just raised $106m at a $2.6b valuation. Here the CEO Nico Laqua sits down with Harry Stebbings and talk through not the journey or building for insurance, but actually more about the intensive workplace culture.

Move over 9-9-6. This is 24/7 where Nico actually lives and works in the office - yes he’s got a mattress there too. The first 30 people that work there all have a Corgi tattoo (let’s face it - It’s a pretty cool logo), and they opened a cafe in the place of an old barber shop in the building, the Corgi Cafe, that’s open 24/7 - were drinks are sponsored by partners. a go-to destination but doesn’t make any money.

My other takeaway is that they are spending $400K a month on Anthropic tokens and zero on OpenAI. I listened to the All-In Podcast too this week where Marc Benioff was talking about salesforce spending $300 million a year on Anthropic. All that said, an interesting interview - if you are interested in the culture. Love to hear part two Harry Stebbings, where we talk about insurance and the challenges in building in a regulated space that’s not Revolut or another fintech. Insurance is quite unique, and not transactional like banking.

On the flip side this interview just oozes experience, hard yards and someone who’s been there to the deepest darkest hard times and come out the other end (burning $8 million a month). Net net, as Andrew Feldman, founder and CEO of Cerebras , “hardware is hard” - sitting down with Harry Stebbings, 20VC. This posts the largest ever semiconductor IPO. Interesting to hear his view on this not being a bubble.

The current AI buildout is the exact opposite—infrastructure is severely behind demand. Cerebras has a $25 billion backlog, and NVIDIA and AMD are facing similar constraints

Like many others, they are not stuck behind the high bandwidth memory constraints, where it’s only made by three people in the entire world (Cerebras use SRAM). As with many Of these technology episodes recently, lots of talk about China and what we should and shouldn’t provide to them to maintain an advantage. And whilst many of us have heard the quote ‘We’re in the business of turning Electricity into intelligence’, Andrew beleives

this philosophy relies on a massive, unproven assumption: that models will simply continue to get infinitely smarter the more energy you feed them. If the capability of these models eventually plateaus, the economic incentive to keep dumping multi-gigawatt amounts of electricity into them will completely collapse.

📺 Watch 📺

Okay, we interrupt this usual coverage to talk about Luce. Ferrari‘s first ever fully electric vehicle that has made headlines world over for so many reasons. Mostly how it looks. to be fair the internet seems to be united, for once. As both a car and technology fan the thing that intrigued me most was that many of the reviews came from technology vendors not just typical car journalists. Heres Marques Brownlee on the Luce. But it doesn’t stop there. Even the Pope got involved.

All that aside dear Internet, thank you for your memes. They’ve been brilliant!

and hear from Sir Jonny Ive directly (yes the apple one) here with Cleo Abram ( not a car journalist) and Ferrari Chief Designer, Flavio Manzoni. Interestingly, Jony is known as a huge car guy too, so this is a good watch (in case you miss his voice from those iPhone ads for years and his rationale for breaking all the norms of a Ferrari).

I shared part one last week. Here’s part two from Rory Sutherland and Tom Goodwin and by pure chance there happens to be an insurance story too! as well there’s lots more about cars, from Tesla to self-driving. Rory is an avid Jaguar fan and also recently test drove the new Jaguar that also caused huge controversy - see his review here.

Back to Insurance, here’s a great story about how an insurance company was dealing with complaints re the length of time claims take to process. The company was about to spend a large amount of money speeding up the claims process (sound familiar) until someone asked a question

What was the actual length of time an average claim takes? Answer: 2 weeks. So they had an idea. Write and tell everyone that claims take 4 weeks. Suddenly all the complaints stopped except for a small number of messages thanking them for their speedy response

Turns out they didn’t have a speed problem they had an uncertainty problem!

📚 Read 📚

Away from cars, Pope Leo is also getting involved in AI with a controversial 42,000-plus-word encyclical (I had to look at what this was!). I’ll be honest, I’ve not read it - but lots of articles about it. The core challenge seems to be

human dignity versus an algorithmic optimisation - specifically we can’t treat society as an efficiency problem, and that human vulnerabilities are not software bugs to be erased

Equally his implied views of the leaders of some of these tech companies Building AI capabilities it’s something to take note of. When the head of the Catholic Church with over 1.4 billion people around the world puts out such a strong message. It can only drive further backlash and uncertainty against AI, not that it didn’t already have a poor reputation.

Interestingly, Christopher Olah, the atheist co-founder of Anthropic delivered a candid critique of the very industry he’s been serving for years and

explicitly called for external, unbending moral voices to regulate the industry, proving that tech insiders are actively looking outside Silicon Valley for ethical guardrails.

All that said, Softbank just pledged €75bn to Build Europe’s biggest AI facility in France, their largest investment outside of the USA. Perhaps in part with regards to some of the narrative around sovereign AI right now and a heavy reliance on the US, This shows a diversification in strategy and mitigation of risk.

“Softbank is proud to make this major commitment to France,” said Son, SoftBank’s founder and chief executive. “With its industrial capabilities, talent base and national ambition, France is uniquely positioned to become a leading AI infrastructure hub in Europe.”

I have covered this a few times already but now it is starting to show up in the news more broadly too. This from Bradley Olson at The Wall Street Journal. When you put a measure in place, people will try and maximise it, in this instance tokenmaxing. Many organizations have now said their AI Blank Cheque (check for my American friends) is definitely over. No surprise given the numbers I shared above from Corgi and Salesforce. Net net, companies are falling into three broad categories:

1. Tiered Access - specifically for non-technical employees

2. Homegrown solutions - to use cheaper internally built open way models

3. Real-time dashboards of usage - putting responsibility back on the user

Microsoft had already announced it had cancelled claude licences and shifting developers back to copilot, many highlighted for similar reasons.

Phil Fersht from HFS Research shares his recent keynote from his New York Summit, “How to AI” which has 6 big takeaways. His presentation linked below. Two of these that stood out to me are:

1. Stop funding AI by cutting the people who make you human, and start leading it as a strategy. Firms that rip out junior headcount to fund AI pilots are not being bold, they are being shortsighted. Learnability is the new critical capability, and the firms that invest in young talent alongside AI will build the culture and identity that the cost-cutters will spend years trying to recover.

2. You cannot deploy AI on top of your enterprise debts. Process debt, data debt, people debt, and tech debt are the four real blockers. 72% of your people are hiding their AI use because they fear being judged. That is a leadership and a talent failure, not a technology problem.

The presentation is well worth a walk through as well as his other four big takeaways. Thanks Phil!

Back to Aaron Levie, CEO Box - he continues (I agree) with the premise that AI will create more jobs than it will remove, as we look to deploy this at scale across enterprises. And not by a small number. Take whatever number of people you thought might be in jobs related to AI deployment in the enterprise and multiply it by 10. Then probably 10 again.

Either way, all told, we’re in for years of AI diffusion, and along with it tons of new roles and areas of work to be done to deploy AI at scale.

For years I’ve talked about insurance being available in Minecraft so my kids get educated about what it is and how they can leverage insurance to rebuild the worlds they’ve created when one of them comes along and destroys up the others, just for fun. What better way to teach kids about insurance?

Here ERGO NEXT Insurance are partnering with TikTok in a bet that shows embedded insurance will shift from retail to commercial. Not a bad bet given that the TikTok shop last year generated over $60 billion in gross merchandise value.

Following the partnership, vendors will now be able to purchase commercial insurance policies directly through the app’s Seller Centre interface.,

TLDR; Its early days. However we choose to partner, we must do so responsibly and with the right controls and guardrails. On one hand, we have Allianz Commercial‘s Shanil Williams push back on predictions that artificial intelligence liability will follow cyber in becoming a standalone product (sitting down with James Thaler from The Insurer here) - my 2c is people will look to exclude this quite quickly, in the same way they have done for hyperscalers. We also have seen a number of new startups in this space like Armilla AI start get traction.

All the way through to a New German decision regarding liability for AI chatbots as highlighted by Jakob Plesner Mathiasen (thanks!). Many folks will remember the Air Canada chatbot from last year, this time following suit and holding a cosmetic treatments company directly responsible for full statements made by its chatbot.

👀 Quick bites 👀

1. ‘I forgot what it feels like to be outside’ - First rooftop intensive care ward opens - Sadly, I’ve spent more time in hospitals that I care to talk about over the last 18 months, but love this initiative.

2. BNP Paribas steps up Mistral partnership to bolster rapid AI defences - Europeans sticking together here

3. Spain blocks prediction markets Polymarket, Kalshi over lack of gambling licences - yes prediction markers are just casinos with different names.

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I!

And thats It! See you next week! 🚀

🎙️Listen 🎙️

Mythos caused a storm when it was first released and since then has found more bugs and issues than ever previously discovered by all of the cyber software ever created. On one hand it’s super impressive and on the other you’ve got to ask why these tools that were created and find cyber incidents haven’t been successful. As the optimist in the room, this for me just means we now have the ability to close these gaps really quickly and move into making better software that’s safer for all.

