Special thanks to our sponsor Sidekick Lab!

Sidekick Lab gives your enterprise a living, governed understanding of your data estate.

Your data can finally speak the same language as your business.

• News

• The Tell

• Pods

• Charts

• Final Thoughts (how do you sell to a machine?)

#1 – Charles Stack published The App Store With No Toll Booth. August 2026.

• My Take: Interesting take on how Anthropic is structuring the future of selling. Selling to enterprises is hard (see the podcast we reviewed this week) … the best product doesn’t always win. It is the sellers who figure out how to get through procurement.

• It will be interesting to see what will happen when Anthropic “owns the river” and increases the price for workflows reliant on the water flows.

• We are early, it will be fun to watch!

#2 – Andy Graham published Why do Chief Data Officers only last 2.5 years. August 2026.

• My Take: 2004 was the first time a person was entrusted with the title “CDO”. People still can’t figure out exactly what that person does or is supposed to do. Whether it is competition with other roles, unclear objectives, or outlandish expectations, this role is a tough one.

• One way to make “what a CDO does” more obvious is to tie it to revenue. Commercialize the data. Make it obvious to the board that you are adding value in the form of high margin incremental revenue. This is way more exciting than improving governance or enhancing MDM procedures.

BONUS: (related to Andy Graham article above) Sadie St. Lawrence published The CDO Role Is Dying. What Comes Next Is More Interesting. April 2026. “The identity of the CDO is not disappearing. It is becoming something bigger, something more consequential, and something that requires a completely different kind of leader than the role originally imagined.”

What else I am reading:

Carlyle-backed YipitData weighs sale at valuation of up to $3bn. August 2026.

Doug Garber published Top 10 Takeaways from BattleFin Montauk–The Edge of Alt. Data & AI. August 2026.

Didier Lopes published OpenBB belongs to everyone. August 2026.

Dan Entrup’s Another Week, Another 673 Companies Now “AI-powered”. August 2026.

Shubhangi Goel published Using synthetic data for AI training is ‘a big mistake,’ says AI pioneer Rich Sutton. August 2026.

Liz Brandt published When Your Data Is Their Profit: Why the Personal Data Economy Is at a Crossroads. August 2026.

Source: Lenny’s Podcast published a conversation with Jen Abel. 84 minutes of enterprise sales alpha. August 2026.

• My Take: This is not directly a data podcast. But getting data & data related products into market is something on which I focus. Enterprise selling!

• I have learned a lot about selling from listening to Jen (her 3^rd time on Lenny’s Podcast!) & following Jen’s X handle.

• In the age of easy-to-build products … selling/distribution/GTM becomes the highest value part of building the business. It can be hard. It can be fun.

• The key is to listen. Enterprise sales is an art, not science. Develop rapport. Gather a lot of information. Get your champion invested.

• I posed a question via X, she addresses that question at 9:30:

• Two entry points – 1. at the very top or 2. finding the individual one step removed from that person.

14 STEP ENTERPRISE SALES PLAN

Getting in (JF: this is the hard part!):

1- Land the meeting, (your job as an enterprise salesperson is to know who to talk to)

2- Intro call, (collect intel)

3- Follow-up call, (get the right people on the call)

4- Prep the pitch/frame for the demo (the more advanced intel you’ve collected, the better)

Proving it (JF: the product must be good):

5- Run the demo, (should be an hour meeting)

6- Post-demo debrief (text your champion right away)

The pilot (JF: pick the right testers)

7- Identify the pilot process, (free if it is a quick PoC, paid if it is a long PoC … credit the PoC trial fee back if contract signed)

8- Prep for it, (define success)

9- Run it, (focus on a small number of power users)

10- Post-pilot session (collect the intelligence from the trial users … how did it go?)

Getting paid (JF: this is where things stall … need to navigate this)

11- Papering prep and procurement timeline,

12- Papering review,

13- Papering procurement process,

14- Signature!

HIGHLIGHTS (84-Minute Run Time)

• Minute 03:00 – interview starts

• Minute 09:00 – getting in & who to target

• Minute 24:00 – every pitch is unique … the art of getting to know the prospect

• Minute 30:00 – Runs through each of the steps

Source: a16z. AI Agents Use 5x More Tokens Than Humans. August 2026.

• JF: more evidence the future will be selling to machines (or selling to enterprises…then selling to machines within those enterprises) … just wait until agents get good at using more data. You’ll see a similar spike in data “usage”…this will go far beyond web scraping into commercially available data.

How do you sell to a machine?

I am in enterprise sales. I really enjoyed the Jen Abel podcast highlighted above.

Going forward there is going to be a huge market for selling data to machines (or machines within enterprises).

Most data companies are positioned to sell to enterprises (long sales cycles, big contracts) … but this is going to change.

This is a bit of a repeat message from ADW #292 “Feed The Machine,” May 2026: the primary consumer is shifting from a human operator to a machine, and the non-human buyer market will be way bigger than the human one.

One buyer takes nine months. The other takes nine milliseconds.

How do you build products & processes for both?

Subscribe now

Special thanks to our sponsor Sidekick Lab!

Sidekick gives your enterprise a living, governed understanding of your data estate.

We provide the ontology for your data to finally speak the same language as your business. (JF: this is a really cool offering, the free demo is quick & the value is quickly obvious).

• News

• The Tell

• Pods

• Charts

• Final Thoughts (The Trust Stack)

#1 – Karan Vora published Capital Intelligence. August 2026.

• My Take: The line that resonated with me: chat is an interface, research is a process.

• An LLM can explain how to compute a Sharpe ratio, and it can write the code, but you should never let it be the place where the math happens. Language models guess. Guessing doesn’t work.

• Karan also highlights a related problem: once machines can crank out 100,000 backtests overnight, the bottleneck moves from intelligence to verification (trust in LLM output is a common theme in recent ADWs).

• When analysis gets cheap, trust in the analysis becomes extremely valuable. Hard to earn. Easy to lose. (We are living this at SymetryML, where the AI queries our DEM instead of expensively & slowly re-scanning your raw data, and every answer comes back auditable.)

• Trust is the whole game.

#2 – John Lathouwers published AI hasn’t driven up market data costs yet. That’s about to change. August 2026.

• My Take: How this plays out will be interesting to watch.

• Market data costs are up, and AI hasn’t even really started pulling on the demand side yet. More agents. More queries. Agents pull more than any human ever could.

#3 - Google wins bankruptcy auction for Spirit Airlines emails, chats, documents. August 2026.

• My Take: This has, justifiably, been a much discussed transaction. I’d love to have been a fly on the wall as the bankruptcy attorney’s went through the process of liquidating this data asset. Seems like $10m is low. But the devil is always in the details with data transactions.

• Perhaps, when we look back as this widely publicized event in a couple years, this will be viewed as the kickoff of every firm at least going through the process of evaluating what their data is worth.

What else I am reading:

Mark Leher published Why Market Segmentation Matters for Community Banking. August 2026.

Lior Barak published How Much of Your Team’s Week Actually Goes Into Getting Data Ready to Use? August 2026.

Jae Leeholland published The Next AI for Finance Frontier: Investment Intelligence OS. August 2026.

Andy Hannah published University Research Data as a Strategic Asset. August 2026.

Kelso Holland and Jackson Hejtmanek published The Trucks Are Talking: Federally Mandated GPS Data From America’s Big Rigs as a Valuable Wall Street Signal. August 2026.

Jamie Comtois published Why Over-relying on AI For Stock Picks is a ‘Recipe For Disaster’ (David Trainer). August 2026.

“The Tell” is a one question survey I will include every week.

Source: The AI Daily Brief published AI Optimism Has a Trust Problem. August 2026.

• My Take: The “trust” word caught my eye given how frequently I am seeing this topic.

• We are seeing the big companies shift messaging from “AI will destroy the world” (…so stupid) to AI is for everyone!

• Basically we are seeing people have low levels of trust of tech executives. “…the tech industry just doesn’t actually understand normal people.” (minute 12:30). This is true.

• And these people have so much power that they need to be trusted, and hopefully are worthy of our trust.

• Creating new things vs automating existing things.

  • I have had conversations with business owners who are focused on using AI for creating new things (products, etc) that they couldn’t have created pre-AI …rather than focusing on the automation of existing processes. More value is generated in the creation of new things rather than the improvement of old things.

• Human ingenuity for the win!

HIGHLIGHTS (19-Minute Run Time)

• Minute 01:00 – messaging from the companies is changing

• Minute 03:45 – Human capabilities will outpace automation

• Minute 06:00 – Gov’t involvement

• Minute 07:30 – Data center communities

• Minute 09:00 – Importance of open source

• Minute 11:30 – Trust of AI executives (largely the same guys as social media execs … yikes!)

• Minute 12:30 – Living a life

• Minute 17:00 – Deep access to personal context makes AI better … but feels invasive

SOURCE: Bessemer Venture Partners published The rise of vertical AI-native services in India. August 2026.

• My Take: Interesting thought piece on where value is accruing and why.

BONUS: Asymmetrix published their H1 2026 Data & Analytics Outlook. August 2026.

Every data-driven decision has to clear three gates.

1. Can I trust the data?

• The oldest question in this industry, and it is getting harder. Lior Barak asks how much of your team’s week goes into getting data ready to use. Too much. Meanwhile, John Lathouwers points out that agents are about to pull on data at volumes no human team ever could. An analyst squints at a number that looks off. An agent does not squint (they are sometimes annoyingly detail oriented).

• Garbage in, garbage everywhere, faster.

2. Can I trust the math?

• Chat is an interface. Research is a process (Vora). An LLM can explain the math and even write the code, but it should never be the place where the math happens. I live this at SymetryML: the AI asks the question, a statistical layer computes the answer, and every result comes back traceable. That is what makes testing thousands of models overnight useful instead of dangerous. Volume without verification is just noise at scale.

3. Can I trust the call?

• The gate machines cannot clear (this is why there will be an increase in human jobs due to AI).

• McKinsey does not get paid for the deck. McKinsey gets paid because a name stands behind the deck when your board asks why. A model cannot sign its work. It has no reputation to lose.

• Jamie Comtois calls over-relying on AI for stock picks a recipe for disaster, and this is why: every call needs an owner (h/t David Trainer).

• Ivy Zelman, who I wrote about in Advising Midas, made her name calling the housing top while everyone stared at the same data. Knowledge was already free in 2006. The call is what got paid.

AI is making the middle of that stack nearly free. Anyone can produce analysis now. Which means the premium is migrating to the gates: inputs you can verify, math you can audit, and a human who owns the call.

Knowledge is free. Wisdom is not.

Special thanks to our sponsor Sidekick Lab!

Sidekick is the Agentic AI layer that gives your enterprise a living, governed understanding of your data estate.

We provide the ontology for your data to finally speak the same language as your business.

• News

• The Tell (new section!)

• Pods

• Charts

• Final Thoughts (the barbell future)

#1 – Abraham Thomas published Looks, Brains, and Money. August 2026.

• My Take: This is Part 2/2 on data quality. Find Part 1/2 here.

• Garbage in, garbage out was easier to police pre-LLM, because the garbage looked like garbage. Now it comes back polished, cited and confident.

• There are a lot of good tips in here for how I use LLMs (i.e. Do NOT let your LLM write to your source of truth, use raw inputs, tell LLMs to check their work, long sessions degrade, etc).

#2 – Sayede Salim Mohamed Ahmed published Data Monetisation Is Not the Same as Data Value Realisation. August 2026.

• My Take: I have talked for years about how organizations looking to “monetize” their data (i.e. sell it externally & drive new revenue) would benefit tremendously from going through the process of organizing their data for sale. The benefits of organizing your internal data are enormous, even if not tied directly to new revenue.

• Value = data + context + quality + accessibility + permitted use + organizational action.

#3 – Keith Rinzler How to turn synthetic data into strategy-grade insights. August 2026.

• My Take: We should use a combination of human and synthetic audiences. Getting high-quality feedback from real humans is extremely valuable. We should always query human customers / experts. Feedback from real humans is not replaceable. But it is expensive & time consuming (AlphaROC is changing this game).

• Once built, a synthetic audience is cheap to query (AskRally does good work here).

• Using the two in combination is powerful. You can prepare for human conversation by asking a synthetic panel thousands of questions. You can ask your synthetic panel to react to competitor ads or new media, think about a new product offering, or just react to a post your social media team is considering. Low cost and even if it is less than 100% accurate, it is so cheap relative to human surveys, it is worth it.

• Each question you ask a real human is of high importance, you only have so many shots, so you don’t want to waste that opportunity. Combining your efforts with real humans with synthetic audiences delivers better results.

BONUS: Check out TableStakesData for a searchable index of every ADW published since 2020.

What else I am reading:

Karan Vora published Capital Intelligence. August 2026.

Dan Entrup published Universal Ignorance. August 2026.

Dan Poppy published Data lessons from inside Meta (Shridhar Iyer). July 2026.

Avalon Zoppo published 9th Circuit Allows Use of Perplexity AI Agent on Amazon. August 2026. (complete ruling).

Saeed Amen’s What I’ve learnt from 13 years of working independently. August 2026.

Alliance for Responsible Data Collection (ARDC) published Request for a veto of the Stealth Crawler Act. August 2026.

Found this cool resource from Axel Stahmer: The Atlas of Alternative Data

“The Tell1” is a one question survey I will include every week.

Source: Signal Drop podcast published The Gate Stays Human: Who’s Accountable When AI-Ops Acts. July 2026.

• My Take: How much can we automate? You can automate the action, but you can never automate the accountability. Lots of talk about the “one throat to choke” (my words) when … not if … things go wrong.

• Play out the scenario. What happens the day it goes wrong. You decide on the “blast radius” … how big is the problem if it goes wrong.

• Who owns what … need to name a person who owns the high blast radius decisions.

HIGHLIGHTS (12-Minute Run Time)

• Minute 01:20 – blast radius is the key

• Minute 06:00 – accountability stays human

• Minute 07:00 – hard stop; really evaluate what is automated

• Minute 08:00 – Q&A

SOURCE: Abraham Thomas published Looks, Brains, and Money. August 2026.

“careless in —> convincing out”

BONUS: Funny one … source: Steve Yegge

The future of research is a barbell.