Casey Newton and Kevin Roose are joined by Nikesh Arora, Chairman and CEO of Palo Alto Networks to discuss how AI Safety is So Back! and specifically the tension between AI safety advocates (who want mandatory government pre-clearance) and tech billionaires, who argue that any regulation will stifle American innovation and allow China to take the lead.

As with everything AI, the velocity of everything is just running multiple times faster!

Nikesh emphasizes that the timeline of cyber threats has shrunk drastically. In the past, hackers might take days or weeks to exploit a known system vulnerability. Today, automated and AI-driven tools mean vulnerabilities are scanned and attacked within minutes or hours of being discovered

An AI backlash with more and more graduation speeches being booed. The latest this week, Eric Schmidt, former CEO of Google . What’s actually going on. Here the All-In Podcast share their view.

I have to agree here. I do think we have a PR problem. If all the headlines just talk about the ‘jobpocalypse’, then why will we expect our next generation to be excited by this. Too many examples this week, from Standard Chartered CEO calling people ‘lower-value human capital’ to Cloudflare CEO, Matthew Prince calling folks ‘measurers’ - Fortune , and then their is Mark Zuckerberg at Meta and how folks are feeling about training models to replace themselves.

“The vast majority of those we laid off last week were measurers,” he wrote. He defined “measurers” as those in middle management, finance, legal, internal auditing, and revenue recognition.

It’s a PR and storytelling problem. it’s why Dario at Anthropic shifted from 50% of all white-collar jobs going to a new narrative recently.

📺 Watch 📺

For a change starting with a watch! In fact the hype around the Swatch and Audemars Piguet collaboration has gone off the scale to riots in some places, leaving many shops closed. I love what they did with Swatch and Omega (to be fair, owned by the same company). Here Rory Sutherland, Vice Chair at Ogilvy shares his view about collaboration and brand partnerships. See below too for an article from The Wall Street Journal or whether this is a smart move or not.

Sticking with Rory Sutherland, this time he sits down with Tom Goodwin for the first of a two-part series with Jon Evans on Uncensored CMO. There’s a lot in here that I found really useful, everything from digital overload ( you’ve heard me talking about being overwhelmed with what’s going on with AI right now), over complication of products and options (I see this all the time) - and why Tom prefers shopping where there is limited choice versus an abundance of choice (e.g. Amazon or eBay). I also loved Rory’s example of his Sage Appliances Toaster and the button that simply said “just a little bit more.”

And why companies need a ‘Director of Trivia’ to sort out all the things that fall between the cracks, or specifically job titles

Even one for Christopher Sandilands on why people love Ryanair - Europe’s Favourite Airline but hate, British Airways.

Away from all this, Ashwinn Krishnaswamy has a short, simple and to the point post on how we may have gone too far with optimisation culture. We measure everything and forget to enjoy it.

📚 Read 📚

Sticking with the jobs backlash and AI booing, this is a great wrap up from The Independent. No need for a summary, you can imagine!

“We want to see new technology create jobs, not destroy jobs.”

On the flip side, Benedict Evans has some good and practical insights here about how we can actually predict AI job exposure, mainly backtest against other big technology shifts in the past. This example from accounting.

We spent a century automating accounting: we built calculating machines, punch cards, mainframes, data processing, databases, PCs, spreadsheets, ERPs, cloud… in fact, we built half of the tech industry around automating this. Yet the number of accountants kept going up.

The bigger test is not about a specific task being automated, however, it is more about a business model change, e.g. the newspaper test or an Uber test. Very much well worth a read.

The age old - rubbish in, rubbish out. But some folks are feeding it rubbish on purpose. Google and others are now fighting mis-information in AI models and specifically search manipulation. In just 20 mins Thomas Germain at the BBC managed this (with just a single article on his website).

I tricked ChatGPT and Google into telling the public that I am a world-champion competitive hot-dog eater. The joke was dumb. The problem is serious.

As Google rolls out countermeasures to stop this, manipulators quickly adapt and find new loopholes to stay ahead of the algorithms (same as the security episode from above). Because these defences are never 100% foolproof, the ultimate takeaway is:

All of us must maintain a healthy skepticism and independently verify critical information rather than blindly trusting AI-generated summaries

Home with ServiceNow for a min, I really liked Chris Bedi‘s wrap up from Knowledge26 - Our Chief Customer Offier. The three things that stood out to me were (and linked to some of this week’s themes). Check the article for all five in more detail.

1. AI value realisation is a J-curve.

2. Nobody believed in a jobless future.

3. Nobody wanted to vibe-code an enterprise platform.

At Google‘s last I/O, Demis Hassabis. CEO Google DeepMind closed his keynote by suggesting:

humanity may be standing at the “foothills of the singularity.”

A bold and exciting claim all at once, especially given some of the AI rhetoric right now. I actually loved Saanya Ojha introduction - Google I/O Day 1 was yesterday, and Google did what Google does best: brought the fireworks, then set them all off at once.

Google meanwhile takes twenty announcements that each deserve oxygen, throws them into one keynote-shaped blender, and wonders why the market only remembers three. The company needs to learn how to tell one story at a time.

That said Saanya does a great job of highlighting the key announcements from the day. You can also watch the full keynote here

I wanted to share this in the back of Rory’s interview above. Ultimately as he said - we play with an Aston Martin Lagonda Ltd or a Ferrari as a child long before we (ok a small number of the population) actually buy the real thing. This is no different, Audemars Piguet are playing a long game with this collaboration, while at the same time building cultural relevance and awareness to a new generation.

It’s already a higher price than most watches and higher again than the Speedmaster collaboration they did last year with Omega. What I actually love is all of the companies that are taking this Royal pop and trying to make it into actually a wearable with straps and so much more. It’s been fantastic to see the speed at which they’ve been coming out and popping up on Instagram. Of course you know actually, wearing an AP, you’re wearing a $400 watch that you can’t even service or even get to the movement. Would I buy one if I could? Absolutely.

Where and how we spend our time is really important. In fact many of us work really hard to protect our time (I’m always striving to get better). Whilst I agree with Eduardo Ordaxs sentiment here in the desire to be continuously learning,

As I said in my comments - we’ve also got to take time out to reflect, be inspired, see and experience new things. I get a huge amount of my inspiration from movies when travelling or TV shows or nights out with friends, cycling, running, fitness - in fact the only common theme here is where I spend time off screen. From Minority Report to The Matrix, making us consider or believe, or Knight Rider in the 80s that had a guy who talked to his watch and who’s car drove itself. Hello KITT!

There’s no point in creating a perfect future if you don’t get live it - including experiencing the ups and the downs along the way.

A good example recently was, I was with a few friends in the legal world, one that had never used ChatGPT and another that had just got his first access to Harvey - both in awe of its respective capabilities. What I highlighted to me though was we often live in a bubble and it’s important to break out of that. Or as I share with folks who I’m mentoring

it’s important that we’re both interested and interesting.

I’ve always been fascinated by this debate. Seema Amble, Partner at Andreessen Horowitz does a great job in breaking down whether or not software is losing its head. I think about this a lot as we see more and more insurance companies and other institutions launch capability on ChatGPT. This piece tackles some really great questions.

• When software goes headless, where does defensibility move?

• The UI was the product. ( I think Amazon all the time as I’ve believed their UI to be terrible but it was a means to get anything delivered within hours, or the call centre agent that had spent 40 years memorising every keystroke, which meant any change was going to be a nightmare)

• What makes a system of record durable? ( a very big debate in insurance given the investments)

As incumbents go headless, they are making an implicit bet that the data layer will remain the source of value. In some categories, especially those that are deeply compliance-bound like financial services, that bet may hold for some time and going headless may be farther away.

👀 Quick bites 👀

1. Trillions of miles of data: Your car is spying on you, and it’s only just the beginning - Thomas Germain BBC

2. Britain’s EV boom hits second-hand bump in the road - Financial Times Kana Inagaki & Laith Al-Khalaf

3. Record 50m freestyle time at controversial Enhanced Games - after all the hype, only one record was broken!

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! clients are ahead. Insurers are behind. Tune in to find out who actually said this it’s our latest catch up

Unfiltered with Matt & Nige

Clients Are Ahead, Insurers Are Behind, and Someone Finally Said It

On Unfiltered with Matt & Nige, Episode 11, we talk conference season and why the best conversations rarely happen on stage, Nigel’s new love affair with Whispr Flow and all things amphitheatres and ancient drainage tunnels…

Listen now

3 months ago · Matt Connolly and Nigel Walsh

And thats It! See you next week! 🚀

🎙️Listen 🎙️

My old friend Robert Pick sits down with Tammy Soares from NTT DATA on all things AI and adoption. Some great insights here, specifically as Bob calls out in his own words.

1. Being deliberate about AI, versus just being fast.

2. Recent downplay of AI for productivity purpose is off-base.

3. The need to more deeply consider the effect of AI on the humans who supervise it. (This has come up a lot in the overwhelming nature of managing agentic teams).

Slowing Down is a Competitive Advantage (Not being a Luddite) - Deliberate Intentional Outcomes, not AI for AI Sake.

Amrit Santhirasenan hyperexponential hosts Scott Stephenson, former Chairman, President and CEO of Verisk for a fascinating discussion on the Underwriting Intelligence Podcast. A practical look at Data Moats - created in decades, not overnight! to a fundamental shift needed in Underwriting operational speed. AI Compresses everything.

Software and point solutions need to be spun up at a fraction of past timeline

importantly, they touch on the shift from buying products to buying outcomes.

Because of the sheer volume of data and the complexity of modern workflows, insurers are increasingly looking to purchase comprehensive, vertical software solutions that deliver tangible business outcomes rather than just analytical outputs

📺 Watch 📺

Creativity, Curiosity and Storytelling. These have been my 3 for a very long time. Here Scott Galloway on why Storytelling is what we need to teach our kids. Jeremy Connell-Waite you will be pleased to see this!

Corporate Executives mentioned Storytelling 469 times on earnings calls in 2025 (up from 354 the year before)

Our very own Amit Zavery sits down with Alex Kantrowitz from Big Technology to get to the crunchy question on will AI replace SaaS. Amit gets straight to it - and importantly, folks are missing the nuance.

AI by itself doesn’t guarantee outcomes.

What makes AI actually work at the enterprise level, is what you wrap around it: context, guardrails, governance, and the ability to take action safely and predictably.

The companies that will win with AI are the ones treating it as a capability to harness, not a replacement for the systems and processes that are already working

📚 Read 📚

More and more folks are standing up to this AI layoffs. Here Andrew Ng shares that it’s more of a PR and expectation play, than math.

To be clear, I recognize that AI is causing a lot of people’s work to change. This is hard. This is stressful. (And to some, it can be fun.) I empathize with everyone affected. At the same time, this is very different from predicting a collapse of the job market.

Importantly, and not enough folks are talking about the upside here. (Please can we choose a word easier to spell!)

Net Job Creation Will Outpace Destruction. There will be an AI jobapalooza!

He’s not the only one. Scott Galloway writes Apocalypse No, citing folks rolling work forces back to pre pandemic levels. Equally, the challenge is more about the speed of transition to the new world and tools.

The relevant policy question shouldn’t be about whether jobs will permanently vanish, but rather whether society can handle the velocity of the transition period as workers adapt to the new tool.

I might be biased here, Bill Devine is both an old friend and we are actively working together in his new capacity at Naitiv Partners, however Bill has put together a really clear and concise article here - quite simply, there is no clear single bullet or monotholic platform.

Carriers are searching for a single solution in a world that requires many. The architecture is wrong. The workflows are wrong. And the governance model most carriers are building will fail the test that matters most.

Well worth a read - some great insights here on the challenge, reframing the problem (focus on outcomes not technology) and how to get started.

Jeff Wenger, EY shared this recently and it really got my attention - specifically how Physical AI may become he biggest underwriting disruption the insurance industry has faced in years.

In an article from Reuters (Michael Levine Geoffrey Fehling and Alex Pappas), insurers are reassessing exposure around robotics, autonomous systems, and AI-enabled operations. Some carriers are already introducing AI-specific exclusions while others are building entirely new AI-focused coverage models.

That’s a major signal. AI is no longer just changing workflows. It’s starting to change underwriting structure itself.

The conversation quickly becomes operational: • How are AI decisions monitored? • What controls exist? • Can actions be audited? • Who owns escalation?

The question is shifting from: “Are you using AI?” to: “Can you prove you can operate AI safely at scale?”

It was only a matter of time - as Belle L. at The Wall Street Journal recently reported - we have a new problem. I’ve said it for a while, first we had application sprawl - now we have AI sprawl. Companies have a new AI problem. Too Many Agents

Gartner predicts that by 2028, the average global Fortune 500 company will run over 150,000 AI agents (up from fewer than 15 just a few years ago).

Despite this looming tidal wave, corporate guardrails are lagging dangerously behind. According to Gartner, only 13% of organizations believe they currently have adequate AI agent governance in place, leaving the vast majority of enterprises completely unprepared to audit, centralize, or secure their automated workforces.

This was inevitable, given all the news around Anthropic Mythos, this week Google warned of the first time hackers used AI to find and exploit a security flaw.

The Era of AI-Driven Offensive Cyber Operations Has Officially Begun. Cybercriminals have used artificial intelligence to discover and weaponize a previously unknown software vulnerability – the first confirmed case of its kind – Google revealed Monday.

Beyond simply writing code, hackers are scaling their operations by building automated pipelines to cycle through commercial AI provider accounts, using anti-detection tools to bypass safety filters and bans. Attackers are pioneering autonomous attack orchestration; for example.

Even the mighty Apple security has been tough to Crack. Mythos Helped Find a Way In according to The Wall Street Journal here

During tests in April, researchers found software issues in MacOS, one of the world’s toughest targets for hackers

Brokers, mortgage, insurance, finance - they all seem to be in the spotlight right now with the role of AI transforming their respective markets. This time James Pickford Deputy Editor of the Financial Times looks at the role of AI in UK Mortgage Broking.

Rather than just disruption, (8 in 10 mortgages are provided through brokers) and pure efficiency gains for a heavy manual and paper based process, we are already looking beyond just this. Insurance Next?lin

David Morris, head of homes at Santander UK, sketches scenarios in which AI over time takes on more of the advised process. It might start to give prompts to brokers or borrowers to say: “Have you considered X, Y or Z?” helping them understand their options.

Finally, always good to share Benedict Evans updated - AI will eat the world’ presentation for his Spring release 79 slides, well worth a read - as we prepare for radical uncertainty.

👀 Quick bites 👀

1. Love this - The many joys of small talk, Pilita Clark Financial Times

2. The AI trick I use constantly - this from Joanna Stern on how she used AI to help create her book - I am not a Robot

3. The insurance industry’s biggest problem in the Gulf is a misconception

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! This week the cost of AI credits vs salaries, what will win?

Unfiltered with Matt & Nige

AI Credits vs Salaries, Token Maxing Goes Corporate

On Unfiltered with Matt & Nige, Episode 10, the boys start recording whilst still laughing, and eventually get into some good chat about spending the same on AI credits as employee salaries 🎙️…

Listen now

4 months ago · Matt Connolly and Nigel Walsh

🎟️ Join Us 🎟️

See many of you tomorrow with Robin Merttens and the InsTech team for our inaugural InsTech ServiceNow event in NYC. Very much looking forward to this.

📆 Tuesday 19th May from 1300 ET. 🎟️ more info here 👇🏻

And thats It! See you next week! 🚀

🎙️Listen 🎙️

Zoe Kleinman sits down with Otter.ai CEO, Sam Liang on how AI should capture everything and every conversation should be recorded!. The ⭕️ anyone?

Liang observes a generational divide: younger “digital natives” are generally more comfortable being recorded and living their lives online,

From my own experience with my teenage kids, they are often more worried about this than my generation. Mind you every time I leave the house, Mrs W used to plug out all AI assistants, now they just don’t get plugged back in. I have to say, I love my AI assistants, whether old or new and theirs plenty to chose from. I suspect consolidation here at some point where these become better and ultimately native on your phones.

📺 Watch 📺

OK, it was going to be impossible to not share what I was up to all week. The ServiceNow Knowledge Event just gets bigger and bigger. This year with over 25,000 customers and partners over 3 days in Las Vegas. WOW!

Here, Jeff Wenger from EY and I tried to share our Top5 takeaways from the event - and yes, this was day 3, so my voice was 100% raspy.

Don’t worry if you missed it - you can still live stream the keynotes and so much more. Hear the Day 1 keynote with Bill McDermott, joined by special guests Jensen Huang, CEO NVIDIA as well as Nick Tzitzon, Holly Briedis, Amy Lokey, Bhavin Shah, Amit Zavery and special guests Raj Subramaniam CEO, and Vishal Talwar, EVP and CDIO of FedEx. Such a phenomonal start to the week.

Amit Zavery led our Day 2 Keynote here and Lauren McManamon our Developer keynote here

Finally, our FSI spotlight led by Gregory Kanevski along with Guests Kelly Caspers SVP, Digital Solutions & Services and Tripp Rex, SVP, Enterprise Risk Framework, U.S. Bank. I say finally - there’s over 80 sessions for you to tune into in the K26 catalog here.

And, while on Knowledge, one of my fav photos from my colleague - Adolfo Tunon with the crew! Talk about a perfect fit!