That argument came out of the closing chapter of my book, Advising Midas, a history of Wall Street sell-side research from the 1950’s to the AI era.

• Advising Midas (Amazon)

• Advising Midas (Gumroad)

These are a few excerpts from Chapter 6.

• “The argument here is that the value does not vanish; it polarizes. It moves to the two ends of a barbell, proprietary inputs at one end and judgment and trust at the other, while the synthesis in the middle collapses toward free.”

• “When processing becomes cheap and universal, processing stops being the edge. If everyone can run the same model over the same feed, the feed produces no alpha; the signal decays as fast as it is found. What remains scarce migrates to three things: proprietary access to data nobody else has, the organization that makes a messy feed usable, and the question, which turns out to matter most. In a world where the answer to any well-posed question is a prompt away, the advantage belongs to whoever asks the question no one else has thought to ask. The analyst’s value moves upstream, from producing the analysis to specifying it. The question is the product.”

• “Consider Ivy Zelman. A buy-side client does not pay Zelman because she has a number he cannot find. The number is public; the model has it too. He pays her for 25 years of pattern recognition across housing cycles, for the judgment to know which number matters this time, and for the authority to be believed when she says so. That last part does not reduce to data. Wisdom is not information. A model can accumulate information at a scale no human can match; it cannot accumulate wisdom, because wisdom is the residue of having been right and wrong…”

Subscribe now

Special thanks to our sponsor Sidekick Lab!

Sidekick is the Agentic AI layer that gives your enterprise a living, governed understanding of your data estate.

We provide the ontology for your data to finally speak the same language as your business.

• News

• Pods

• Charts

• Final Thoughts (Table Stakes Data – The Searchable History of ADW)

#1 – Grigorios Reppas published The Data of Human Behavior Has Grown Up. August 2026.

• My Take: Alternative types of data are being adopted more widely. It is now table stakes for any serious investor. It is becoming table stakes for anyone involved in a wider range of pursuits, from commercial real estate to the public sector.

• Beyond using a single data source to drive insight, as tools improve, combining multiple data sets to drive higher conviction insights is becoming more possible. And from a value decay perspective, how you combine those multiple data sources unique to you and difficult to replicate. When making big decisions, no rock should be left unturned, especially when access to “the rocks” get much easier.

#2 – Dan Goldberg published California’s DROP (Delete Request and Opt-Out Platform) went live. August 2026.

• My Take: Thanks to Dan for highlighting this … I had missed it somehow. One request from a California resident now reaches 600+ registered brokers, and all of them are legally required to act on it within 45 days. 350,000 people signed up quickly.

• So what is the impact? There are still some open questions, but the data a company generates itself likely gets more valuable, while data assembled from other people’s exhaust gets riskier. And of course, there is the operational headache of just reacting to all the requests.

#3 – Matteo Guillot published Every AI Strategy Is a Bet on Architecture. July 2026.

• My Take: How your information is organized is key to how well AI will work for you.

• I am living a small version of this in my day-to-day (see Final Thoughts from ADW #299), the value of my AI tools expands as they “learn” more of my context. They (it?) still get confused and make dumb suggestions, but they are getting better and adding a ton of value. This is what I understand the “harness” to be … all the information & wiring around the model that makes the model useful to me.

• The big model developers want to control as much of this as possible, but the users want to keep this information for themselves. That is the struggle. Matteo suggests three reasons users’ harnesses survive.

• Something I think about … if I build a great harness and spend years “teaching” the models what I do and how I do it … what happens if (when) Anthropic 3x’s the price? Can I take my context with me to another model?

BONUS: Dun & Bradstreet published Enterprise AI Returns Continue to Advance, But Only 6% Have the Data Ready to Scale Them. July 2026. “Only 6% of organizations report their enterprise data is fully ready to support AI at scale.”

What else I am reading:

Gary Drenik published AI Is The Canal. Forward-Looking Data Is The New Water. July 2026.

Dan Evans published Everyone’s Impressed by the Model. Almost Nobody Can Reach Their Own Data. August 2026.

Federico Trotta published Why Your RAG Pipelines Are Only as Good as Your Data Acquisition Layer. August 2026.

Databricks published AI Applications in Finance: A Practical Use Case Guide. July 2026.

FactSet launched A multi-layered AI stack powering faster and smarter decisions for financial professionals.​ August 2026.

Source: Prasanna Vaidya of GydeBites published an interview with Evan Reich. Episode: Building Proprietary Intelligence Systems in an AI-Driven World. July 2026.

• My Take: Where is your competitive advantage? Domain understanding and workflows. Proprietary intelligence is the combination of data, context, reasoning. As the gap between models shrinks, the important part becomes the data inputs and how the models understand your workflow.

• Not everyone has a strong understanding of data and workflow with which they engage. You need granular domain knowledge to keep iteratively testing the system. AI doesn’t have wisdom, understanding, judgement, so you need to be very intentional about teaching it. It will take everything literally and is very detail oriented.

• 5 Things:

1. Data (the input ... deterministic checks)

2. Structure (schemas, etc.)

3. Graphs & ontology (tie everything together)

4. Reasoning layer (domain knowledge applied)

5. Harnesses (keep reasoning bounded)

HIGHLIGHTS (21-Minute Run Time)

• Minute 01:30 – what is proprietary intelligence

• Minute 04:00 – AI ready data ... make everything very explicit

• Minute 07:45 – people fear the unknown and would rather wait

• Minute 10:30 – 5 steps

• Minute 15:00 – how to make sure systems are accurate & working properly

• Minute 17:00 – practical advice from Evan

• Minute 19:00 – rapid fire questions

Source: TableStakesData (by John Farrall … see more detail below in Final Thoughts)

I pulled every issue of Alternative Data Weekly back to October 2020, parsed out every source I have ever pointed readers at, and asked the pile what changed. It came to 2,500+ sources.

BONUS: Andrew Chin from AllianceBernstein published AI and Alpha: Why Technology Alone Won’t Be Enough. July 2026.

I spent a weekend doing something I have wanted to do for years. I pulled every issue of this newsletter back to October 2020, parsed out every article, podcast, paper and post I have ever pointed you at and put the whole thing into one searchable table.

It comes to 2,500+ sources, and it is live now at TableStakesData.com. You can search it, sort it, and filter it by topic, by year or by publication. Where I wrote a comment at the time, the comment is attached.

Then I did what you do once you finally have the data. I asked it “what changed” & “what was most interesting”.

#1. The conversation turned in July 2023.

• That is the month AI passed “alternative data” as the thing I was citing most, and it never gave the lead back. ChatGPT launched November 2022. So it took this industry a few months to catch up.

• Counting only sources with AI in the title, which is the strictest measure I can apply to my own reading, AI went from 0.25 citations per issue in 2021 and 2022 to 4.64 in 2025 and 2026. That is 18.7x. The looser measure gives 11x. I am quoting the stricter method because it produces the bigger number.

• Over the same stretch, the word “alternative” in a title fell by almost half.

#2. The selling never changed. The buyer did.

• Here is the number that surprised me most. Coverage of data monetization, meaning how firms package and sell their data, is as common today as it was five years ago. 78 citations then, 79 now. It has not moved.

• Everything around it did. Marketplaces and data discovery, which in 2021 looked like the future of this industry, are down 46%. Crypto went from 22 citations to one (I still love crypto). So the money did not go anywhere. The buyer changed. It used to be a fund hunting alpha. Now it is a company trying to feed a model.

• Data is still sold, not bought. I hope this changes with the advent of agent-buyers.

#3. AI may have already peaked, at least in here.

• Sources with AI in the title ran at 5.10 per issue in 2025. So far in 2026 they are running at 3.87, a 24% decline and the first time in four years the line has pointed down.

• One year is not a trend and I would not bet the house on it. But if the volume of AI talk is topping out while the money keeps moving, that gap is where the next few years get interesting.

One last thing. Almost nobody at the top of my citation list in 2021 is at the top of it now. The people writing about this industry turned over about as completely as the subjects did. Two names appear in every single year since 2021: Ben Lorica and Auren Hoffman. If you have been reading them the whole way too, you chose well.

The full version, with the charts and the methodology, is at www.tablestakesdata.com/state-of-alt-data-2026.

Thanks for reading!

Share

Subscribe now

Special thanks to our sponsor Sidekick Lab!

Sidekick is the Agentic AI layer that gives your enterprise a living, governed understanding of your data estate.

We provide the ontology for your data to finally speak the same language as your business.

• News

• Pods

• Charts

• Final Thoughts (MCP = Margin Compression Protocol)

#1 – Maribeth Martorana & Andy Graham published Enterprise Data Is Becoming the Next Competitive Advantage. July 2026.

• My Take: Once your internal data is organized (i.e., governed, structured, compliant, etc.), then the data’s value compounds when using AI tools.

• The right question from the executive level is, “how do we increase the value of our enterprise data assets?”. I’ve said for years the value of just getting your own data house in order is worth the effort, whether you formally commercialize the data externally or not.

#2 – Srinivasa Mathkur published The Context Gap: Why Data Products Anchor AI Success. July 2026.

• My Take: The models are very good, but they need very specific context to be great. This is where humans add a ton of value. “Knowing a lot in general” and “knowing enough about this specific situation, right now” are very different things. Models are very good at the former and need a lot of help to be very good at the latter.

• When the “customer” of the output is an agent, not a human, this amplifies errors early in the process.

• Srinivasa takes us through the journey of creating an environment where you get the best output from the tools. “What does a model have when it answers you”? From semantics to governance to data quality, and provenance/lineage … these are all essential to generating trustworthy output.

#3 – Adrian Krebs published How Hedge Funds Source Web Data in 2026. July 2026.

• My Take: Web scraping is like a free puppy, not free beer. The maintenance is what gets you.

• When funds are thinking about “build vs buy”, they should think about how the collected data will be used, how the pipes will be maintained, and if the “what is collected” or the “how it’s collected” is proprietary to their workflow or a commodity input.

BONUS: Matt Ober’s The Cost of Information is Expensive. July 2026. “The game of data is costly when it’s exclusive data.”

What else I am reading:

Simon Burton published Discussion with The Terminalist. July 2026.

Matthew Bernath published The State of Data Monetisation in African Financial Services. July 2026.

Benn Stancil published Is “jobs” a four-letter word? July 2026.

Joe Reis published To Every Agent Its Own Database. July 2026.

Sarah McKenna commented on Public Search Results and Responsible Data Access. July 2026.

Bilal Jafar published Point72 hires AI expert to predict weather for traders. July 2026.

James Evelegh published How to unlock the full value of your data. June 2026.

SOURCE: Ethan Kho’s Odds on Open podcast published Ex-Balyasny PM: Quant is like Blackjack, Fundamental is like Poker. July 2026.

• My Take: Ying Hua, founder of Implied, breaks down how top multi-manager hedge funds synthesize quantitative discipline with discretionary analysis to extract repeatable market alpha.

• I really enjoyed this interview. The investment world is changing quickly (isn’t it always) … Ying shares some valuable perspective.

• Quant-amental ... what does that even mean? Using quant and quant data to inform fundamental analysis. Historical pattern matching = quant. Things that are not pattern matching increase in value.

• AI tools shift focus to “art” … things that are well studied can be outsourced. What does analyst do?

  • Gather data

  • Process data

  • Analyze data (hard for LLMs to do)

• Q: How quickly can you update models? A: Not fast enough. (Full disclosure: this is what we do at SymetryML.)

  “We changed our process so that every model can be updated within two minutes.” (01:30)

• Forecasting a loss in insurance is “art” (interesting example of alt data usage for insurance use case). Another example shared of using credit card data (19:00).

• Another person who thinks these tools will expand the demand for investing talent.

• A differentiated view today pays more.

HIGHLIGHTS (70-Minute Run Time)

• Minute 01:00 – interview starts.

• Minute 05:30 – the importance of sizing.

• Minute 08:00 – insurance example (“quantamental”).

• Minute 12:00 – can fundamental investing be automated?

• Minute 17:00 – does demand for investing talent expand or contract as a result of AI tools?

• Minute 21:00 – why can’t I just tell Claude to do all this for me?

• Minute 30:00 – how do you make sure the models don’t commodify your business?

• Minute 35:00 – MCP connecting data … chaos ensues

• Minute 38:00 – AI picks up regime changes?

• Minute 39:45 – First thing you’d advise a fundamental PM to do with AI?

• Minute 42:00 – what moats expand (quality of talent)? Ask good questions. Reading people.

• Minute 44:45 – willingness to learn the technical side.

• Minute 49:00 – fundamental is poker; quant is blackjack.

• Minute 53:00 – choosing the right sector is key.

• Minute 58:00 – philosophy for building wealth.

• Minute 65:00 – traits of the good and the bad investors.

• Minute 69:00 – importance of thinking from first principles.

SOURCE: Andre Retterath published The Analyst Role Is Dead, No? July 2026.

• My Take: Data quality is key. The job of the analyst is changing. Really cool review of use cases across the board.

BONUS: LSEG published AI’s Role in Wealth Management. July 2026.

• My Take: Good data is key. Wonderful series of use cases for a very relationship-driven industry. Many would think this to be counterintuitive.

MCP = Margin Compression Protocol

The question of whether data providers will plug into AI agents is answered.

Asymmetrix surveyed the industry: 56% have launched an MCP server, 40% are considering one. 96% in. In the same month, S&P launched Adaptive Retrieval, LSEG shipped its MCP connector, and Bloomberg explained why it built MCP and then pointed it at itself. Wildly interesting timing, as in the same week The Terminalist explained what all this means for the industry.

Everyone has pipes. How does it get priced? The same Asymmetrix survey shows answers ranging from a free add-on to a 50% subscription bump. That wide of a range indicates no one knows.

My take: Once an agent can easily query five vendors through the same connection, “our data is special” becomes a testable claim, checked in seconds, thousands of times a day. Bloomberg sees this, which is why their MCP faces inward. Keep the agent, keep the customer. S&P and LSEG are betting the opposite: open the pipes, win on trust, citations, and token efficiency.

Both can’t be right over the long run.

My thought is that some version of Jevons paradox will happen and the amount of data being “used” in various ways will explode. Data-hungry non-human agents can consume data at a pace no human can comprehend. Price per query might collapse, but volume goes vertical. The fight is over who captures it.

For data owners, reachable by agents is now table stakes.

The edge is being the source the agent goes to when the answer has to be right.

Special thanks to our sponsor Sidekick Lab!

Sidekick is the Agentic AI layer that gives your enterprise a living, governed understanding of your data estate.

We provide the ontology for your data to finally speak the same language as your business.

• News

• Pods

• Charts

• Final Thoughts (what will be valuable in the future?)

#1 – The Terminalist published Bloomberg’s Last Stand. July 2026.

• My Take: I am assuming most ADW readers have already seen this piece as it was published just over a week ago. It is a must read when The Terminalist publishes anything. I set aside some time last weekend to read through this. There is a lot of insight here. I’ll focus on the idea of data FOMO.

• I’ve attempted to articulate the idea that alpha used to come from access to data (buying the $30k BB license, knowing where to look) … now that, in theory, everyone has access to all the same data … the alpha is going to come from organizing the data, structuring the workflow, and asking the best questions. FOMO shifts from missing the right data to missing the right question.

• From the sell side’s perspective, they will create value by creating unique proprietary data sets … (surveys, expert conversations, channel checks).

• From the buy side’s perspective, they will create value by organizing workflows & data assets in such a way that their questions get the highest quality answers. Rather than creating your workflows around the data assets & tools like Bloomberg, the best buy-side firms will create the best workflows and fit the data assets into those workflows.

• Now you just need to ask the right questions (that’s the hard part, right?).

#2 – Chad Sanderson published The Shift Left Manifesto - v2. July 2026.

• My Take: This is an update to The Shift Left Manifesto - v1, published March ‘25.

• “Shift left” means fixing and organizing your data earlier in the process, ideally, right where it’s created, instead of waiting to clean up the mess later.

• The mantra of ship new stuff fast meant decentralization … which means no single team has the full picture. Which causes a lot of new problems, but new stuff is getting shipped quickly, so this is good, right? The answer in part is “shift left” so that the decentralize teams are working off the same ground truth.

• As more people are engaging with AI tools that give everyone the ability to get answers, terms like evaluation, traceability, provenance suddenly become hot topics.

• At the end of the day, some human needs to stand behind the answer with confidence. Moving closer to the source (ideally all the way to the code) will help.

BONUS: Vin Vashishta published The Context Moat Is Real, But Most Enterprises Don’t Have One Yet. July 2026. “…the moat is moving from the model to the proprietary knowledge that makes models and agents valuable.”

What else I am reading:

Extraordinarily sad news: Elizabeth Pritchard Obituary :Transformational Chief Data Officer. July 2026.

PwC & Eagle Alpha published Data monetization for CFOs: How to find, assess and execute opportunities. July 2026.

Simply Wall St. published Is TransUnion’s (TRU) Alternative Data Push Quietly Redefining Its Core Competitive Edge?. July 2026.

Rishabh Raj published Alternative Data May Soon Become Mainstream Investing Tool, Says S&P Global’s Aditya Sharma. July 2026.

Bruce Kappler published The Supreme Court Expands Privacy Rights to More Squarely Encompass Your Digital Footprint. July 2026.

Matt Robinson published AI Turns Plain English Into Backtests: Lord Abbett’s Tal Fishman. July 2026.

Ergest Xheblati published The Demand Perspective in Data. July 2026.

Book Alert: Roopika Risam published Data Empire: The Power of Information to Organize, Control, and Dominate. July 2026. Excerpt: 2126: The year data will be controlled by citizens, not tech companies.

SOURCE: Daniel Beach published an interview with Hugo Lu, Agentic Data Engineering Is Here — But Can It Close the Loop?. July 2026.

• My Take: This was a bit technical for me, but interesting to hear some of the pain points these two describe and what they are doing about it (Lu’s Founding story of Orchestra).

• The opportunity: how people build data pipelines is changing. You can make a big impact if you are ahead of the curve.

• I like how Hugo talked about dealing with the massive change happening in the world. Bottom line: Stay focused. Simplify. Every enterprise has multiple data tools. These need to work together. This won’t change.

• Last one … discussion of what happens when you are fully locked into a single model provider and they move to increase price? This is when, not if.

HIGHLIGHTS (33-Minute Run Time)

• Minute 01:00 – Hugo Lu of Orchestra; background in data engineering.

• Minute 07:00 – importance of testing; little things creep in.

• Minute 09:30 – training junior developers.

• Minute 12:00 – the background on Orchestra.

• Minute 15:30 – Agentic Data Engineering.

• Minute 20:00 – shift toward speed and simplicity.

• Minute 25:00 – what’s next? Staying up with all the changes.

• Minute 27:00 – focus on the good user experience for the basics … observability, monitoring, alerting, etc.

• Minute 29:30 – dealing with token costs.

SOURCE: ICONIQ published 2026 State of AI Report: The Builder’s Economy. July 2026.

What will be valuable in the future?

Elon says everything will be free.

If intelligence is free and everyone has access to all the information, do we all end up like the people in Pluribus, or is there something others will desire? What will people find valuable? When everything is abundant, how do we stand out?

People will continue to chase status.

If you can deliver something that will help them increase their perceived status, they will pay you. The most obvious is that if you can help them make more money (more money = more status) … then you will get paid.

Kevin Kelly put together an interesting piece on the idea of value when as things get cheaper (interesting: this was originally published in 2008 … he updated for 2026 and not much changed).

Better Than Free

His words (my words)

1. Immediacy — (now where was that thing?)

2. Personalization — (I value the effort someone put in to make this just for me)

3. Interpretation — (what does all this mean? Why should I trust you?)

4. Authenticity — (community)

5. Accessibility — (I don’t want 50 user names & passwords)

6. Embodiment — (live events)

7. Patronage — (support creators — ADW)

8. Findability — (the right info at the right time)

“…these new eight generatives demand an understanding of how abundance breeds a sharing mindset, how generosity is a business model, how vital it has become to cultivate and nurture qualities that can’t be replicated with a click of the mouse. These generatives are very human skills, by the way, which AIs are not very good at – yet.”

Subscribe now

Special thanks to our sponsor SymetryML!

Increase Research Throughput Without Increasing Infrastructure or Compute Costs.

SymetryML helps funds interrogate massive real-time data to uncover usable signals faster, without the compute costs, pipelines, and data-engineering burden of conventional analytics.

We still have 3 open slots for a free pilot.

Get details and claim a slot: hello@symetryml.net.

• News

• Pods

• Charts

• Final Thoughts (second brain)

#1 – Kunal Mehta published We Don't Have a Data Problem. We Have a Signal Problem. July 2026.

• My Take: The bottleneck is no longer data, or even access to data. It is attention and context. Can you sift petabytes of noise for the one piece that matters? Can you do it in real time? Can you do it at scale?

• Kunal describes the goal as a “living picture of what is actually happening.” Building that “mosaic”, reading the situation, and handing the decision-maker only the piece they need is a genuinely hard problem. My former company ModuleQ spent years working on exactly this. I learned firsthand how hard it is to do, and how powerful it is when done well.

• Two lessons from that experience still hold. The more context you give the system, the better it performs. The more you engage with it, the better it gets. The winners in this next phase will not be the ones holding the most data. They will be the ones feeding their systems the most context, with an engine on top that delivers only the right information, at just the right time.

#2 – Meltem Demirors published The World as Model. June 2026.

• My Take: I was watching the MLB draft earlier this week when I heard a scout say that MLB prospects need to get to a place (NCAA school, HS team, baseball camp, etc.) that will start to collect their data. The right baseball data collection equipment will capture spin rates, swing speed, exit velo, etc. In this new world of baseball, these numbers will get you found by the scouts (or ignored).

• Meltem argues that capital follows information density. Just like how an MLB scout might find you once you have a dense collection of real data about your baseball talents, there is value to be found in data that is collected. Yet, there are large parts of the world that have not yet been digitized. Industrials companies offered as a case in point. We should build for industrial companies what HFTs built for markets. Native resolution data collections that will demonstrate real value in short order.

#3 – Vinit Agrawal & Jason Nogueira published How alternative data drive investment insight and alpha potential. July 2026.

• My Take: Investment analysts can dig deeper into the companies and industries they cover than ever before. AI tools are making engagement with all sorts of data more impactful, traditional and ‘alternative’ sources alike.

• Implemented correctly (and this is not easy), teams can separate signal from noise to strengthen confidence in their investment theses.

What else I am reading:

Jody Hesch published Fundamentals of Engineering: Part 1 (Part 2: Here). July 2026.

Cindy Lin published MCPs Are Becoming Finance’s New Plumbing. July 2026.

Davis & Cole published How the Growing Use of Alternative Data Sources Is Reshaping How Professional Traders Identify Market Inefficiencies. July 2026.

Dan Entrup published Strategic Strategery. July 2026. Related announcement: AggKnowledge Launches Healthcare Data Product.

The Economist published Kevin Warsh’s new Fed. July 2026.

Source: Stan Altshuller interviewed Pouya Taaghol: How Alternative Data and AI Are Predicting Market Moves Before Everyone Else. July 2026.

• My Take: Alternative data is not always found in private, hard to find places …sometimes it is just public data, organized really well. Different people will interpret the same data differently. Which keeps things interesting.

• Pouya Taaghol founded Big Data Federation. He started by counting cars in parking lots. He highlights the benefit of hiring science guys vs finance guys to unlock value. He suggests that the US stock market digests information and predicts stuff. There is value to be found here.

• Alpha decay:

  • 1- Panel data (panel can change) … can you make the panel stable?

  • 2- Surveys (only includes the people who respond).

  • 3- Factual data … (TSA data, Semiconductor Association, etc.).

HIGHLIGHTS (33-Minute Run Time)

• Minute 01:00 – interview starts; Pouya’s background

• Minute 06:00 – Science guys vs finance guys

• Minute 09:00 – getting outside the noise and the benefit of independent thinking

• Minute 12:00 – details on the BDF product

• Minute 16:00 – same data but different conclusions

• Minute 17:00 – UltraBlue Capital (fully automated long-biased fund that trades using BDF data)

• Minute 22:00 – alpha decay

• Minute 28:00 – Pouya’s journey

SOURCE: SSRS published Voices from the Data Community: How 2025 Has Impacted Public Data Users. May 2026.

• My Take: The recent interruptions of government data flows are an opportunity to improve how we work with data. This should cause users of gov’t data to want to broaden the data sources we use to determine how we are doing as a country (i.e. include more alternative sources of data).

Related article: The Economist published Kevin Warsh’s new Fed. July 2026.

BONUS: Pouya Taaghol published The Evolution of Retail Brokerage. July 2026.

• Phase 1: Electronic Trading – Online trading replaced traditional brokers, making investing faster and more accessible, though commissions remained high ($30–50 per trade).

• Phase 2: Commission-Free Trading – New revenue models enabled free trading, simplified the user experience, and fueled a massive increase in retail investing.

• Phase 3: AI-Powered Investing – As commission-free trading and modern apps become standard, the next competitive advantage is AI: delivering personalized insights, automation, and intelligent investment decisions.

Second Brain: I can only hope it is better than my first brain.

I have spent the past few months building a “second brain” ... an AI system that remembers what I read, who I talk to, and what I decide. The lesson: the more the system knows about me, the better it works. Context is the fuel. I can only imagine what happens once glasses record my everyday movements and the system starts learning how I actually operate (and starts helping me get better).

Which got me thinking more broadly about this idea.

MLB scouts will tell a young prospect to get to a place where they can start to build a corpus of data. Produce the right data (spin rate, exit velo, etc.) and they will find you.

Jim Simons was early in this game … from Zuckerman’s The Man Who Solved the Market (source):

Now the same sequence is arriving for the physical world. Meltem Demirors argues in The World as Model (featured above) that machines, pipelines, power systems, and buildings generate continuous signals across hundreds of modalities, and most of them go uncaptured. The ones that are captured rarely arrive at a resolution where data becomes predictive instead of descriptive. Her example: a temperature sensor polled every fifteen minutes tells you a machine ran hot. A millisecond vibration signature tells you a bearing is failing three weeks before it does.

So here is the question for anyone running a company, and it is not “what data do we need?” Simons could not have answered that in the 1970’s, and you cannot answer it today. The question is: what signals is your business letting die unrecorded? The interest rate history nobody copies down, the vibration signature nobody captures, the customer interaction nobody logs ... none of it can be bought back later.

Missed data is the only truly unrecoverable asset.

Everything feeds the machine. The only thing it can never learn from is what you didn’t write down.

Subscribe now

Special thanks to our sponsor SymetryML!

Increase Research Throughput Without Increasing Infrastructure or Compute Costs.

SymetryML helps funds interrogate massive real-time data to uncover usable signals faster, without the compute costs, pipelines, and data-engineering burden of conventional analytics.

We still have 3 open slots for a free pilot.

Get details and claim a slot: hello@symetryml.net.

• News

• Pods

• Charts

• Final Thoughts (Distribution)

#1 – Gurbinder Gill published An LLM Without Memory Is Just a Very Expensive Prompt. July 2026.

• My Take: The winners in the AI age will be those firms that understand their own businesses well enough that they can build the workflows that compound over time (and workflows that deliver deterministic outputs when that is appropriate).

• But … this requires persistent storage, and a consistent memory layer that is always on (suddenly losing context = arghh!). As this baseline gets bigger, and when talking about the very detailed workflows of complex orgs, this stuff quickly gets complex, expensive, and slows down the response times.

• Getting all the key-person context & knowledge (data, data, data!) into the system is the win.

#2 – A-Team Insights published How Much Hedge Fund Alpha Is Lost Before the Model Even Runs? June 2026.

• My Take: This is a problem as old as data itself. Dealing with bad data, finding it, cleaning it, organizing, it … what a pain.

• I thought this discussion around point-in-time to be particularly interesting. Everyone interprets that term differently. I found there was always a ton of confusion.

• Of course, AI takes this all to a massive scale. Good processes early in the stack help tremendously.