📚 Read 📚

Mike OS - have you seen it? How have you not. It’s been all over my feed all week and keeps going. It’s an open source legal AI platform, an alternative to Harvey and Legora. There’s a tonne of posts online about this and how it’s blown up. Equally, given it’s open source, theres add on’s, forks and much more. See here from Mike, and his original post 2 weeks back here. I am intrigued as it’s very similar to insurance when it comes to working in a regulated industry. Not many industry-specific models have taken off but if I was to pick two, it would be legal and medical. Watching carefully and with great interest, as is the press.

The multi billion dollar question’s, everyone is asking. A great read from the team at McKinsey & Company, Arun Gundurao, Vasanth “Krish” Krishnakanthan Rory Walsh, Sanjay Kaniyar and Sid Kamath

No Bob in account’s cant vibe code a new PAS and claims core on the school run. I’m not going to try and summarise this, theres too much goodness here - needless to say, theres not a single facet of the insurers operations where we can’t improve (not so marginal gains). However, this perhaps may be the biggest question.

Lets stop digitising (ok if that what we need to do to keep everyone on board), but lets work backwards from a world where software can (and is allowed to act). First start in a function, then a division or line of business, and then joined up across multi agent systems and self healing, autonomous workflows.

Strategy over Tooling: CEOs must stop asking “How do we add AI?” and instead ask how work should flow in an environment where software can act

Give Claude eyes!, thanks Allie K. Miller. I’m loving all the hacks and ways people are using these. To be fair they are flooding LinkedIn, and as Allie suggests, this should just be a skill. Give it a week or two and who knows!

Agentic commerce conversation seem to come and go. I like this framing from Angela Strange and Stripe framework, from discovery through post purchase. Another mention for the cost of arguing going to zero. Maybe ok if you are a retailer, but what if you are an insurer or law firm. What happens to the cost of legal advice or insurance?

It’s quite hard to convince people to take on new super cool AI capability while at the same time telling them they wont have a job. The narrative continues to evolve, here Dario Amodei switches from the AI blood bath to a term thats been used quite a bit recently too - Jevons Paradox, a term Satya Nadella of Microsoft used in Feb last year. The thinking now;

As AI removes the friction of doing the work, the value shifts entirely to deciding what work is worth doing. Human judgment, taste, and high-level strategy become more valuable because they are the only parts of the process that cannot be scaled infinitely by a GPU

We talked about taste, grit and imagination last week.

With AI, all of our roles and functions changed forever. None more-so than the CIO. Here, A. Ravi M., SVP and CIO at Box shares some of his insights from cost and operations to revenue generation and how the role of Customer 0 (I think we used to call this, eating your own dogfood or drinking your own champagne) - in this case Box on Box (at ServiceNow we have the NOW on NOW programme)

Ravi’s other interesting point for me here is around stewardship. Gone are the early AI days of letting 1000 flowers bloom. Now we need to work out what we focus on and what we don’t, while balancing innovation with the need for compliance and controls. I’d argue we all have multiple hats these days, and that’s here to stay.

With all the noise around Job cuts, this is an interesting read, post Gartners recent survey results of 350 enterprises. Put simply, there is no correlation between AI-driven layoffs and improved ROI. None.

“Workforce reductions may create budget room, but they do not create return.” Helen Poitevin, Gartner.

The real win is in using AI to create and unlock value, not just get more efficient. Don’t be the laggard (only focus in and down). Reminds me of the Google X conversations - start with the hardest thing first. Yes, there will be some replacements, some roles will go - others will emerge, eg what Aaron Levie is advertising for at Box - here - AI Business Automation Engineers.

What will win though is up skilling our folks to work with and alongside AI, and ensuring people are where they need to be (and are expected). As I have said before, the floor is being raised, not just the ceiling.

👀 Quick bites 👀

1. Kalshi valuation soars to $22bn - Prediction markets continue to soar!

2. Sam Bankman-Fried’s Venture Bets Would Have Made Him $100 Billion Richer Had He Stayed Out Of Prison

3. Trucordia launches Trucordia Go - soon to be on ChatGPT too. The list is certainly growing.

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! Finfluencers get their comeuppance and why distribution is the real moat!

Unfiltered with Matt & Nige

Finfluencers, Fake Certainty, and Why Distribution Is the Real Moat

On Unfiltered with Matt & Nige, Episode 09, we talk money, moats, and giving advice with ZERO accountability. 🎙️…

Listen now

4 months ago · Matt Connolly and Nigel Walsh

🎟️ Join Us 🎟️

Final opportunity, join Robin Merttens and the InsTech team in our breathtaking Hudson Yards HQ with a whole host of insurance partners and friends from Naitiv Partners, INSTANDA, Concirrus, hyperexponential and EY and many more! very much looking forward to this.

📆 Tuesday 19th May from 1300 ET. 🎟️ more info and to register, see here 👇🏻

And thats It! See you next week! 🚀

🎙️Listen 🎙️

A great interview with Pravina Ladva, Group Digital and Technology Officer for Swiss Re who sits down with Peter High on the move from Experimentation to Scale. Lots of great insights here, from taking processing from 40 days to 4 or 5. The key for me is enabling people not just showing up with the great new shiny tech. For Pravina, this is 80% people, 20% technology and of course, none of this possible, without having a brilliant data foundation.

Here Harry Stebbings 20VC sits down with AppLovin CEO, Adam Foroughi and this was brilliant. Not from the tech perspective, but the honesty in leading a company, the all in nature the managing a drop of 92% in a single year, the toll on his personal life, health and how he spends time now.

Interestingly, the psychological burden of managing “Agentic Workforces.” Unlike human teams that go home, AI agents operate 24/7.

This has created a new type of founder burnout where the feedback loop never pauses, leading to a state of permanent cognitive oversight that makes it nearly impossible for leaders to mentally log off. Lot’s of folks I have talked to recently talk about similar challenges and the overwhelming nature and pace of change right now.

📺 Watch 📺

Continuing with Harry Stebbings - here he speaks with Anjney Midha, founder of AMP PBC. Again, the thing that intrigued me (beyond being an early investor in Anthropic after they had 21 No’s), is the personal side of things. Now I appreciate you need to be in a position you can do this too, but Human Alignment is harder than AI Alignment.

Anjney expresses deep vulnerability regarding his relationship with his parents in India, admitting that after moving to Stanford for college, he often went weeks without calling or texting them

Hmm, I feel Im about to enter this phase with my kids - who just text for a password! That plus aging.

It also reminds me of the story from Sahil Bloom a while back when he works out how many more times he would get to see his parent if he stayed on the West Coast. Answer, 20. The Tail End concept which he wrote about, as did Tim Urban back in 2016.

I have zero regrets moving back from NY to be with Mum over the last 18 months or so, before we lost her a few weeks back.

In fact, Scott Galloway posted a chart on this yesterday about where folks spend their time with the caption, hits hard. In a world buzzing with Tech, spend time with and hug your loved ones.

📚 Read 📚

This was hitting the news all over the place this week. With great power(AI) comes great responsibility! What’s super interesting is that the AI seemed to ignore all the guardrails and instructions specifically defined here. 9 seconds later, the entire database was gone (since recovered I understand)

PocketOS , which provides software for car rental businesses, suffered a massive outage over the weekend after the autonomous artificial intelligence tool wiped the database and all backups in a matter of seconds.

Now imagine this in your enterprise Global Insurer or Bank. Guardrails and Governance is the number one and number two conversation right now for our clients, specifically with our AI Control Tower. Alexis Bliek Anna Mazzone

Highlighted here in practice, on the back of Pravina’s podcast. Data, Legacy platforms and security concerns are all holding us back - in a report titled - Artificial Intelligence Appears to be Ready, But Most Insurers Are Not. Not sure I fully agree, I doubt any industry is perfect. AM Best Senior Director of Industry Research and Analytics Sridhar Manyem adds:

AI systems can produce unreliable outputs when underlying data is of poor quality, fragmented across legacy systems, insufficiently governed or lacking appropriate context. Insurers that have invested in modernizing their legacy systems and have robust data governance will find it easier to integrate AI into their workflow.

This too from Joseph C. Peiser, Global CEO for Aon Risk Capital. Joe’s got some really great insights here, specifically that brokers that win the AI era will be the ones with the most trading data and market leverage, not the most expertise.

Artificial intelligence will erase knowledge as a premium in commercial insurance broking but will not level the competitive field.

Again, all roads lead back to data - and in this case trading data, coupled with decades of relationships. Whilst I don’t agree about levelling up (I think this new found capability for agile, tech first brokers the opportunity to catch up quickly), more broadly - brokers have one of the biggest opportunities across the market given Joe’s comments around being ‘aggregators of business’ and contract certainty. One for you Alison De Spong , Tim Schreck and Steve Dowling

And it’s not going unnoticed by the worlds regulators. This from the Australian Australian Prudential Regulation Authority

In a letter to industry published today, APRA warned that governance, risk management, assurance and operational resilience practices are not keeping pace with the scale, speed, and complexity of AI adoption.

This goes from a literacy gap (reminds me of Cyber 5-10 years ago) to an over reliance on each vendor, with no joined up view. Ultimately, responsibility sits with the exec and board from an operational and oversight perspective.

Boards have strong interest for AI’s potential benefits but many lack the technical literacy required to provide effective challenge to management on AI related risks and oversight.

Speaking of AI, this intrigued me. Theres a trust and knowledge gap emerging here. Lots of links back to what we teach kids in school, but this latest insights from KPMG showed that:

6% of 18 to 24-year-olds stated they would pay higher premiums to ensure a human is involved in their claim process

The need for empathy in the claims process (or anywhere it’s not familiar or you are in control of it) means people are more skeptical. In this particular case, the worries centre around AI being able to make Fair and Ethical AI decisions.

Only 36% of consumers said they would accept a fully automated claims journey even if it meant their insurance cover would be cheaper.

The tone is changing (but I do think it eb’s and flows), so on one hand - consumers will pay more to avoid AI (above), on the other - the end of ‘Complementarity’. The current earnings season shows this veneer is peeling away. Major banks like Citi and Bank of America are now explicitly citing AI as a tool for “eliminating work” and reducing headcount to improve the bottom line. It some regards, I think they have to. It’s expected.

“A.I. gives us places to go we haven’t gone,” said Brian Moynihan, CEO Bank of America

In other examples, Citi has pledge to remove 20,000 roles through efficiency & technology. Zoom out and head back 20 years, you can name any technology evolution from the ATM, Mobile Banking, Mainframe, x86 servers, Internet, IVR’s to all changing where and how we work.

Steven Alexopoulos predicted an alternative view - A.I. would lead banks to an initial “profit surge,” quickly followed by a “fortune reversal period” in which customers rely on the technology to find higher-interest-paying bank accounts and cheaper loans, crimping lenders’ ability to make money and ultimately leading to mass layoffs and bank closures.

Which led me to this cheery summary from Peter H. Diamandis I spotted on substack

OpenAI ChatGPT continues to leverage it’s 800+ million active users. With Matthew Cresswell and team at MONY Group plc continuing to add new capability to their recently launched service. This week after Simply Business and Aviva both announced, it’s the time for Mortgages and NatWest Group. That’s quite teh collection of Financial Services products stacking up here now. Same questions apply though re trust from the claims piece above.

Customers Bank CEO, Sam Sidhu recently led his quarterly earnings call, well his digital clone did! Others are doing the same, including DXC Technology. I wonder who noticed here - deepfakes and voice cloning come to mind here too! Sam said he did this to prove they were not just experimenting anymore.

Elsewhere, a similar story unfolds of areas being addressed, reducing loan time, betters more accurate risk insights, faster development and much more.

“It’s going to benefit our investors. It’s going to benefit our customers,” Sidhu said. “Our regulators will hopefully also be happier over time, because they’re going to see us reducing risk as well.”

Now all this AI every where, despite our best hopes is not free (just look at the hyperscaler earnings this last week) - and many companies are in the same boat as Uber and has already burned through their AI budget for the year. We’ve talked about Tokenmaxing on here before and more.

Uber exhausted its entire 2026 AI budget in just four months. This happened because the rollout of AI coding tools (specifically Claude Code) was too successful.

You could argue the flip side of this quite easily, especially when you see the results:

Uber has transitioned from using AI as a simple copilot to an agentic model. Currently, 70% of committed code at Uber originates from AI, and roughly 11% of live backend updates are written entirely by AI agents without human intervention.

What I think all industries will ask themselves more now is - what is the ROI of AI, both now and in the future. Reason I say this is we are very much in an experimenting phase even fro the more advanced organizations. It may be in many cases, humans with AI are the right answer vs AI Agents only, both from a cost and society perspective. Reminds me of the phrase from Aviva years ago when everything was Digital Transformation. Digital First, but not everywhere.

This from Tracy Lee Kus, CEO Global Broking at Aon made me sad. Maybe angry. No not at Tracy, but the continued perception that follows insurance around. It’s dull, boring, grey or worse, pale, male and stale. Some of this possibly, is it changing, YES, is it fast enough, NO!, but I agree on one thing - it’s the very oil to society. Nothing moves without Insurance, we just need to do a better job showing what we are up to and it’s purpose.

As my former 11:FS cohost, Sarah Kocianski used to say - we have self driving cars and drones! Thats super cool!

It’s been’s over a decade since I wrote this - The Insurance Brand - Time for a Refresh. I’m not sure how much has moved on or not. I too sat at the careers fair at my sons school a while back, next to me - Greg Brown - him representing Oxbow Partners - a super cool boutique insurance consulting firm, and me - leading insurance at Google. I’ll leave you guess the rest. You are not wrong. We have work to do.

👀 Quick bites 👀

1. Former Apollo risk chief says some new life insurers could struggle in downturn

2. AI advancing faster than expected as AIG builds multi-agentic solution: CEO Zaffino AIG

3. California to begin ticketing driverless cars that violate traffic laws

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! Bonus Saturday episode. These just fill me with joy if Im honest, it’s two good friend chatting through the market and working out the impact and opportunity. This week, AI has gone mainstream and both of us are busy Vibe Coding!

Unfiltered with Matt & Nige

AI Advice Goes Mainstream, Who Owns the Interface?

Unfiltered with Matt & Nige | Episode 08 🎙️…

Listen now

4 months ago · 1 like · Matt Connolly and Nigel Walsh

And thats It! See you next week after ServiceNow #knowledge26 in Vegas!

🎙️Listen 🎙️

Is SaaS Cooked? the big question on everyones mind and that has hammered the markets over the last few months. Harry Stebbings sits down with Aaron Levie for an incredible discussion and one I’ve looked forward to since he mentioned it was being recorded. I don’t want to spoil the surprise but - we are going to use more software, not less.

Aaron emphasizes that for enterprises to truly gain leverage from AI, they must redesign workflows for agents rather than people

He also talks about AI is the new UI, especially important given the headlines around headless recently. Early on he also jumps into this - token maxing and how spend will shift.

AI spend will shift from traditional IT budgets into regular operating expenses (hello line of business) and We are about to see the creation of 500,000 to 1 million “Agent Operators”.

Harry’s takeaways here - brilliant as always.

Still thinking about about building your own Open Claw or personal Chief of Staff. This from the The AI Daily Brief with special guest Nufar Gaspar who returns to introduce Agent OS, the latest free AIDB training program for building a personal agentic operating system that travels with you across any tool, model, or harness.

Nufar talks through the key things you need, in just 30 mins, from identity to context to how you should experiment, test and iterate. Well worth a listen.

📺 Watch 📺

Another solid Cheeky Pint Episode (starting to like that title more and more, maybe next is a Cheeky Nandos! - one for my UK readers). Bret Taylor needs no introduction, but include CEO and Co-Founder of Sierra, Chairman of OpenAI, co-created Google Maps, ex FriendFeed, Quip, Facebook and invented the like button!

In this, John Collison and Brett get stuck into the Transition to Outcome Based Pricing - to re-align interests, the move to Agent Harness over Traditional API’s and finally, the Productivity Paradox in corporate structures. Interestingly, his view on a move away from specialist to generalists intrigued me - especially given we are all levelling up with AI by our side now. Worth a listen!

Simon Taylor shared good insights on the Plaid + Perplexity integration. I’m a huge fan of PFA’s and super curious about using AI for Financial insights (see FT article below too). Or as Simon Taylor says

The insight from this chat is how power users are already using personal finance agents to do REALLY useful things, is a sign of what comes next for all of us. And for anyone building consumer finance products, a roadmap.

Jeff Grimes at Perplexity and Brian Dammeir at Perplexity share more. The numbers don’t lie - 13% of people use PFM’s, 35% of people ask AI about Financial Decisions.

The weekly news show is back, Rory O’Driscoll Jason M. Lemkin and Harry Stebbings are back talking about the $60b Cursor acquisition by SpaceX and how Anthropic has hit 1T valuation in a secondary market valuation, higher than OpenAI given their success and lead in the Enterprise space. Finally the talk of the Agent Fabric which is something we have talked about a lot - we don’t need more point solutions - thats just noisy. We need AI Orchestration for the Enterprise (oh hello ServiceNow!)

Need’s no explanation - I love Jon Grays videos on LinkedIn - more running, just like you and I - less Blackstone. Quite fun indeed. 100% know your audience.

📚 Read 📚

Earlier this month, McKinsey & Company released their AI Manifesto. This from Alex Singla here. And his opening line is what I see and hear day in day out.

AI doesn’t create value. How you rewire the enterprise to use it does.

The tools are broadly all converging in different guises and ways - how we get this into the hands of users every day is what we need to accelerate - and it’s going to take a while. There’s a lot to do across all enterprise’s.

The constraint is no longer the technology. It’s the operating model. If workflows, decision rights, and system architecture stay the same, AI will improve productivity at the margins. It won’t change performance in a meaningful way.But when companies reimagine work end to end, how decisions are made, how systems interact, how teams are structured, AI becomes part of how the business runs.