BONUS: Meltem Demirors published The World as Model. June 2026. “The best representation is not necessarily the richest one. It’s the simplest representation that preserves the information required to accurately reconstruct or predict the system’s behavior.”

“Every representation is ultimately a decision about compression.” (my company SymetryML - & this week’s sponsor - solves this … no need to sample or compress … no trade-off decision required).

What else I am reading:

Dan Evans published Everyone’s Building Agents. Almost Nobody’s Built the Foundation. July 2026.

Olga Berezovsky published The Model Is Smart. Your Company Is the Problem - Issue 322. July 2026.

Abraham Thomas published On Data Quality. June 2026.

Jason Derise published Summer Inspiration 2026: Books, Podcasts, and Newsletters Shaping Data and Investing. July 2026. (thanks for the ADW mention!)

Nehhaa Purohit published Building Mature, Decision-Driven Data Organizations. June 2026.

Source: Tristan Handy talks with Claire Gouze, co-founder and CEO of nao Labs. The context engineering playbook. July 2026.

• My Take: Interesting interview with a founder of nao Labs. It is pronounced “nao” … like “now” … not pronounced “n. a. o.” ...

• Recent pivot from “Cursor for data strategy” ... to “context layer” (13:10 …talked about the pivot). Basically wanted to focus on things people are excited about. The idea of a context engineer has legs (18:20). They published a context engineering playbook.

• Evaluation frameworks … intriguing to Tristan:

  • Methods ... not what to put into your context.

  • Be focused on first use cases.

  • Start small, set up scaffolding, eval framework running…

• Machine vs human context (28:30).

  • Where does this go? Everything ends up in .md files & needs to be as close as possible to daily work. Then, whose job is it to edit .md files?

  • This is important because adding relevant context increases quality.

In my experience, the more context, the more narrow the use case. You wind up creating a bunch of agents … because each project (even if very similar) has slightly different context. This is where deep domain experience makes a massive difference. I think nao’s pitch is saying are the source of context truth for agents.

HIGHLIGHTS (50-Minute Run Time)

• Minute 01:20 – Claire’s background

• Minute 04:00 – custom build vs using the available tools

• Minute 06:00 – nao background

• Minute 07:30 – how are coding agents going to impact consulting world?

• Minute 20:00 – context engineering playbook

• Minute 24:00 – Ramp example

• Minute 25:45 – who does this work?

• Minute 28:00 – machine vs human context

• Minute 40:00 – plugging into production data + plugging into production context

• Minute 45:00 – files system is all you need (or is it?)

• Minute 48:00 – thoughts on open source

BONUS: Marc Andreessen on AI, California, and the Future of America | Joe Rogan. May 2026.

• My Take: I listened to this entire 3-hour podcast while driving last weekend. I really enjoyed it. Specifically this part:

“Andreessen argues that AI should be understood less as replacement technology and more as a universal layer of cognitive augmentation, giving individuals access to capabilities that previously required teams of experts.”

• This validated the experience I am having with AI. I am experiencing the same cool things and shortcomings Andreesen described. Assuming he is on the forefront of all things AI, this was very affirming for me.

SOURCE: Modern Data 101 published The Modern Data Report 2026: The Data Activation Gap. June 2026.

Research from 540+ data practitioners on the gap between data platform capabilities, operational reality, and business outcomes.

Distribution. This is what I am good at & what I enjoy doing.

Today’s Top Founders All Have One Thing in Common

How to break through the noise? Even when proven founders are using a proven technology to solve real problems, it is tough to get the attention of the decision makers.

Getting the word out. Causing someone else to take action.

When Elon bought Twitter, I got the sense that he really didn’t want to, but was in too deep that he would have gotten sued if he pulled out late in the deal. This has proven to be a massive positive for him. He has 240M+ people who he can reach.

240M+ people!

Talk about getting the word out.

Then … telling a compelling story is powerful (and harder than one might think)!

Funny:

Special thanks to our sponsor SymetryML.

Bring real-time ML to your clickstream data, through Claude or existing workflows.

100TB+ of clickstream? No Problem. SymetryML lets analysts interrogate their massive clickstream data in seconds, without repeated warehouse scans, heavy MLOps, or a full data science team.

Ask questions in plain English. Run advanced analytics and ML.

We’re opening 5 free pilot slots for teams working with large clickstream datasets.

Get details and claim a slot: hello@symetryml.net

• News

• Pods

• Charts

• Final Thoughts (250)

#1 – Abraham Thomas published On Data Quality. June 2026.

• My Take: Another must-read from Abraham (teased as a “Part 1”). He shares with us the different levels of data quality: granular, aggregate, fitness, value. These are laddered and reliant on the prior step.

• The key question is about how the data will be used. That drives how we should think about “quality”.

#2 – Jon Stojan published The Real Advantage Isn’t More Data. It’s Better Context. June 2026.

• My Take: Data access used to be the source of advantage (the internet helped solve this). Then it was the ability to process that data (AI levels the playing field here). Now it is organizing it the right way, giving it the context of what you are doing, and asking the best questions of the data.

BONUS: Rob Mannix & Mauro Cesa published How quants are getting the most out of Claude. June 2026. “…quants are discovering the tools require skillful handling”

What else I am reading:

Peter Baumann published Data Contracts - Putting Data Governance into Practice. June 2026.

Modern Data 101 published The State of Data Products (Q1). March 2026.

Parabellum Investments Acquires Crux Informatics to Transform AI-Powered Data Infrastructure for Financial Markets. June 2026.

Dan Entrup published Raising Cannes. June 2026.

Source: Ben Lorica of The Data Exchange Podcast interviews Chang She, co-founder and CEO of LanceDB. The Data Stack Wasn’t Built for AI — Here’s What Comes Next. June 2026.

• My Take: Ben is one of my favorite follows in the data world.

• This podcast resonated with me because Chang talks about how multimodal exhaust data that enterprises spent a decade shoving into a back room is now proving to be valuable (finally, right?) … mostly because the tooling to finally draw signal out of it exists.

• Bottom line: data is more valuable because of AI. Agents are driving this to massive scale (100x++).

HIGHLIGHTS (40-Minute Run Time)

• Minute 00:45 – Chang shares background and thoughts on why LanceDB is needed.

• Minute 02:00 – Vector Databases discussion (what works and what does not work)

• Minute 05:30 – LanceDB – multimodal data (“open”)

• Minute 10:00 – the “bay area bubble”

• Minute 14:00 – various types of multimodal (data types, workloads, ops)

• Minute 17:00 – the massive scale is causing problems

• Minute 20:00 – agents & memory

• Minute 22:00 – “lightweight” … make it easy for agents to engage with massive amounts of data.

• Minute 27:00 – building more capable agents

• Minute 28:00 – CXO thoughts (pain points around new AI data silos)

• Minute 30:00 – data stack in a few years (data tooling ”middle layer” will start to melt away with agents taking over)

• Minute 34:00 – the rise of open weight models

BONUS: Neudata’ Stuart Broughton hosted Explore data for sale: Winners, failures & the decisions that made them . June 2026.

SOURCE: Jim Rowan, Beena Ammanath, Nitin Mittal, Costi Perricos of Deloitte published The State of AI in the Enterprise. January 2026.

• My Take: This first chart is interesting … hopes were too high for everything but productivity.

BONUS: Asymmetrix published the results of their MCP adoption survey.

There is a full report available.

Happy Fourth of July for my American readers.

Some “alternative data” facts about the USA (AI helped me here).

1. America was a data project from day one. The Constitution (Article I, Section 2) requires a head count every ten years. The first census in 1790 found 3.9 million people. The next one, in 2030, will count around 345 million, roughly 88x larger. We may be the first country to write recurring data collection into its founding charter. The original alternative dataset was the census.

2. The greatest coincidence in the data. John Adams and Thomas Jefferson both died on July 4, 1826, hours apart, on the exact 50th anniversary of the Declaration. James Monroe followed on July 4, 1831. Three of the first five presidents died on Independence Day. Run those odds.

3. We’re celebrating the wrong date. Congress actually voted for independence on July 2. Adams predicted July 2 would be “the great anniversary festival.” The text was adopted on the 4th, and most delegates didn’t sign until August 2. We commemorate the metadata, not the event.

4. The oldest “new” country. America feels young at 250, but the Constitution (in force since 1789) is the oldest written national constitution still operating. Most of the world runs on a newer one than we do.

5. The 22-million-row dataset that vanished. During the 1976 Bicentennial, a wagon-train time capsule holding roughly 22 million Americans’ signatures was stolen from beside the speaker’s platform while President Ford spoke at Valley Forge. Never recovered. The original lost dataset.

Special thanks to our sponsor Brickroad!

Alpha isn’t found in a catalog. Brickroad’s information frontier agent finds alpha suppliers at the frontier - companies generating your target data who’ve never marketed it. Sign up for a 7-day Pro trial Here.

• News

• Pods

• Charts

• Final Thoughts (this, not that)

#1 – Didier Lopes published Data is THE moat. June 2026.

• My Take: I like the expansion beyond, 1- data you’ve licensed, and 2- your internal data … to 3- “derivative data”. This is how you combine and work with (correct, decide, combine, query) the data, that is what makes your data more valuable. Said another way, asking the best, or most unique interrogations of proprietary data is going to be a huge moat in coming years (shameless plug: SymetryML makes this easier).

• See ADW #177 for early thinking on this idea … the idea that a combination of data assets that is a mosaic unique to you (and nearly impossible to replicate). AI will take the ability to uniquely combine and query disparate data assets to the next level.

#2 –Peter Baumann published Data Contracts - Putting Data Governance into Practice. June 2026.

• My Take: I am seeing governance issues around who gets access to what information even at a small company like mine. I cannot imagine the complexity at a large organization. But I can imagine the power once these guardrails are in place, this makes it worth the effort.

#3 – BigEye published What is shadow AI? June 2026.

• My Take: I’ve been seeing this issue discussed more frequently. People use their personal AI at work (I am guilty of this). This can create problems. Traceability, privacy, exposing confidential information … this stuff is not good.

• AI is moving fast and it is nearly impossible to keep up. You need to balance 1- getting the most out of these powerful technologies and, 2-keeping a required governances perimeter around the use of these tools.

• Network level blocking is tough. Good luck telling employees they can’t use AI tools. BigEye suggests the management of this should include authorized tools … tell employees to use AI, but this specific AI tool … and implementing the governance/access from the data layer, rather than the application layer. These are tough problems. These are important problems.

BONUS: Dylan Anderson published Fable and Mythos Are Here (well kinda). So What’s Next for Human Work? June 2026. “The moat isn’t the technical part anymore; it’s all the stuff that came around the technical part: the logic, the structure, the process, the implications, the action, the communication, etc.”

What else I am reading:

Auren Hoffman published disposable software: software is now just paper plates. June 2026.

Nathan Goldschlag published A New Threat to Economic Data. June 2026.

Arvind Narayanan and Sayash Kapoor published Why AI hasn’t replaced software engineers, and won’t. June 2026.

Ian King published The City of London’s vanishing analysts. June 2026.

Anthony Goldbloom published Account Scoring Should Explain, Not Just Rank. June 2026.

Dan Entrup published Vibemaxxing. June 2026.

Matt Robinson published AI Pushes Trading Into Thinner Markets. June 2026.

Source: Nicolai Tangen sits down with Sridhar Ramaswamy, CEO of Snowflake. HIGHLIGHTS: Sridhar Ramaswamy - CEO of Snowflake. June 2026.

• My Take: I’ve enjoyed this podcast series. As a massive investor in everything, Norge’s CEO Nicolai Tangen gets great guests. He also lets his guests talk (which my podcast hosts don’t do). Nicolai interview style is rapid fire questions of high-level thinkers like Sridhar.

• Interesting: NBIM runs 2 petabytes in Snowflake and ~3 million queries a day. Norges spent “years” cleaning up their data…

• Snowflake separates storage from compute.

• Software development has always been a “craft” like being concert pianists … no longer. Native coding agents are amazing.

• Data more accessible. AI is helpful in the data modernization effort.

• “Privacy is like exercise”. Talk about it, don’t practice.

• Ramaswamy’s point: the manual data engineering that’s gated alt data for a decade, cleaning it, re-plumbing pipelines, migrating off legacy systems, is collapsing toward near-zero. A one-column pipeline change went from a week to an hour. Migration went from years to weeks.

• Lastly, I really like the message around resilience to failure. Sridhar’s keys to good life: hard work, resilience, malleability.

HIGHLIGHTS (31-Minute Run Time)

• Minute 01:00 – Snowflake background; separate storage from compute

• Minute 05:00 – Anthropic as a competitor (and other competitors)

• Minute 07:00 – Snowflake benefitting from AI revolution

• Minute 09:00 – AI make the data you have much more accessible

• Minute 10:30 – MCP

• Minute 11:00 – Agents changing how you work

• Minute 13:00 – barrier created by messy data (this is getting better rapidly)

• Minute 15:00 – GDPR discussion

• Minute 16:00 – data centers in space?

• Minute 17:00 – Sridhar background (Google, Neeva, Snowflake, etc)

• Minute 20:00 – how software engineering is changing

• Minute 23:00 – “weekly war room”

• Minute 26:00 – setting culture

• Minute 27:00 – Sridhar growing up; education

• Minute 30:00 – advice to young people; resilient to failure

SOURCE: Freeman Lewin published Neither Alone, Both in Sequence: Human-Agent Collaboration, Intellect, and the Information Frontier

Note: Freeman Lewin’s Brickroad is also the sponsor of this week’s ADW.

• My Take: This is deep. The amount of information is expanding at a rate beyond our ability to comprehend…and our efforts to harness the information increases the rate at which information grows.

• Agents become the answer.

• Exponential thinking is hard for humans.

BONUS: Reminds me of this article from Jordan Morrow: You Are Living in an Exponential World. Stop Thinking Linearly. June 2026

I am overwhelmed with good options.

How to pick this, not that. When is the timing just right?

Yordan Ivanov published The 4 Questions That Decide if Your Idea Deserves Your Time. June 2026.

1. Start with the pain

2. Research

3. Think of different solutions

4. Talk to the people it affects

My Take: I add articles and interesting things I come across to a Claude skill I created called “crucible” … this places the article in a 2x2 box (see below).