In the same way, is the narrative on the AI ‘Jobpocalypse’ changing too. I’ve seen it now multiple times. A good read here from John Burn-Murdoch at The Financial Times. We are not going to need less people - but more - especially in the short to medium term - and they are also just going to be different. Job’s and roles change.

Spreadsheets automated away the jobs of bookkeepers but boosted financial analysts.

Interestingly, in the same way - ATMs didn’t replace bank tellers, but iPhones did (much later on)

My old friend Sai Vishnubhatla, CFA over at USAA Inc wrote a really good piece on Agentic AI in insurance and why its harder than many expect. I think this is fair for most regulated industries. I also agree on the point with regards to vertical vs horizontal, and why we win here.

That is what I call The Deposition Problem.An adjuster can be deposed. An underwriter can testify.An actuary can defend their methodology. An AI agent cannot.

Importantly, we need to rethink and reimagine, not just bolt on.

There was a lot of outrage last week with Meta looking to use tracking software. However, I’m with Tom Goodwin here. Especially as people have pushed for the return to office post Pandemic (yes we are still talking about that) for the purpose of creativity and connection. I’m 100% an office guy, i love people, the commute the thinking time, the sights, sounds and smells.

Having worked with Contact Centers for decades, where AUX time was tracked (coffee, loo, etc) I don’t mind what people track (but you would need to look at most evening’s too to get the full picture) - what I often pause for is thinking or writing time or that brilliant coffee or water cooler moment. Not sure where those keystrokes are tracked.

I think this was probably always inevitable, albeit how it got there, is quite amazing. Caroline Hyde and Rachel Metz over at Bloomberg share more here.

By making an educated guess about the model’s online location based on knowledge about the format Anthropic has used for other models,”

Another insurer joins OpenAI, this time the turn of Travelers for Small Business Insurance across the USA and UK. There is now a whole plethora of insurers here and while I agree distribution has changed and will continue to, I’m really curious to how many insurance customers are actually here now. Either way, looking forward to seeing how this changes the servicing and acquisiton landscape over the coming years

’Launching this insurance app in the US and the UK for small businesses in ChatGPT is our latest step in meeting our customers where they are,’ says chief executive

This was a great share from ServiceNow‘s Chief People and AI Enablement Officer Jacqui Canney - a story by Angela Haupt at TIME that talked through how leveraging AI heavily erodes confidence. However whats more interesting is this;

Participants who pushed back—editing, questioning, or rejecting AI-generated suggestions—reported the opposite: greater confidence and a stronger sense that the final output was truly theirs.”

We are not outsourcing thinking here, the opposite if its used and considered in the right way - we are accelerating our ability to process and think, while maintaing the muscle to pushback, challenge and much more.

👀 Quick bites 👀

1. Insurance Act fails to curb disputes as litigation jumps 180% in a decade: Mactavish report

2. New York sues Coinbase and Gemini over prediction markets

3. Google to invest up to $40bn in Anthropic

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! Later than usual this week (my fault), but fun as ever!

Unfiltered with Matt & Nige

Unfiltered with Matt & Nige is a straight-talking podcast about insurance, innovation, and everything in between. From tech trends to family life, training, and life on the road.

By Matt Connolly

And thats It! See you next week! 🚀

🎙️Listen 🎙️

I’ve not shared an InsTech podcast for a while - but the energy and sheer joy in Sasha Haco voice is brilliant. I even pinged Robin Merttens and Matthew Grant as I do on a regular basis. Also great to see an Entrepreneurs First success story too. Well worth the listen to hear what is going on at Unitary and congrats on the early success here!

The The Voice of Insurance himself, Mark Geoghegan who sits down with Sean McGovern, Deputy Chair of Lloyd’s and CEO AXA XL, or as Mark says;

Today’s guest is one of the best qualified people in the London Market.

Particularly interesting post the evolution of Lloyds given the new on BP2 recently. From Moving beyond the binary of Lead and Follow where enhanced underwriting allows followers to rely more on the Leads, removing duplication to prioritising Market Simplification and Digitisation, specifically the importance of the Core Data Record (CDR). Well worth a listen! (one for you Pierre Njeim, Cert. CII Liam Sweetman John Malivoire)

The Market of the Future is not just about technology, but about maintaining its status as the worlds hub for complex risks.

I’m also still really enjoying The AI Daily Brief thanks to Richard Gunn. Just short sharp, good insights on whats going on. Check it out here, simple quick listen - this one on how the best companies are using AI (and the gap between leaders and laggards), others on the ROI of AI, metrics, news and so much more!. Super helpful on your commute, dog walk, treadmill.

📺 Watch 📺

I think we have been entering the ‘claudifciation’ phase of AI, Matt Connolly and I spoke about it on the pod last week. Not many meetings start these days without somone saying, I’ve put this through Claude or.. More on this another day. t’s the default go to as it just works and creates visualisations and UI to make things understandable. Answers alone is not enough OpenAI.

I found this super interesting (I like apps, fitness and AI) and shows Peter Yang building a mobile fitness App in just 16 mins on Claude plus Pencil. Super cool! Is this the beginning of hyper personalised apps for us all?

The Anthropic Mythos release has caused much stir and interest. Genius, Marketing Hype or both. I love this debate show as you know from Harry Stebbings Rory O’Driscoll and Jason M. Lemkin - and as Rory states, as you discuss stuff - he often changes his mind as you talk through these. It’s a great way to learn in my view - I have strong opinions, loosely held.

Mythos certainly caused a stir - was it not released because there is not enough compute, or it’s too dangerous in the wrong hands or... either way, I think it’s seen unprecedented response from the USA and UK regulators and government’s given the number of legacy zero day issues identified - some dating back 27 years.

Machine Gun vs. Rifle: While older models could find bugs with human steering, Mythos can agentically scour massive codebases at incredible speed. This shifts the threat from a “rifle” (targeted) to a “machine gun”

This plus loads more - from OpenAI Ad’s to the SaaS trap. I liked Jasons POV here both on hiring and the 60% test.

The “Jason Test”: If an AI agent is only 60% as good as a standalone native AI tool, the company cannot charge extra for it; it must be bundled for free

Travis Kalanick re-joins the All-In Podcast crew - the part that most interested me here was the ongoing Ban of Data Centers - now across 30 states. Are tides turning on AI firms and Data Center builds from a society perspective?

Local Resistance: Data centers are becoming a target for populism and NIMBYism. Across the U.S., projects are being voted down or banned (e.g., Maine’s recent ban) because of fears over electrical rates, water usage, and the perception that they only benefit the “rich elite”

📚 Read 📚

I love this from Union Square Ventures. Fred Wilson shared that Spencer Y. had just joined and in this post shares who are the agents doing all the work, they all have names! I actually spoke to a good friend over the weekend who had done something similar. Importantly they are designed to fit with how USV works.

Completely aside - also made me pause and think, how many agents are now just responding to comments on LinkedIN - hmm, more slop or useful conversation or slaves to the machines? What are we actually engaging in sometimes?

This is how I use/view AI today. An accelerant that raises the floor, not just the ceiling. You know what Im going to say and what Im going to pick up on - so why not codify this so we use the time to discuss/debate/evolve and much more. Much work to be done, but I love the thinking and idea here.

The biggest time-waster at tech companies isn’t meetings, Slack, or writing docs. It’s trying to predict what your leader wants.

I think we will see every organization do this - how long for, I don’t know but for now - it’s all in or nothing. AIG no different with Peter Zaffino sharing that they are doubling down for their complex risk landscape. The complexity right now from Geopolitical risk, increased catastrophe activity and a need for better capital management - lead to a stronger pull to AI. I fully share his sentiment.

Before my brief pause, the sheer pace and evolution of new things was just mind blowing - so actually nice to unplug, process and come back. I wake up each morning thinking - what a time to be alive!

The pace of technological change over the past year, noting that breakthroughs in large language models, computing power, and agentic and autonomous AI are fundamentally reshaping how risk is assessed and managed, developments that Zaffino said are “unlike anything we have seen in our lifetimes.”

From an enterprise perspective, I always love Aaron Levies, CEO of Box perspective on enterprise software and adoption. Here he talks about a new type of role in organizations going forward, in the era of Agentic Transformation

This person will need to figure out what are the highest leverage set of workflows on a team are (either existing or new ones) where agents can actually drive significantly more value for the team and company.

This also from Daniel Hulme, Chief AI officer at WPP - simply borrowing his intro here and building on his 28,000 agents in production today. Worth a read.

People talk about agents as if the hard problem is building them. It isn’t. The hard problem is verifying them. Knowing what they’re actually doing, whether they’re doing it well, and whether they’ll keep doing it well when conditions change.