Most things are interesting, but not urgent . Claude “notes” them and (in theory) resurfaces them when the time is right.

Example: I came across an article sharing best practices on how to most effectively market your case studies. We at SymetryML are pulling together case studies on how our tool improves engagement with massive amounts of clickstream data. We do not have this report ready at the moment. When it comes time to release the study, The theory goes that Claude should surface that relevant article (currently “parked”) and include best practices into my thought process.

Another thought:

Special thanks to our sponsor Brickroad!

Alpha isn’t found in a catalog. Brickroad’s information frontier agent finds alpha suppliers at the frontier - companies generating your target data who’ve never marketed it. Sign up for a 7-day Pro trial Here.

• News

• Pods

• Charts

• Final Thoughts (Skeuomorphism)

#1 – Sonal Goyal published The Identity Crisis: Why Entity Resolution Is the Missing Foundation of Every Data Product Stack. June 2026.

• My Take: this stuff is really hard, and this stuff is really important. Making sure you have high quality input data goes a long way to making sure you have trustworthy output. The author shares some best practices and advice for implementing this in practice (and yes, human-in-the-loop is still a requirement).

#2 – Patha Anbil published The AI Era: How Deep Learning is Overcoming Data Scarcity in Medical Imaging. June 2026

• My Take: This caught my eye because SymetryML’s sister company Decentra Health plays in this space. Medical research is difficult, in part because of the need for privacy. Pulling together massive datasets that are robust enough for deep research is a challenge. Get the federated learning benefits without the associated risks. Every field where data is scarce or sensitive is making the same shift, from shipping the data to shipping the statistics.

#3 – Dan Poppy published DuckDB’s agent moment (Jordan Tigani). June 2026.

• My Take: This is a transcript of a conversation between Tristan Handy and Jordan Tigani. This line caught my eye: “Data size and compute size are two different axes.”

• The part I keep thinking about is how things change when the agents arrive (theme in recent weeks has been engaging with non-human data buyers/users). Agents don’t query like we do. A “swarm of agents branching, querying, and throwing away work hundreds of times a second,” running in the background, flagging the weird number before a human ever opens the dashboard. That’s a very different workflow. We are using new tools to accelerate old workflows, when workflows should be turned on their head (see Skeuomorphism in Final Thoughts).

BONUS: Dan Evans published Trust Doesn’t Come with the Dashboard. June 2026.“The gap between what your data says and what your people believe is often wider than leadership realizes, and it rarely closes on its own.”

What else I am reading:

Daria Cupareanu published How I Cut Prospect Research from 30 Minutes to 2 (Using AI). May 2026.

Abhimnav Agrawal published AI is Rewriting the Economics of Outsourcing. June 2026.

Alex Wilhelm published Observability overload is drowning engineers. June 2026

Interlatent published An Overview of Modern AI Robotics from First Principles. June 2026.

Elena Verna published The Mom-and-Pop SaaS era has arrived. June 2026.

Source: Rob Sarnie from “What the Heck Fintech” interviewed Evan Schnidman: How Evan Schnidman Went From Professor to AI Founder to Leading Fidelity Labs (Part 2). June 2026.

Part 1 – June 9, 2026 (link here) – 27 mins

Part 2 – June 16, 2026 (link here) – 26 mins

• My Take: great to get the current startup view from Evan’s perspective. He shares some background on himself (academic to entrepreneur … his comapny Prattle was first to use AI to analyze earnings calls… 2019 sold to Liquidnet).

• He does talk about GTM. Startups are typically very good at product/engineering. Incumbents are good at GTM. Big companies already have the brand and the rails. I am seeing this ... there is SO much noise in the world right now, breaking through and getting the attention of the right people is key. Easier said than done. This is what I am good at, and what I like to do.

HIGHLIGHTS (Part 1: 27-Minute Run Time)

• Minute 01:45 – interview starts; Evan’s background.

• Minute 07:45 – evolution of AI; and more recent impact.

• Minute 11:30 – academia path vs industry path

• Minute 14:45 – Prattle tech & background (in the Fed: role matters more than the individual)

• Minute 23:00 – what is interesting in the startup world (fintech specifically)?

HIGHLIGHTS (Part 2: 25-Minute Run Time)

• Minute 01:00 – interview re-starts.

• Minute 03:00 – Fidelity Labs background (relationship business).

• Minute 07:00 – know where the puck is going.

• Minute 11:00 – what is the latest from Fidelity Labs; portfolio overview

SOURCE: Anthropic published How Anthropic enables self-service data analytics with Claude. June 2026.

• My Take: I agree that “data foundations” is the base layer.

• Engaging with data via a Claude interface is intriguing, but typically 95% accurate is not good enough for most use cases.

• And this will struggle when interrogating petabytes of data (right?).

Skeuomorphism.

What a cool word.

How does this relate to AI?

Early movies were just like plays, but on a big screen. Early web pages were just like magazines, but online. The new medium arrived, and we made it look familiar.

Just like in the early days of the internet, no one had yet thought of social media. Entire jobs/companies/industries are going to change.

For me, AI is doing my research and rote processes much better. This is the same workflow I would have done 5+ years ago, just faster and better. This is not peak AI. I/m trying, but I am not there yet.

How does this relates to SymetryML?

Today’s standard model for data-driven alpha:

Those with deeper pockets to buy more data and run more rows win. The unit of work is an ever-expanding collection of billions of rows. AI might help you ask different questions, or process them faster. But it's still the same race.

SymetryML changes the unit of work entirely.

SymetryML changes from “re-scan the rows” to a “statistical twin” that never grows with new rows, just updates.

The compute constraint disappears. And when the constraint disappears, so do the old questions. You start asking questions that weren't previously possible.

Special thanks to our sponsor Brickroad!

Alpha isn’t found in a catalog. Brickroad’s information frontier agent finds alpha suppliers at the frontier - companies generating your target data who’ve never marketed it. Sign up for a 7-day Pro trial Here.

• News

• Pods

• Charts

• Final Thoughts (@Alt_Data_Weekly)

#1 – Multiple authors from Deloitte published TMT Predictions 2026: The AI gap narrows but persists. June 2026.

• My Take: There is a ton of good information here. From a data perspective, there is a consistent them that data governance, access, observability, etc is essential to making AI work well. Most companies have not gone through this, sometimes painful, process.

• I have said in the past, if you own valuable data (and most companies do), even if you never externally commercialize the data, the organizational process you need to go through with your data is extremely valuable.

#3 – Inbar Rose published Why Truth and Trust Are No Longer the Same Thing. June 2026.

• My Take: Interesting articulation of one of the bigger challenges in the age if AI. Trust takes years to build and can be lost in an instant.

• If your job is to make sure the “system is working”, your job has changed.

• The consequences of low quality are bigger...if your search engine provides a bad result, you just press back-button. If an AI system attempted to resolve a billing dispute, and does so incorrectly, the customer is seriously pissed.

• The author introduces the idea of a “reversibility budget” whereby you only take a human out of the loop when it is possible to reverse the action without too much consequence. Billing questions? … maybe a human should eyeball whatever the customer is getting as the “trust tax” on that correspondence is very high.

#3 – Asymmetrix published What MCPs mean for Data & Analytics Providers. June 2026.

• My Take: I’ve heard of MCPs for some time and always felt like I kinda understood what they were. I’ve recently been using a few and incorporating them into my workflow. I am now more viscerally understanding what they are and why they are valuable. This summary from Asymmetrix was as good a description as I have come across.

• My sense is, from the data provider prospective, there is a jevon’s paradox angle, meaning that the customer will consumer far more data when they can easily get that data into their AI workflows. How do you then capture that value? That is the challenge. My experience has been that I am not even aware what data is coming from which provider (and I really don’t care…I just want it to all work together & get my a trusted response). Consumption based pricing can result in opaque, erratic bills, … “workflow based pricing” sounds cool, but no one knows what that means. Fixed-fee leaves money on the table for the data vendor.

BONUS: Michael Horrigan, David S. Johnson, and Maggie Meinhardt published Restoring trust in economic statistics: Why it matters and how we fix it. May 2026. “On one hand, replacing data in surveys with declining response rates with higher quality and more reliable alternative data may improve trust. On the other hand, the mixing of various data sources to produce published estimates may produce a “black box” type of opaque understanding, even among technical users, that results in a feeling of a lack of transparency.”

What else I am reading:

Ergest Xheblati published How Data Creates Massive Leverage in Opaque Marketplaces. May 2026.

Eliot Raman Jones & Mya Jheeta published Will SEC reporting proposal supercharge alt data providers? June 2026.

Vin Vashishta published What Google Did To Websites Is Happening To Your App Right Now. June 2026.

Jasmine Sun published The independent writer’s advantage in the age of AI. June 2026.

Source: Brian Lamar & Andrew DeCilles of Signal & Noise podcast published an interview with Dan Entrup. The Research Industry’s Self-Inflicted Wound | Signal & Noise Ep 36. June 2026.

• My Take: Bad data is a scourge in the age of AI. Dan’s company is a great example of the value created when you can produce/create high quality data.

• Expert networks waste time reaching out to the wrong people as they work to find the right expert with whom to connect. Dan shares how “bad data” leads to negative results. Not just wasted time but, more importantly, a bad “customer” experience for the very experts (the respondents) you are trying to engage.

• Two products:

  • Update & enrich (cleaning existing data)

  • Discover (recruiting new people)

• Panelist retention is key metric. Another important ratio is outreach to successful connection.

• Interesting discussion of AI moderators. The good & the bad.

• Pro Tip: Sign up for your own panel. Eat your own dog food.

• This is not a winner-take-all market, but there are A LOT of new companies being started in this space.

• Side note: if you care about the data world, you should subscribe to Dan’s “It’s Pronounced Data” newsletter.

HIGHLIGHTS (54-Minute Run Time)

• Minute 01:30 – interview starts; Aggknowledge background.

• Minute 06:30 – the problem with the expert experience (Aggknowledge founding story).

• Minute 13:00 – the current system invites fraud and bad data; Dan quantified his experience.

• Minute 20:00 – client requests are not always rooted in reality; quality vs speed.

• Minute 22:00 – how does Aggknowledge work?

• Minute 27:00 – what does success look like?

• Minute 31:00 – AI moderators.

• Minute 35:00 – focus on the problem you are solving for prospective clients.

• Minute 41:00 – what is happening in the market with regards to expert network related companies?

BONUS: Metrics & Mayhem podcast published Signal Drop: Position Before The Page. May 2026.

This is a podcast about the challenges of observability work. The best operational decision-makers get positioned before the moment. As it relates to my life, SymetryML focuses on putting SRE’s in a position where they can more thoughtfully react to alerts by sending them the alert earlier (“moved left”)and with more context.

SOURCE: Every published The Eight Levels of AI Adoption. May 2026.

• My Take: I tried Fable 5 and blew through all my credits in about 5 minutes. It was like using a flamethrower to light a candle.

On this scale created by Every, I am at level 4/5.

I set up a new X handle (@alt_data_weekly) and created an automated workflow where there is a automated post upon publication of this newsletter, and then one more post the following week.

Why?

1. To see if I could build it.

2. To see what kind of traction I can get on X.

Feedback welcome.

Thanks!

Subscribe now

Special thanks to our sponsor Brickroad!

Alpha isn’t found in a catalog. Brickroad’s information frontier agent finds alpha suppliers at the frontier - companies generating your target data who’ve never marketed it. Sign up for a 7-day Pro trial Here.

• News

• Pods

• Charts

• Final Thoughts (Teaching your AI)

#1 – Asymmetrix published The next wave of Data & Analytics M&A: research firms are buying their moats. June 2026.

• My Take: Proprietary inputs become a key value layer. Proprietary inputs cannot be replicated and it gives you an edge. This value is amplified as AI commoditizes the organization and synthesis layer. If you have proprietary data, you are well positioned.

• I really liked the idea that you still need an expert human to be the “authoritative voice that establishes their data as the industry’s reference point”.

• Non-sponsored take: Asymmetrix does good work in the space and is a must follow for “stakeholders in the Data & Analytics industry”

#2 – Ben Lorica published Your Enterprise Data Deserves Better Than a Chatbot. June 2026.

• My Take: As systems get bigger and more complex, you need to watch how multiple signals behave together to avoid painful disruptions. Watching single values against static thresholds is monitoring (“univariate”). Watching the behavior of entire systems is the needed next layer (“multivariate”). AI is amplifying this need.

• Disclosure: this is in line with what my company SymetryML is doing.

#3 – Hugh O’Connor published What I Learned at Robotics Summit: The Physical AI Data Market Is Scaling Faster Than Most People Realize. May 2026.

• My Take: the next hot job is data sourcing expert. Hugh shares with us his takeaways from the recently attended robotics summit. These physical AI tools need training data. He compares the maturity curve this market is experiencing to the maturity curve experienced in the early days of alternative data for hedge funds.

• AI requires good data. This is becoming viscerally obvious to me as I get better at using these tools. Someone has to oversee that process.

BONUS: Larry Ellison (via Vivek Sen) on the need for proprietary data. “… for these models to reach their peak value, you need to train them not just on publicly available data, but you need to make privately owned data available to those models as well.” May 2026.

What else I am reading:

David Rotman of MIT Technology Review published A reality check on the AI jobs hysteria. May 2026.

Jordan Hauer published How We Work With Data Providers Today. May 2026.

Paul Bloom published Moneyball for book publishers and Substack writers. May 2026.

Dan Entrup published Futurum + ETR: The Future of Tech Sector Intelligence. May 2026.

A-Team Insights published Where is the Edge When Everyone Has the Same Alt Data? May 2026.

Charlie Simionescu-Marin published I’m pretty bullish on agentic commoditisation of code ripping up closed-source software moats. May 2026.

Source: Coatue Co-Founder Thomas Laffont, sat down with Boris Cherny creator and Head of Claude Code at Anthropic. May 2026.

• My Take: Things are changing quickly in the AI space. It is interesting to hear from the front lines. It is good to know, when Boris was asked about his thoughts 3-5 years out, he responded that he thinks one month at a time as things are changing so quickly.

• The singular focus at Anthropic: Enterprise coding. How to “productionize” it.