Great to see this from Pierre-Alexandre Kamienny on the 113 Insurance startups across 17 cohorts backed by Y Combinator. He shares three mega trends across these below. Full report here. Of these only 9 inactive but also 9 acquired! One for your Florian Graillot Sam Evans

1. Covid tripled the pace.Before 2020: ~4 insurance companies per year. After: consistently >13.

2. The US is in a league of its own. 98 of 113 companies operate primarily in the US. India is second with 5.3.

3. Two mega-categories are emerging. AI-native MGA/carriers, and AI-native brokers, in between tooling.

AGI is one of those topics that either excites or scares people. Sticking with Union Square Ventures, Albert Wenger shares his Economic Model of AGI which gives a good summary of what the future might look like when we do get there.

This is a first-cut model meant to build intuition and invite discussion, not a calibrated forecasting tool. Also note; it is entirely vibe coded!

Either way, a super interesting read and model, and this is just the very beginning of the debate.

My intuition that you need both competition and redistribution via a basic income or negative income tax, and that neither alone is sufficient seems to be supported by this model. Looking forward to feedback!

The never ending question here - what do my kids do, regardless of their age. What should we teach them now and in the future. I’ve always promoted - creativity, curiosity and story telling - and stand more and more firm by this every day. Like Keith Zhai, I have young(ish) kids, almost 17 and 14.

What is going to make a human being valuable ten years from now?

I like his three things - Imagination (similar to my creativity & curiosity), Taste and Grit. I’ve heard taste come up a few times now in this debate and really like it. Grit reminds me of one of my fav quotes, that will 1000% required in the age of AI.

Passion may fuel your fire, but tenacity will get you to your destination. We need both.

He also quotes James Dyson of Dyson fame - noting it took him 5,127 prototypes to get the bagless vacuum right. But closes out beautifully.

The machines are getting better at everything. Which means the humans need to get better at being human.

Finally, all the advice points to good sleep for mind, body and soul (looking at you Mark Gammon!). However, CEO’s are now blaming sleepless nights on their own senior team!

For CEOs of companies with revenues over $5bn, the biggest source of anxiety is their own “ego-driven” C-suite.

importantly, a constant state of fire-fighting prevents them from focusing on the firm’s overall long-term strategy with 60% of CEO’s saying that immediate, near-term issues and time-sensitive deadlines are “all-consuming.” Feel’s more true now than ever before.

👀 Quick bites 👀

1. FCA criticised over using sensitive data in AI trial with Palantir Technologies

2. Nearly 160,000 uninsured cars seized on UK roads

3. Not all AI agents are created equal from Hamza Farooq & Jaya Rajwani

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I! Ok we are rubbish, but also having way too much fun. Take 2, Take 3.. no - Matt wont lets us - ... straight in - goofballs and all!

Unfiltered with Matt & Nige

Peak AI Slop: When AI Starts Talking to AI

Unfiltered with Matt & Nige returns 🎙️…

Listen now

5 months ago · 3 likes · Matt Connolly and Nigel Walsh

🎟️ Join Us 🎟️

Tickets are selling going quickly - join us in NYC at our Hudson Yards HQ on the 19th May with our friends at InsTech and partners NAITIV, Concirrus, INSTANDA and hyperexponential thanks to all the hard work from Susi Ozkurt Robin Merttens and the teams! Register here

The future of Insurance will be orchestrated not built.

🚶 And Finally 🚶

And finally, a reminder for our upcoming charity marathon on the 9th May in London for Insurance United Against Dementia and Alzheimer’s Society. What was supporting it, is now in memory of my little old Mum. Proud of these guys, Oscar Walsh, Sam Collins and Leo for doing it (yes I’m joining them along with Emma, Adam and Nicola).

While we have smashed our target, were not stoping there and will keep raising funds to battle this cruel cruel disease. Link below if you would like to donate,

Team Scrambled Legs - here

And thats It! See you next week! 🚀

🎙️Listen 🎙️

I’m a huge Bill Gurley fan, so really enjoyed this episode. Ordered his book (Running Down a Dream) while listening immediately. Looking forward to getting into this.

Gurley argues that while the first wave of AI was about the “cool factor” and new startups, the real financial value is shifting toward established incumbents.

Importantly, curiosity (my fav word as you know) and the ability to learn how to use these tools to 10x your output is now more important than a static skill set.

This is a new one for me, but when i saw the title - and both Adam Grant and Brené Brown it was a clear must listen!. Although, not 100% convinced on the first one’s title - Mission vs. Ego: The Dangers of Narcissistic Leadership, it actually turned out to be super interesting - recorded live at SXSW.

Everything from Shame based Fear, Mission over Ego to the competence trap. I also really enjoyed the point about everyone Brené Brown met at St. Jude Children’s Research Hospital and how everyone started with, ‘I’ve been curing cancer for ... ‘ Worth a listen!

You can identify a healthy leader by their ability to say, “I don’t know,” and their willingness to let the mission—rather than their own image—dictate the strategy.

I’ve talked about this a lot previously - we don’t teach kids the basics - when it comes to Life Skills - e.g. changing a plug, changing a car tyre, how do use a bank accounts, what’s Insurance for!. I was surprised to hear Ted Dintersmith share with Scott Galloway were not even getting the AI part correct right now!.

Ted covers loads, from Schools are Optimized for the Wrong Century. The current system prioritizes rote memorization, standardized testing, and following instructions. However, in an AI-driven economy, these are the exact skills that are being automated.

Education needs to move away from “content delivery” and toward “creative problem solving” and “critical thinking,” which are the only areas where humans still maintain a competitive advantage over machines.

By offloading the commodity work (summarizing, basic drafting) to AI, students can focus on higher-level tasks like ethics, strategy, and cross-disciplinary synthesis. I couldn’t have said this better myself!

If schools ban AI, they are essentially graduating students who are “illiterate” in the most important tool of their future careers.

📺 Watch 📺

Is AI growing up, or have as the guys said - the accountants entering the room? This week with OpenAI killing Sora and Anthropic hitting a $6bn month (not a typo). Interestingly, AI Agents are being used in ways we hadn’t wanted too - e.g. the data leak at Anthropic regarding its new model - Mythos - meaning AI agents are creating a mega-threat for enterprise security - when Mythos itself was being promoted to CISO’s and more.

In this episode of Venture with Grace, our very own Amit Zavery sits down with Grace Gong for a conversation on The Future of Work: AI, Ethics, and the Human Element. One thing I firmly believe to be true is the point around AI as a collaborator, not a competitor who is going to take my job. We need to bring the Human Element, empathy, ethics and nuance that AI is not replicating (remember the post recently from Mustafa Suleyman on Empathy).

The common theme here from the Bill Gurley interview, Ted Dintersmith and this is the focus on both continuous up skilling and lifelong learning.

I guess I know if i need you Amit Zavery, then I’ll come find you in the Pickleball court 🎾

A little fun break (see what I did there!) - if you’e been on socials recently, you will have seen the Kit Kat drama that unfolded, as a truckload of over 400,000 F1 themed Kit Kats were stolen on route from Italy to Poland. The Internet has gone wild and Kit Kat themselves are having some fun too!, check it out below (just one of many examples) on Instagram!

How to turn around a bad situation! Have you tracked yours?

📚 Read 📚

A fascinating summary from Simon Taylor on how the CEO of Brex, Pedro Franceschi runs the company. The business was acquired by Capital One for $5.15b earlier this year. Is this the future for all employees? here

Pedro Franceschi built an autopilot on OpenClaw that answers that for him. It screens everything and organizes his world around two concepts: people and programs.

Speaking of processes and support, Colin Payne over at the Financial Conduct Authority shared this from WIRED - I, Rulebot: How AI is turning compliance into software

Policy as Code is the practice of translating an organization’s internal rules, legal requirements, and operational controls into machine-readable code.

Importantly, this brings Real Time Logging, Continuous Screening and Observability. Aligned clearly to what ServiceNow is doing with our AI Control Tower and why it’s been such a hit with regulated industries. One for Anna Mazzone and Alexis Bliek

Greg Brown and the brilliant team over at Oxbow Partners shared their take on the lates Lloyd’s five-year strategy. A return to basics. As always, what matters now is execution, execution, execution.

The Corporation therefore resumes its role as enabler rather than a disruptor of risk placement.

The Lloyd’s strategy sets six statements for its 2031 ambition:

• Be recognised as the most capital efficient marketplace for complex risk

• Command a share of wallet proportionate to its inherent strengths

• Operate across the full spectrum of commercial and speciality risk

• Be predictable, transparent and investable

• Retain and attract global talent and sophisticated capital as a matter of course

• Sit at the forefront of innovation in underwriting, capital and oversight

One for you Pierre Njeim, Cert. CII John Malivoire and Christian Bowyer

With the ongoing AI Boom, the demand (and cost of building) Data Centers has dramatically accelerated. I heard recently that the cost was in excess of $10b per Gigawatt and rising.