• He joined Anthropic Labs on the prototyping team. I’d say they had some success. They created 4 things:

  • Claude code

  • MCP

  • Desktop app

  • Skills

• These products all had immediate PMF internally

• “Agentic coding” … the agent writes the code…you describe what you want. This is the future. I am seeing glimpses (see “final thoughts” below) and still feel like I have keys to a Ferrari but am using AI like it’s a Vespa.

HIGHLIGHTS (12-Minute Run Time)

• Minute 01:30 – interview starts, background on Claude Code creation

• Minute 04:00 – mission is to build safe AGI (model has to interact with the world)

• Minute 04:45 – the work of a software engineer has changed

• Minute 07:00 – how to manage agents?

• Minute 09:00 – “Seeing a giant speed up in an engineer’s productivity, by hundreds of percentage points … yet we are still bottlenecked by engineers.”

SOURCE: Shaili Guru published Drift on Foundation Models vs. Your Own Models. May 2026.

• My Take: drift is hard to detect and expensive to observe. This caught my eye because detecting drift is what we do really well at SymetryML.

I’ve read about AI agents the way you read about weight loss, it sounds great, it is motivating, but it never quite happens.

This spring that changed.

With a big assist from Rhys Fisher, I’ve been building what amounts to my own little operating system (Flow OS) on top of Claude Code. Not a chatbot. Something closer to a junior employee who never sleeps and slowly learns the way I do things.

Every time I do a piece of work: prep for a meeting, scan a competitor’s earnings, pull together a deal memo, the system writes down how I did it. It becomes like a playbook. The next time a similar task shows up, it reaches for that playbook instead of starting from scratch. The longer I use it, the more it sounds like me and the less I have to explain.

That’s the part that I’m finding really interesting. This system seems to compound. The knowledge lives in plain text files I can read and edit (and lives on my computer and private GitHub), it’s not inside some model I’ll never see. So, in theory, I am not locked into Claude, I could switch and keep the work and my “context”.

I’m not going to oversell it. Some days it’s magic. Some days it is off-base & frustrating. As I “teach” this system my job, it is becoming painfully obvious to me that I am not a process-oriented person. I am inconsistent and there is a lot of context-type-stuff in my brain that a computer might never be able to incorporate.

But the trajectory is real & upward, and for a mildly technical person running a couple different streams … 1-GTM at a startup, 2-ADW, 3-advising a couple of firms, 4-plus all the personal stuff that goes along with life … this has proven to be a valuable tool.

Lastly, like Thomas Laffont from Coatue, I am trying to wrap my head around what this looks like in 3-5 years and struggle (maybe the only similarity between me and Boris from Anthropic … see above podcast).

Subscribe now

Special thanks to our sponsor Brickroad!

Alpha isn't found in a catalog. Brickroad's information frontier agent finds alpha suppliers at the frontier - companies generating your target data who've never marketed it. Sign up for a 7-day Pro trial Here.

• News

• Pods

• Charts

• Final Thoughts (Get in the Flow OS)

#1 – Arno van Driel & Mike Shi published Observability was built for humans. AI agents need something different. May 2026.

• My Take: I was the at the Linux Foundation’s Observability conference in Minneapolis last week and had a chance to hear Mike Shi present about about his company Clickhouse.

• In the alternative data space, I have consistently seen more references to data sellers setting themselves up to sell to non-human agents. This non-human-buyer market is going to be a far bigger market than the human-data-buyer market is today. This change is showing up in other areas of the data world as well.

• This theme carries into observability (& cyber-security) as well. Data in today’s systems are too big, too fast, too complex for any human to effectively monitor for anomaly, drift, and events that matter. The pressure to catch this stuff earlier and earlier is growing.

• We humans learned to sample or compress data because the raw data got too big, too complex, and too expensive to keep whole. I’d say there should never need to be a “cost vs business case” tradeoff decision. The business case should win. Keep the full fidelity, and let the machine read every bit of it.

• AI agents will love you for it.

#2 – Mark Wator published I’ve spent almost 20 years in the market data industry, and I’ll say it plainly: there has never been a better time to be in this business than right now. May 2026.

• My Take: I love the opening line, and agree! One thing that resonated was that “industries that previously had never been exposed to these data intensive workflows begin to adopt them”.

• This is happening slowly, but surely, as AI tools make these data driven insights more accessible. Price, and exactly how the right data products get delivered remain questions, but broader adoption is happening.

#3 – Ben Lorica published What Upwork, DoorDash, Meta, EY, and Fundrise reveal about agents. May 2026.

• My Take: data is the context that is the real power behind all the models. But your data is scattered everywhere and, I’d argue, the most important data/context might be in the human’s brain with no real way of being captured by the model.

• But even so, what we are seeing is the “…assembling, cleaning, connecting, and continuously refreshing domain-specific data at that scale is the genuinely hard part.”

• To be done right, this needs to get done on a scale and speed beyond human comprehension. There is value that should accrue to whoever does this. It might be a “real estate data” agent that becomes the trusted source. That agent (non-human) might contract with all the various data sources. Then clean, organize, score, and specifically make the best real estate data readily available to the model (or the next agent in the chain). This is hard. Value will, and should, flow to this level.

BONUS: Rachel Lynn Karry published Bruin Just Bet $1 Billion on the Part of Sports Fans Never See. May 2026. “The more environments they power, the more data they gather. The more data they gather, the more intelligently they can optimize monetization, engagement, and performance. And the more deeply embedded they become within the ecosystem, the harder they are to replace.”

What else I am reading:

Jordan Morrow published Reimagining the Workspace: Why AI is an Invitation to Innovate, Not Downsize. May 2026.

Jeremy Hermann & Duncan Gilchrest published What TK got right about data at Uber. May 2026.

Stephanie Wang from Mongo DB published Making Your Metrics LLM-Ready. May 2026.

Jason Derise published How can good ideas break out of AI noise? May 2026.

Source: Matthew Bernath published Pricing, Packaging and Selling Data Products. May 2026.

• My Take: Matthew posts these short podcasts packed with good insight (disclosure: I am an advisor to Matthew’s firm, Alternata).

• One takeaway from the NY BattleFin event a couple weeks ago is that there is a lot of data coming to market. More firms are seeing the opportunity to generate high margin revenue from exhaust data that is valuable in other domains. The question Matthew and Alternata are addressing is, how do you most effectively price, package, and sell this data product?

• Often the commercial reality is opaque. Often, there are no directly comparable offerings, and you might have little sense of the value your data is bringing to the buyer.

• Matthew suggests three pricing models:

  • Subscription access (direct feed, API, etc)

  • Usage-based pricing (per query, per API call, etc)

  • Outcome/performance-based pricing (tough to structure, but good upside for data seller; low risk for data owner)

• Other procurement traps include:

  • Data products don’t always fit cleanly into existing corporate buyer processes.

  • First buyer not always best buyer (these can be built around friendly intros, rather than something that can scale, the question is what can scale? The question is not what does buyer #1 look like, but rather, what do next 10 buyers look like?)

HIGHLIGHTS (12-Minute Run Time)

• Minute 01:00 – podcast starts

• Minute 02:00 – pricing discussion

• Minute 05:30 – packaging discussion

• Minute 07:00 – product discussion

• Minute 08:30 – first buyer not always the best buyer

SOURCE: I attended BattleFin in NY mid-May. I really enjoyed the presentation from Increase Alpha‘s Sid Ghatak. These two slides in particular. You can get all the slides by emailing: sales@increasealpha.com

My Take: Just because Gen AI is good for a lot of things, doesn’t mean it is good for everything. This is a good summary of which is which.

BONUS: Gergely Orosz and Jessica Salmon published State of the software engineering job market in 2026. May 2026.

I continue to think AI will be a job creator.

I took a big leap forward in my use of AI tools in the last month. Largely thanks to spending time with friend Rhys Fisher (The Cognitive Shift) who is out on the front edge of how we use these tools.

Using Rhys’ Flow OS, I am creating real value and find myself being more productive, and not overwhelmed.

What is Flow OS? It is like guard rails for Claude code. The system brings a persistent memory system and philosophy towards a way of agentic work. One that makes me owner of my memories (no vendor lock-in). All the context I share stays on my local machine or my private GitHub.

I still feel like I have been given keys to a Ferrari but am driving around like its a Vespa. My lack of imagination is the biggest hurdle.

Please let me know your thoughts on the below examples and if you know other ways I can set up automated workflows to give me an edge.

Getting better!!

Example: contact context agent. I organized & uploaded all my contacts (LinkedIn connections, ADW subs, personal contacts…all securely on my local machine). I am being reminded how everything fits together. Serendipity comes easier with great context. One concrete example is that this agent looked at an upcoming conference agenda, picked out a presenter who was first degree LinkedIn connection, but I really didn’t know that well. This person was also on my ADW distribution list. This gave me a chance to reach out prior with a thoughtful comment…we set up a meeting at the conference, which resulted in a partnership conversation that is getting started. I would have missed this entire chain of events.

Example: competitive intel agent. I created a list of companies that I want to keep track of. I first set up a full research report on each (found a great tool to use for this, and then spent quite a bit of time tuning the process, so this “agent” now does this work for me). Today the system sends me an email every weekend with all the most recent news, blogs, podcasts from each competitor, summarized for me, with the angle of how that commentary might specifically relate to SymetryML (red, yellow, green relevance). Any new company I come across, I just ask it to be added to this process, the full competitive intel report is created and the company is added to the weekly summary.

Example: publishing agent (more on this in coming weeks). I started a book many years ago and finally got it finished with Claude’s help. This was actually more time consuming than you might think.

Example: distribution agent. I will aways personally write the ADW. While I am sure an AI agent could do a version of this work for me, I learn by pulling this all together. Plus, I enjoy it.

The process is this: I send a Substack email Friday 7am ET, and do a single post on LinkedIn. I’ve set up a system such that, when the ADW email is sent, Flow OS breaks the ADW down and pulls together a series of related X (Twitter) posts. The goal is greater distribution. I have not yet automated this as I need to work to improve the social media posts Flow/Claude creates, but there will be a time when my newly established Alt Data Weekly X handle will have automated short summaries of the ADW. Open to ideas here…still fine tuning the output.

If you want to get in the Flow OS let me know. Perfect for small teams looking to make AI work for them. Happy to discuss and learn with you.

Subscribe now

Special thanks to our sponsor Brickroad!

Alpha isn't found in a catalog. Brickroad's information frontier agent finds alpha suppliers at the frontier - companies generating your target data who've never marketed it. Sign up for a 7-day Pro trial Here.

• News

• Pods

• Charts

• Final Thoughts (I love email)

#1 – Didier Lopes published Why an Apps Marketplace for a Financial Workspace. May 2026.

• My Take: I’ve used all the data marketplaces at various times in my career, with very little success. So this article caught my eye. Perhaps this is the moment when marketplaces will finally deliver on their long await promise due to the advent of open protocol standards like MCP. I like the idea of focusing on the workflow rather than the data (the details of which can get messy). Assuming you are working with data you can trust, yes trust still matters, the value is in the workflow.

• Formal announcement from OpenBB found here.

#2 – Michael Spencer published Tensions mount over AI Consequences on the Future. May 2026.

• My Take: I am an optimist on AI and think there won’t be enough humans to do all the jobs humans will create using AI. After two data- & AI-related conferences in two weeks, I think that point-of-view puts me solidly in the minority. When Citadel’s Ken Griffin is feeling “fairly depressed” after watching AI agents work, it is worth revisiting my opinion. The question is, what do these smart Citadel employees do when they are given a tool that 100x’s their productivity? These aren’t the kind of people who will work less hours. They will create new ways of working, and new ways of making money, and new ways of adding value.

#3 – CDO Magazine published What Happens When AI Agents Become Data Consumers? Monte Carlo CEO Barr Moses Explains. May 2026.

• My Take: Far more data will be sold to non-humans than humans in coming years, if not months. The non-human data market will be far larger than the human data buying market in both quantity of data and value of the data being transacted. Most data sellers are not set up for this type of transaction and a massive opportunity in coming years is creating the infrastructure (governance, trust, etc) to make these types of transaction happen.

BONUS: Jody Hesch published Make Yourself Fungible. May 2026. “You get what you give. So, make giving a habit.”

What else I am reading:

Maribeth Martorana published Why Most AI Initiatives Fail Before They Scale. May 2026.

Ergest Xheblati published Creating Leverage with Data: How Nielsen Ratings Changed TV Shows and Ad Buys. May 2026.

Auren Hoffman published if you can’t get a job today, it’s your fault. May 2026.

Scott Galloway had the audience. Here’s what he came for. May 2026.

Carbon Arc announced they are launching a new consumer transaction panel.

Source: Hugo Bowne-Anderson of Delphina’s High Signal podcast interviewed Chris Fonnesbeck The 100-Year Lead: What Baseball Teaches Us About the Future of AI. May 2026.

• My Take: I like baseball and have been wanting to listen to this podcast since I saw it was published.

• I like how he frames “observing causes” rather than “observing outcomes”

• Move from outcome-base statistics (like RBIs for a batter, or wins for a pitcher) and move towards process-based statistics (hitting the ball hard, or spin rate). The key is to focus on what is predictive vs what was random & knowing the difference. In other fields, this might be called separating signal from noise.

• The Hawkeye camera seems to have been a big breakthrough that is powering all this data analysis. These are the cameras in every MLB stadium that capture everything … and I mean everything … about the game.

• To me this really highlights the importance of asking the right question. Paired with a Hawkeye camera, baseball manager & a data scientist can do a lot of interesting work. In fact, the game has literally changed as a result of these questions (the shift has been outlawed, terms like “run value per event” have been invented, OPS is more important than simple batting avg).

• Wouldn’t it be interesting to have a personal Hawkeye camera that can measure what you are doing and how you can do generally better in life?

HIGHLIGHTS (56-Minute Run Time)

• Minute 04:00 – interview starts/Sabremetrics.

• Minute 06:30 – observation data science / Bill James.

• Minute 09:30 – culture change / why baseball?

• Minute 18:00 – thinking differently. Methodological advancements. Hedging risks.

• Minute 20:00 – the science of expressing value of a player (ie WAR).