Global insurance premiums linked to data centres are expected to rise to $24.2 billion by 2030 from an estimated $10.6 billion currently

The lack of capacity may be one of the limiting factors here for future build outs.

“They’re not just buildings, they are highly integrated systems where you have power, cooling, hardware, and software that all depend on each other,” Swiss Re’s Head of Engineering and Nuclear Jimmy Keime told Reuters.

This concentration makes a single data center site a potential systemic loss event for an insurer, comparable to a major city-wide catastrophe. Add in traditional risks, Hail, Tornado and more and we have a perfect storm (pun intended)

The TL;DR here from Ethan Mollick, don’t treat AI like every other enterprise wide software platform - it’s weird (and thats a good thing!)

If the IT department is the only gatekeeper for AI, the technology will be reduced to boring, low-value automation rather than a competitive advantage.

We talk a lot about guardrails and governance, without this - we are strangling the potential of AI by forcing it into the rigid, risk-averse frameworks of traditional IT management.

Meanwhile, in the EU, staff have been banned in using AI for Images and Videos. The primary driver behind this ban is the need to foster and maintain “citizens’ trust.”

The European Union’s main institutions have banned staff from using artificially generated videos and images in official communications.

It’s been met with mixed reactions. “Responsible use beats abstinence,” said Walter Pasquarelli an adviser to the OECD who also researches AI-generated content at the University of Cambridge.

👀 Quick bites 👀

1. MAS develops AI risk management toolkit

2. Mass robotaxi malfunction halts traffic in Chinese city

3. ‘From high flyer to dead parrot’: former billion-dollar eco-shoe brand Allbirds sold for $39m

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and I!

Unfiltered with Matt & Nige

AI Is Moving Fast. Adoption Isn’t. The Expectation Gap Is Getting Dangerous

Unfiltered with Matt & Nige is back 🎙️…

Listen now

5 months ago · 1 like · Matt Connolly and Nigel Walsh

🎟️ Join Us 🎟️

Finally, join us in NYC at our Hudson Yards HQ on the 19th May with our friends at InsTech and partners NAITIV, Concirrus, INSTANDA and hyperexponential -

The future of Insurance will be orchestrated not built.

And thats It! See you next week! 🚀

🎙️Listen 🎙️

A great start this week, if you have seen my LinkedIn feed, then you know I’m a huge fan of Waymo and am super excited that I see them regularly now around the streets of London. So was delighted to see and listen to this Cheeky Pint episode with John Collison and Waymo Co-CEO, Dmitri Dolgov.

Whilst driving is pretty much done, this episode covers everything form the new Waymo Driver and Vehicle to things that still cause them challenge (Cold weather across teh stack!, hello London!), to launching 4 cities in 1 day!. With 3,000 Waymo’s on the streets covering 4 millions miles per week! If your an AV nerd, worth the listen (looking at you Matteo Carbone)

Building on our $10bn GTM playbook (below), this was also interesting from Harry Stebbings conversation with Shaunt Voskanian, CRO Figma. Some great insight here, and as always - Harry put together a great summary/key takeaways here. I liked this: Move from Curious to Prescriptive. I also liked #5, but I think you need both, especially in the enterprise space.

While curiosity is essential, top-tier enterprise sellers must be prescriptive. Customers are busy and want insights—they want to be taught what your best customers are doing that they haven’t figured out yet. Building a champion today requires educating them and bringing a point of view that makes their team more effective.

📺 Watch 📺

In the news cycle, it feel everything has centered on Anthropic, OpenAI and Google. The likes of Llama from Meta Mistral AI and Perplexity have all but disappeared. At least from the headlines. That changed with Perplexity launched the Perplexity Computer, a digital worker in the cloud. Interestingly, CEO Aravind Srinivas talks about the importance of Orchestration over consolidation here.

Like most things these days (eg ClawBot), whatever gets released - Anthropic seem to be matching super quickly with Claude.

And while watching this, it autoplayed into this.form Aaru and co founders on CNBC and just wow. Interesting, they roll into the impact of GLP1’s and alcohol consumption too. Cam and Ned, at just 20 and 21, the company already has a billion dollar valuation.

Aaru aims to replace traditional human surveys and market research with digital simulations.

If this is our future generation, them I’m super excited and upbeat, including their giving back and pro bono work

Benedict Evans sits down with Matt Turck on The MAD Podcast / Data Driven NYC and talks about OpenAI’s moat and strategic challenge ahead! I also like the bridge analogy here and how

AI can automate these “fuzzy” tasks that were previously too small or complex to justify a dedicated SaaS tool

The challenge still exists of course to go beyond pilot, and into production at scale - 1997 again anyone?

Final watch this week from our very own Paul Fipps on what it takes to build a $10bn GTM engine. I loved this (Yes I’m biased) and nails it from the outset - we have to be obsessive, in the same way folks have heard me say fall in love with the problem, not the product, as well as how we helping to eliminate Soul Crushing Work.

Organizing Around the Customer, Not the Org Chart

📚 Read 📚

This from German magazine, IT Finanzmagazin has a good summary of how Agentic AI devours insurer back offices: Agents automate underwriting, claims, and incoming mail with insights from HDI Group, Thomas Kuhnt, DEVK Versicherungen Markus Kalkowski , Gothaer Solveig Schulze HanseMerkur Horst Karaschewski , ERGO Georgina Neitzel and many more share their insights and perspectives. I love this concept. A phrase Ericson Chan from Zurich Insurance used in his keynote last week too at Insurtech Insights

In 2026, insurance companies are moving away from isolated digital tools toward an integrated “AI operating system”

One for my friends Kerstin Weihe Katharina Leypold Dr. David Reinisch Julia Deibert Jonas Wiest and the entire team!

Speed is crazy right now - founded in 2024, Dreamer already acquired by Meta - Dreamer was set up to help build Personal AI Agents. What intrigues me here, is not just the talent acquisition, but the length’s BIgTech is going to bring people back to or keep people on their platform. I feel AI and AI Agents have enabled the era of letting a thousand flowers bloom!

The Founder + CTO at HubSpot, Dharmesh Shah has a good take here too on this.

I think the headline says it all here from Alex Karp at Palantir Technologies:

Palantir’s billionaire CEO says only two kinds of people will succeed in the AI era: trade workers — ‘or you’re neurodivergent’

Importantly, success will move away from those who follow a standard “playbook” or academic path toward those who possess unique, non-automatable expertise. This is why I say to people curiosity is a superpower for this and the next generation. From a physcial perspective, there are many things that AI or Robotics cant touch yet - even things like the last mile delivery (at scale) - so while you might be able to drive a truck across the country in autonomous mode, running up the stairs or going through a porch etc. not yet

Equally, neurodivergent people are naturally artists who look at problems from unique angles because they cannot easily follow conventional rules.

This from Julien Bek, Partner at Sequoia Capital makes a fascinating read.

The next trillion-dollar company won’t sell AI tools. It will sell the work itself.

The shift from Co-Pilots (helping you) to AutoPilots (doing it for you) is amongst us, eg ServiceNow Autonomous workers and agents. Every founder building an AI tool is asking the same question. That said, for years people have said the same about Google, Amazon, Facebook etc - Having worked on the inside, I think theres much less risk of that, more of them hiring the people that can do it (as seen here).

what happens when the next version of Claude makes my product a feature?

Interestingly, the largest area identified here is - Insurance Brokerage. Make what you want of the following. Just because it could, doesn’t mean to say it would - and certainly not flicking a switch overnight - especially in this instance. It’s a people and trust based business.

Insurance brokerage ($140-200B). The largest dollar market on this list. Standard commercial lines are highly standardised: the broker’s value-add is essentially shopping across carriers and filling forms, pure intelligence work. The distribution layer is incredibly fragmented, tens of thousands of small brokers each running the same process, so no single incumbent controls the customer relationship. WithCoverage and Harper are interesting newcomers.

Finally, I woke to two posts - one from Brian Solis talking about what we really want Robots to do here, and another showing them actually doing that very thing from UniX AI. As the old saying goes:

The future is already here, it’s just not evenly distributed

👀 Quick bites 👀

1. In the latest line of insurers to join the race - Aviva Launches on Chat GPT App

2. Buffett’s Berkshire takes US$1.8 billion stake in Japan’s Tokio Marine

3. Congrats to my friend William V. Powers at Cambridge Mobile Telematics who secures $350 million from TPG, Allianz to make driving safer

🎙️ Tune in 🎙️

Tune in to the latest episode of Unfiltered, with Matt Connolly and Nigel Walsh

This week: Insurtech Insights After Dark: Scale, Sprawl and Going Beyond Panels and I even get a mention of Crocs - just ask Matt Connolly what he has on his feet!

Unfiltered with Matt & Nige

Insurtech Insights After Dark: Scale, Sprawl and Going Beyond Panels

Unfiltered with Matt & Nige is back 🎙️…

Listen now

5 months ago · 2 likes · Matt Connolly and Nigel Walsh

And thats It! See you next week!