• Minute 24:00 – what’s new with baseball metrics; data collection technology.\

• Minute 32:00 – skill sets needed and ways of thinking that make people good at this. Baysian thinking.

• Minute 40:00 – PyMC (2 book recommendations).

• Minute 45:00 – some interesting examples from baseball stats.

• Minute 50:00 – is baseball a leading example for wider industry and how data will be used?

• Minute 52:00 – how to get into baseball metrics.

SOURCE: FINOS published The 2025 State of Open Source in Financial Services. May 2026.

Conclusion: “AI is both the fastest moving and most strategically significant open source technology for financial services. Adoption has shifted from hesitation to investment, positioning GenAI as a foundational enabler of productivity, customer experience, and innovation. Yet, challenges remain in scaling prototypes into production-ready systems and finding individuals with the skill sets that can make this happen.”

I love email.

It is my primary source of receiving information.

I jealously care for my inbox and am careful to delete useless notes and filter junk.

Much like your social media feed, if you let the information flow control you, it will become noisy and useless.

Take control, be diligent and ruthless about unsubscribing, marking junk, deleting nonsense.

Source of some inspiration for this: Tomasz Tungus published The 6 Messages That Actually Matter. May 2026.

Subscribe now

Special thanks to our sponsor:

Announcing Alternative Data Consensus estimates - Turning fragmented alternative data into trusted insights – Details HERE.

• News

• Pods

• Charts

• Final Thoughts (face-to-face)

#1 – Duncan Gilchrest & Jeremy Hermann published The Data Context Trap. May 2026.

• My Take: These guys do a good job of framing problems and offering the solution. I feel like I am living these problems on a very small scale as I try to incorporate Claude into my day-to-day.

  • Problem 1: Collection is much harder than pointing Claude Code at a few sources (yes!).

  • Problem 2: Validating that context is right is hard (I am finding the more I use the system the better it gets at understanding my context).

  • Problem 3: Maintenance is a full-time job that nobody wants (why can’t this thing just do this automatically?).

• But you need to address all three: If collection is incomplete, validation can’t catch what’s missing. If validation isn’t automated, maintenance becomes manual review of every piece of context after every change. If maintenance lapses, the context you collected becomes actively dangerous.

#2 – Dominika Michalska published AI-Ready Data Is Not Decision-Ready AI. May 2026

• My Take: It will be interesting to see where the blame falls when an AI systems makes an error that costs real money/jobs/careers. “Can we responsibly act on this output?” Where should human accountability fit into the workflow? We need transparent, reliable output so human judgment can then be applied to make the final decision.

• “Decision ready AI” is a data architecture issues where accountability has been designed into the system. My sense is the answer is low-level decision that won’t compound are where these AI systems will start, gaining trust and growing from there.

BONUS: Related to Dominika’s article above is Dylan Anderson’s The Unfettered Capitalist Vibes of AI. May 2026. “That requires a different accountability model. Most companies haven’t built one. They’re still operating on the assumption that the person who generated the analysis is also the person who’ll catch the error in it, and that assumption is becoming less true as AI adoption increases.”

What else I am reading:

Robert Norcross published Private Markets Have Institutionalized, but Operations Have Not. May 2026.

Brickroad Thesis The Data Multiplexer for the Agent Economy

Roman Ostrovski, Harish Gaur and Antoine Amend of Databricks published MCP Marketplace brings real-time intelligence to agentic applications. May 2026.

Thoughts this was cool for Substack people: Your Substack data, visualized in seconds

Source: Bilal Hafeez of The Macrohive Podcast interviews Deepak Gurnani published Why Data Quality & Problem Framing Drive Investment Edge Using AI. May 2026. (Full 42-minute episode here)

• My Take: This is a good high-level conversation from a seasoned institutional investor about working with data with the goal of generating alpha.

• 5 steps to successful application of an AI model for investment mgmt (everyone focuses on #4 …. But the real edge is steps in Steps 1,2,3)

  • 1- Problem framing

  • 2- Data curation

  • 3- Feature engineering

  • 4- Model training

  • 5- Evaluation

• Conventional data (lots of relatively clean data) … the edge = how do you interpret this data.

  • vs

• Alternative/unstructured data (typically messy data) … the edge = have a clear hypothesis; combined with good quality controls.

• Importance of multiple data vendors for double checking sources.

• The number of vendors publishing transcript and vendor calls has exploded with the advent of AI. How do you apply the different NLP methods … there are meaningful differences & investors have gotten much more sophisticated with how they engage with transcripts and textual data sources.

• It is not the data…it is how you work with the data.

HIGHLIGHTS (13-Minute Run Time)

• Minute 01:00 – interview starts, five steps

• Minute 03:15 – data types (conventional vs alternative/unstructured)

• Minute 07:00 – working with complex, messy datasets

• Minute 09:00 – working with unstructured text/transcripts data sources

SOURCE: Joe Reis published April 2026 PDC State of Data Modeling Survey Results Are In! May 2026.

• “What 334 people said about the state of data modeling in April 2026.”

Who Actually Owns This?

What actually works (and yes, something does)

BONUS: Dominika Michalska published AI-Ready Data Is Not Decision-Ready AI. May 2026.

See above write up.

Face-to-face.

Happy to have attended the BattleFin conference this week. Like the Neudata & Eagle Alpha events, these are great opportunities to network, hear about the next big thing, & renew old friendships.

AI is great, but it won’t replace this type of interaction.

Subscribe now

I look forward to attending BattleFin’s Discovery Day New York May 14, 2026.

Book HERE.

• News

• Pods

• Charts

• Final Thoughts (AI is like money)

#1 – Ergest Xheblati published The Data Solutions Architect. May 2026.

• My Take: #1- Find high leverage problems. #2- Collaborate and get buy in. #3- Architect, design, implement (in other words, make it happen).

• People with this mindset … and the ability to use the tools available … will thrive.

• From a “data person” perspective, you need to understand the business to get #1 right. Find those high leverage problems where your skill & expertise can make an impact.

#2 – Dylan Anderson published Issue #56 – Redesigning Your Systems for AI. May 2026.

• My Take: this was really helpful. I’ve tried to make AI a part of my workflow beyond just a super-powered Google search and writing helper. Can it do high quality research reports? Yes … with the right prompts and feedback. But is takes time & a real understanding of workflow to do this right.

• It helps a ton to have 5-years of ADW “context” to share with AI. This is the framework Dylan shares:

  • Understand direction – my answer: I need to grow SymetryML

  • Pick one or two processes – my answer: meeting preparation (one example of many)

  • Redesign the workflow – my answer: pull together a bunch of context and ask the right prep questions.

  • Align with the tools – my answer: trying Flow OS has been really helpful to me; I have too many tools and I need to narrow this.

  • Assess, celebrate, and embed governance – my answer: to do this right, it takes way more time than I expected.

  • Feedback, scale, and rethink the model – may answer: constant iteration.

• Related thought from my favorite marketing writer & thinker, Jared Blank from Gobbledy

#3 – Carla Fried published Privacy Comes at a Cost for Fintech Borrowers. May 2026.

• My Take: want the money? Then share the info.

• This is not always as straightforward as it sounds (the “alternative data that crosses a line” section of the article … yikes!).

• I feel like we are moving in the right direction. All the additional data that is available about an individual’s credit worthiness should lead to more people accessing loans, which is a positive thing. But there is a privacy tradeoff which I feel like should be a manageable balance.

BONUS: Cornellius Yudha Wijaya published The ML Skills That Still Matter in 2026. April 2026. “In 2026, the most valuable data professionals will not be the ones who chase every new tool. They will be the ones who can build, evaluate, question, and explain AI systems clearly.”

What else I am reading:

67 Bricks published Building a Defensible Data Business: The Publisher’s Playbook. May 2026.

Cassie Kozyrkov published Attention Is All You Have. May 2026.

Matthew Jensen published It’s Time to Talk About Private Credit. May 2026.

Kevin Kliesen (StL Fed) published Comparing the FOMC’s Estimate of R-Star with Alternative Estimates. May 2026

Amanda Fioritto published We automated data validation — Here’s how we did it. April 2026.

The IMF published The United Kingdom Completes the Transitions Plans Under the IMF’s Special Data Dissemination Standard Plus. April 2026.

Source: Stan Altshuller of the Momentum in B2B Tech with AI Podcast interview Sarah McKenna of Sequentum. How AI and Alternative Data Are Changing Investment Decisions | Stan Altshuller | Ep 10 Momentum. April 2026.

• My Take: Sequentum is a great story in the alt data space. They have been bootstrapped since day one. Now over 100 employees & (always) profitable.

• Alt data is driving 95% of investment decisions (minute 2:40).

• Web data with quality and consistency.

• Discussion of major use cases. Finance use cases are basically the same as all other use cases (gov’t, corporate, etc). Google trends use cases examples shared. Risk alerts example used.

• Human in the loop … different decisions made using the same data.

• Probabilistic vs deterministic:

  • Probabilistic is very difficult to get high quality, consistent results.

  • Deterministic old school way, you know what you are getting.

• The key is transparent, clear, compliant processes.

HIGHLIGHTS (31-Minute Run Time)

• Minute 01:30 – interview starts, Sarah’s background

• Minute 03:30 – use cases for web scraped data (google trends, risk alerts, Tesla, etc)

• Minute 11:00 – Probabilistic vs deterministic

• Minute 14:00 – LLMs are awesome, but not perfect (most still need evidence-based process)

• Minute 19:00 – going from ideation to requirements

• Minute 23:00 – Captchas & being thoughtful about how you are collecting information

• Minute 26:00 – what is new at Sequentum?

SOURCE: Dylan Anderson published Issue #56 – Redesigning Your Systems for AI. May 2026.

• My Take: I’ve said for companies monetizing their data that, whether or not you ever make a dime from your data, the process of getting your data organized is worth the effort.

• Similarly, with AI, the process of really understanding your processes and how the work is done, is worth it … whether AI changes your day-to-day life or not.

BONUS 1: Michael Spencer published What AI Infographics say about the future of AI? May 2026.

• My Take: I think AI will create more jobs than there are humans to do the jobs.

“…nearly two million workers in the Philippines now work in call centers, up every year since 2016 and through the AI boom. This is Jevons paradox in action: as AI makes call center work cheaper and faster, companies are buying more of it, not less. Lower cost per interaction does not mean fewer interactions. It means more customers served, more channels opened and more markets worth reaching. The technology that was supposed to shrink the industry is fueling its expansion.”

Saw this quote about AI, and it reminded me of this having money:

“More money will only make you more of what you already are.” - widely attributed

“I personally feel like AI simply makes someone 10x what they already were.” – Daniel Beach

Source: Marketoonist:

Subscribe now

Special thanks to our sponsor, Alternata.

Alternata is the data monetization expert. We do the work to turn enterprise data into recurring revenue for data owners, and connect data buyers to novel signals they can’t find elsewhere.

• News

• Pods

• Charts

• Final Thoughts (When Intelligence Becomes Free)

I will be at BattleFin’s May 14th New York Discovery Day! Look forward to seeing you there!

#1 – Charlie Ko published Everything looks like a nail when AI is a hammer. April 2026.

• My Take: Charlie does a good job articulating some of the challenges with AI, namely that you need transparent, reliable, predictable, repeatable output. Gen AI is not always good for that type of output.

• If you go by the theory of keeping things “as simple as possible, but no simpler”, there are a lot of tasks we are assigning to LLMs that are better suited for ML.

• As we get better at using these new tools (and some old tools as well), we will figure out where to assign resources so the job gets done in the most efficient way possible.

• Disclaimer: Charlie is the CEO of SymetryML’s sister company Gemsen.

#2 – The Terminalist published The Distribution Fallacy — Moats, Mechanisms, and Misreads. April 2026.

• My Take: This is (again from the author) insightful writing. Dense. Don’t start reading unless you have some time & attention to commit.

• I like the historical analogy of cartography.

• The author uses the phrase “moats were laddered sequentially”, which I took to mean various tools came together, over time, to create defensible businesses. We really have no idea how AI will change the day-to-day for knowledge workers because this “laddering” is happening so much more quickly.

• It will seem obvious in hindsight, but we are all on the same ride, building the tracks as we go.

#3 – Tristan Handy published Five things I believe about the future of analytics. April 2026.

• My Take: I have been digging into the fact that we will be selling stuff to non-humans (AKA agents) and how does that change things, see points #3 & #4.

• I acknowledged in last week in ADW’s “Final Thoughts” that I was still trying to figure it out and don’t have a clear answer, not that anyone does at this point.

• Also, this is the first time I’ve heard the term “harness” in the context of data (see point #5). This is “the code that determines what information is stored, retrieved, and presented to the model.” Seems important.

  • 1. Analysts are going technical

  • 2. Data usage will explode

  • 3. Analytic agents are happening now, and moving fast

  • 4. Agents consume dramatically more than humans

  • 5. Harnesses are a leverage point

• Lastly, Tristan cites this article which I thought was interesting (and another example of a person who thinks AI will create more jobs). AI’s effects on programming jobs. March 2026.

What else I am reading:

Doug Laney published AI Models Are Commoditizing: Your Data Isn’t. April 2026.

FCA published Open finance roadmap: our vision for a smart data future. April 2026.

Dan Entrup published $100K Dataset. April 2026 (BTW ... this is an unpaid plug … check out Dan’s company AggKnowledge … I really got a lot of value pulling together valuable data).

Brad Oyler published Development Is Now About Automation, Semantics & Context. March 2026.

Source: Clara Grigol from Techsalerator published From Managing Billions to Building Data Infrastructure - Nomad Data’s CEO Brad Schneider. April 2026.

• My Take: How to differentiate when data becomes more accessible?

• Nomad Data – focus on what you’re trying to understand, and go from there (find the relevant data) … not the other way around, “I think I have this data … now what can I do with it?”.

• Nomad started as a marketplace & evolved into DRM (Data Relationship Manager).

• Manage inputs of your data in the age of AI strategies. This is the “data supply chain".

• Interesting one here: Customers weren’t asking Nomad for a DRM, but seeing the common problems among many clients … the DRM becomes obvious (lesson: listen to your customers).

• Changes driven by AI include that licensing model for data will change. Annual subs is historically how this works … this is changing. Moving towards a consumption-based pricing model.

• Thesis: Most companies don’t need more data, they need to use existing data better.

• Need to know what to do with data … domain experience will pay off!

HIGHLIGHTS (31-Minute Run Time)

• Minute 01:00 – interview starts; Brad’s background

• Minute 05:30 – Nomad Data

• Minute 08:00 – DRM (Data Relationship Manager)

• Minute 12:30 – AI strategy impact on data strategy

• Minute 16:00 – What does the process look like when working with Nomad

• Minute 18:00 – Samples of how HF and insurance users engage

• Minute 19:00 – Starting Nomad during COVID

• Minute 23:00 – AI driving demand … predictions

• Minute 28:00 – better process = better decisions

SOURCE: Bright Data published Data for AI 2026. April 2026.

• My Take: 97% of AI organizations now depend on real-time web data infrastructure, yet 90% face restrictions that constrain their initiatives.

• This report uncovers the architectural shift forcing AI leaders to rethink their data strategy and the solution an overwhelming majority are turning to for speed, reliability, and compliance at scale.

BONUS 1: A bunch of people from Stanford published Artificial Intelligence Index Report 2026 (423 pages! … I think only AI will ever read this … Claude .. can you summarize this for me?) … FWIW … 0 mentions of “Alternative Data”.

Too much good stuff … I browsed a few slides…here is the single slide I will share.

Slide page 193.

BONUS 2: Didier Lopes published Fragmentation kills Alpha. April 2026.

Matthew Jensen’s When Intelligence Becomes Free. March 2026.

There will never be enough compute.

This is like saying we have too much bandwidth. There is no such thing.

The more bandwidth we get, the more cool stuff we (humans) will stream.

The more compute we get, the more cool stuff we (humans) will build.

This is the best visualization of that “Jevon’s Paradox” feature of supply/demand of compute.

Human ingenuity for the win!

Jevons’ Paradox: When a system is made more efficient, taking less of an input to produce its output, demand will often rise so much that consumption of the input increases.

Subscribe now

Special thanks to our sponsor, Alternata.

Alternata is the data monetization expert. We do the work to turn enterprise data into recurring revenue for data owners, and connect data buyers to novel signals they can’t find elsewhere.

• News

• Pods

• Charts

• Final Thoughts (One Throat to Choke)

#1 – Steven Carroll published Alternative Data: uncovering opportunities amid the bluster and chaos. April 2026.

• My Take: Alt Data is no longer new and, frankly, never really lived up to the hype. Steven offers a few reasons why. Too many vendors. Too tough to wrangle. Ultimately, the juice is not always worth the squeeze.

• I remain optimistic that data will increase in value, particularly proprietary sources of data that can provide a useful signal. AI can make the data more accessible to more people (and agents!), expanding the TAM to reach the lofty levels everyone expected a few years ago.

#2 – Adrian Krebs published Direct Data Sourcing at Scale. April 2026.

• My Take: Web scraping has been a hot topic lately. LLMs getting to the point where they can do the work, and cloud compute is making things affordable … which puts this type of data sourcing within reach for everyone.

• How do we differentiate if anyone can collect this type of data? First is following industry standards (this is table stakes for any serious player), and next is to do it consistently. Then, it is to focus on delivering alpha, or unique signal, rather than just access. This is hard to do and even harder to do well.

#3 – Daniel Hall published How Alternative Data is Changing Consumer Lending Decisions. March 2026.

• My Take: In theory, using more sources of data will expand access to capital (i.e., I pay my phone bill every month, therefore I am a good credit). See our sponsor from this week, Alternata…they specialize in this stuff. Alt Data is another piece in the puzzle and does not replace existing traditional data sources; it rather enhances and tells a better story about the person or company in question. One interesting use case I’ve heard (not in the article) is that a person who regularly lets their phone battery go to zero is a bad credit risk. Related article: Federal housing agencies to allow alternate credit scores in mortgage applications.

• Privacy becomes an issue, but if the person wants to give the bank access to their detailed shopping information, then that is the consumer’s choice.

BONUS: America’s AI Action Plan. July 2025. “High-quality data has become a national strategic asset as governments pursue AI innovation goals and capitalize on the technology’s economic benefits.”

What else I am reading:

Don D’Amico published Your data vendor has a side deal. June 2026.

James Timmons published Quitting my job to research AI software design. April 2026.

Julia Meigh published Inside the Fed’s Data Playbook: Alternative Datasets Driving Policy. April 2026.

Scot Hamilton published HIPS Don’t Lie: [Expletive] Tony Gwynn. April 2026. (The Guardian’s Jose Ramirez is this generation’s version of Tony Gwynn).

The Terminalist published The Distribution Fallacy — Moats, Mechanisms, and Misreads. April 2026 (I will give this a close read before next Friday’s ADW … didn’t have time this week but wanted to include).

My friends at Gemsen (sister company of SymetryML) are presenting at Fintech Sandbox at 11 am ET on April 28. Check it out!

Source: Stan Altshuller interviews Jack Killea of Maiden Century. How AI Is Democratizing Alternative Data for Hedge Funds | Stan Altshuller with Jack Killea. March 2026.

• My Take: This is a good conversation between two guys who have been in the space for a while. They discuss, among other things, how AI is changing the way hedge funds discover and use alternative data.

• Q: How has the data consumption model changed for HFs? A: Things are changing quickly (matter of weeks!).

• For HF’s & institutional investors, alternative data (3^rd party sources) is now table stakes. Need to meet customers where they are.

• Why now? How have things changed?

  • People are getting better at marrying various data sets to get a better read on reality.

  • MC has helped data vendors expedite sales to quantamental investors.

  • People are getting better at designing the data and the nuances of how data gets pulled together and used for investment purposes.

• Too much stuff to look at. Most doesn’t generate signal…interpreting all the information and simply knowing where to look is a big deal.

• Ideas for new data providers:

  • B2B spend has been the holy grail for the past few years

  • Make it less challenging to work with (Maiden Century pitch)

  • Alt data has not been commoditized. There are infinite ways to combine the data to generate novel signal.

• AI’s impact on data strategy is essentially that the barrier to finding the next important data point has been lowered.

• AI gets better the more you work with it.

• One last one … I am a believer that AI will increase employment…. at 31:45 Jack says “headcount will go up as a result of AI”.

HIGHLIGHTS (36-Minute Run Time)

• Minute 01:30 – interview starts, Jack’s background.

• Minute 02:30 – data consumption models are changing.

• Minute 05:45 – different types of customers & some background on the past decade of alternative data.

• Minute 12:00 – does AI make working with data more scalable?

• Minute 14:00 – separating signal from noise.

• Minute 16:00 – new data partners that are interesting.

• Minute 22:00 – AI is shrinking the gap between big and small investors.

• Minute 25:00 – the flywheel of working with AI.

• Minute 29:30 – proactive agents using MC data?

• Minute 32:45 – will AI manage money for us?

SOURCE: Peter Baumann’s Strategic Patterns for AI Transformation. April 2026.

My Take: These AI tools are changing how we work (I think more slowly than everyone expects, but yes, things will change) … this is a good thought piece on how that might happen.

BONUS: Barr Moses & Oren Yunger published Chief Data Officer Responsibilities in the AI Era: What CDOs Are Telling Us (and What It Means)

What the winning CDOs are doing:

• Uplevel technically.

• Own agent-readiness, not just AI-readiness.

• Take ownership of AI observability.

• Become the enterprise guardrails function.

• Drive revenue, not just risk management.

One throat to choke.

I’ve been trying to educate myself about how humans fit into an AI future.

Humans will be in the loop. Increasingly in roles where trust, judgment, and accountability matter most. The customer still wants a person who owns the outcome. One throat to choke.

Even BCG is saying a version of the same thing in their consultant-speak (AI Will Reshape More Jobs Than It Replaces): AI is more likely to reshape jobs than erase them. The work doesn’t disappear; it moves toward what AI can’t fully deliver. Relationships. Context. Confidence. Responsibility.

There will be disruption. There will be opportunity.

If you have a role that resembles the knocker-uppers as alarm clocks came of age, move yourself closer to the customer, closer to the decision, and closer to accountability.

In the age of AI, the best seat will be the human seat in the relationship.

Source 1: BCG

Source 2: Alex Imas via Timothy B. Lee.

Subscribe now

Special thanks to our sponsor Alternata!

Alternata is the data monetization expert. We do the work to turn enterprise data into recurring revenue for data owners, and connect data buyers to novel signals they can't find elsewhere.

• News

• Pods

• Charts

• Final Thoughts (selling to non-humans)

#1 – Bryan Thomas published The data challenge impacting federal AI adoption. April 2026.

• My Take: Getting your data organized to give AI models the right access and context is key to successful AI projects. I see this on the smallest scale (my day-to-day), I can only imagine how complex this must be at the government level.

• I’ve argued in the past that the process data owners need to go through to monetize their data (see our sponsor this week: Alternata) … whether firms ever generate a dime of new revenue from their data, the process of getting your data organized is extremely valuable. AI is amplifying that benefit.

#2 – Julius Cerniauskas published What Would Happen If Web Scraping Were Banned? March 2026.

• My Take: Web scraping has a marketing/image problem. I thought it was interesting that many firms that might complain about web scrapers actually benefit by having their data made available to more people, thanks to the scrapers. Web scraping does a lot of good, and despite a few bad actors who handle it poorly, people like Sarah McKenna are working to develop industry-wide standards that will go a long way towards improving the reputation of this practice.

#3 – Matt Johnson published The Golden Record Is Dead. Long Live the Golden Agent. April 2026.

• My Take: The golden record is key to doing data the right way. I look forward to the day our Master Data Management (MDM) is better at matching and takes some of the drudgery off the hands of the human data steward. The golden record becomes a dynamic, context-aware agent rather than a dumb matching engine. Can’t wait!

What else am I reading:

Dan Entrup published The Beast Has Awoken. April 2026.

Vignesh Krishnan published Market Research fragmentation is a problem. Can vertical integration be the solution? April 2026.

Matt Robinson published Balyasny, BlackRock Mine Internal Data for AI. April 2026.

Eric Simon published Out with Data-Driven, In with Decision-Driven. April 2026.

Rebellion Research published Neudata vs Databento: Two Very Different Approaches to Alternative and Market Data. April 2026.

Apptopia is now the exclusive mobile data provider for Yipit. April 2026.

Jordan Hauer is launching The Data Market Network: DMN. April 2026.

Source: Data Faces podcast interviewed Michael Meyer of Snowflake Storytelling Is the Most Durable Data Skill. March 2026

• My Take: I really enjoyed this conversation. It resonated with me as I am discovering the importance of telling a good story in my role at SymetryML (and just how hard this is to do well!).

• What is the story, and how does that communicate what needs to be done? Michael shared the benefits of understanding the product from a deep technical level. The key is to listen to customers & pick up key phrases from those customers…this will give you “the hook”.

• Get the emotion involved.

• Semantic layer - emphasis is on context. That is why the semantic layer is getting attention today. We are moving toward a universal semantic layer … you need consistency across the organization. Semantic layer - what is this (16:00). Defining and making relationships clear. Business rules. Clear, consistent language, but so much more. Not just the what, but the why.

• AI is changing the role of data engineers. Can be much more efficient. Spin up quick samples. Helps with the art of the possible.

• Advice for young people, as you build skills, going from raw data to something that is consumable by the end user/customer (again, this is very hard to do). It is helpful to really understand the underpinnings of the data.

• Overall, be curious.

• Lastly, network in the real world (I couldn’t agree more!).

• Michael also wrote Joe’s Brew Reviews book.

HIGHLIGHTS (36-Minute Run Time)

• Minute 01:10 – Michael’s background & being an author.

• Minute 03:45 – why move from practitioner to more of a marketing role; Walt the data janitor.

• Minute 06:30 – changing mindset.

• Minute 09:00 – get the hook. Listen to customers.

• Minute 11:00 – switching back to solutions engineering

• Minute 12:00 – what is the semantic layer, and why is it getting attention today?

• Minute 21:00 – how does AI change what data engineers do?

• Minute 27:00 – what should early career professionals focus on?

SOURCE: A big group* of MIT researchers published Crashing Waves vs. Rising Tides: Preliminary Findings on AI Automation from Thousands of Worker Evaluations of Labor Market Tasks. March 2026.

*Matthias Mertens, Wensu Li, Adam Kuzee, Jonathan Rosenfeld, Brittany S. Harris, Meiri Anto, Harry Lyu, Martin Fleming, Neil Thompson

• My Take: I am a firm believer that AI will create more jobs than there are humans to do the jobs. The jobs will look different, and there will be disruption, but in the next 10-20+ years, AI will be a net positive for job growth.

Said better by Elanor Hawkins of Axios … MIT study challenges AI job apocalypse narrative … “The study challenges the idea of a sudden AI-driven employment cliff and instead points to a slower, more uneven reshaping of work.”

Related: Craig Whiting published The Age of the Engineer Is Here. The Numbers Prove It. April 2026.

BONUS: @twtayaan published The 3 Pillars of Observability. April 2026.

Inspired by this article from Kyle Poyer: Your next customer might be an AI agent. April 2026.

I have given this a lot of thought and am not sure how this will play out.

Several companies are trying to make this happen, but no one will really know what works until it is obvious … and everyone will then claim it was obvious all along.

I have built my career in enterprise sales. Essentially, relationship sales where you get to know the buyer, their company, their role, & you typically stay connected post-sale, if not for the life of the engagement. You are the human accountable for this engagement. You are the “one throat to choke”.

Selling to an agent is a different thing.

The keys to a non-human buyer (according to the article … and it makes sense to me) are transparency, clearly stating what your product does, and how much it will cost. In my mind, not many products are like this.

Even SaaS products like Salesforce CRMs, where the buyer could adapt the messy Salesforce interface to fit their workflow, and ultimately end up using about 5% of the capability of the system (but still paying the full price). In theory, that will go away, and the agent tool will customize itself to your workflow.

At the end of the day, a human needs to be accountable for the results (right?).

It is clear to me that this is not clear to me. It will be fun to watch this play out.

Subscribe now