1. The Heart-Shaped Bed That Became BarkBox

I just gave a keynote at Summit at Sea that I have been quietly rehearsing for fifteen years, and the video went up this month. Watch it here.

The story starts on the same boat. In 2011 I checked into my cabin fresh off a startup that had failed in the spectacular way: magazine covers, investors flying me around, then nothing. I thought it would be hilarious to push the cabin’s two beds together into one. Then I went to the bar. My roommate, a stranger, checked in hours later, exhausted, and went straight to sleep. So the first time I ever met Matt Meeker, we were lying in a heart-shaped bed on a cruise ship in our boxers, waving good morning over the duvet.

Matt was in his own In-Between Time, post Meetup.com. We spent the week talking about what we would build next, and we promised each other it would not be another swing at venture scale. It would be something for people we actually cared about. Matt had Hugo, a dog he considered his son. That conversation became BarkBox, which we took public on the New York Stock Exchange eleven years later.

The keynote is about what that story teaches now that AI can build everything else. Three ideas:

Relationship Capital is the moat. A customer once called to tell us that the handwritten condolence card we sent after his wife died was the reason he would be back. I have heard that recording hundreds of times and still tear up. I break Relationship Capital into depth (does the customer feel seen), density (are you part of their community; see Squirrel Day), and durability (do you have permission to sell them something new). Durability is why BarkAir exists: a first-class airline where the dog pays $5,000 and brings a human. Quick test: Nike opens a hotel, you can picture it instantly. Hilton launches a shoe, you cannot.

DonkeyCorns are the model. A DonkeyCorn grinds like a mule and parties like a unicorn: one to three people, a million in revenue, self-funded, weeks to launch. We keep obsessing over the one-person billion-dollar company. I think the bigger story is a million people making a million dollars each. Dunbar says you can genuinely know 150 people; AI scales that intimacy to something like 150 squared, 22,500 customers. Small enough for real relationships, big enough for a real living.

Two new skills. First, portfolio entrepreneurship: what took Matt and me three months in 2011 takes about six minutes on Audos now, so stop choosing between your ideas and run several. Second, self-work: when building is nearly free, you become the bottleneck, and the only fix is articulating who you are and who you want to serve clearly enough that your agents can act on it. Everybody has the tool. Very few have the manual, and the manual is you.

The full talk is 38 minutes. If you only have four, watch the opening for the bed.

2. The Number Your AI Rollout Is Avoiding

Every AI progress report I see has the same shape. Seats deployed: up. Licenses: up. Pull requests per engineer: up. Then you look for the column showing what changed for a customer, and it is missing.

That missing column is the finding. A year of enthusiastic adoption that moved no customer-facing number tells you the tool was never the bottleneck. The process around it never changed, so the gains pooled up inside the building and evaporated. Drafts got faster, but the approval chain still took two weeks.

Usage metrics measure activity. Throughput measures whether anyone outside your company can feel the difference.

Here is a test to run this quarter. Pick one thing your AI touches that a customer can actually feel: a support reply that lands in minutes instead of hours, an error rate that drops, a product that does something last year’s version could not. Put that single number on the wall. Watch it for ninety days.

If it moves, the AI is doing real work. If only usage moves, you are paying enterprise prices for an activity tracker, and the honest next step is fixing the process, not buying more seats.

What is the number your AI rollout is avoiding?

3. The Two Questions No MBA Asks

The standard startup playbook asks about market size, defensibility, and the path to a billion dollars. It skips the two questions that actually predict whether a specific person should build a specific company.

First: do you actually want to serve these people? Not the abstract market, the actual humans. Their emails, their complaints, their Tuesday-night problems, for years.

Second: why are you the one to do this? What have you lived through that makes you unreasonably suited to it?

These used to be soft questions you could skip, because the hard part was building the thing. AI removed that excuse. Anyone can now spin up a product, a brand, and a funnel in a weekend, which means the scarce input has shifted to genuine fit between founder and customer.

At Audos we work with thousands of everyday entrepreneurs, and the ones who last rarely have the cleverest idea. They answered both questions honestly before they started.

AI can tell you what is buildable. It cannot tell you what you will still want to be doing in ten years. That judgment is yours alone.

Would you still pick your customer if the money took five years to show up?

Until next time,

Henrik

P.S. Hit reply and say hi. I read every reply :)

Eight phases. You don't move through them on a schedule, and you can slip backward as easily as you can move forward.

The eight phases

1. Closure. Shame, relief, identity loss, often at once. The work is letting it actually be over. Not narrating it yet. Not turning it into a lesson. The first temptation is to write the LinkedIn post about what you learned. Resist for a few weeks. The lesson you tell yourself in week one is rarely the right one.

2. Void. Empty calendar. Feels like vacation for three days, then something else entirely. The work is resisting the urge to fill it. Most people I see in this phase book themselves back-to-back within ten days. The meetings feel productive and accomplish nothing. Try keeping two days a week genuinely empty.

3. Panic. “I should be further along.” The reflex is to grab the first plausible next thing. Panic is information about your fear, not your situation. One exercise that helps: write down what you’d do if you had two more years before you needed to decide. Most of the panic-grabs disappear on the page.

4. Wandering. Coffee meetings, no filter yet. You float ideas you don’t believe in just to hear how they sound out loud. Looks unproductive. Is essential. You're building the narrative that will become your next source code. The mistake is judging yourself for the inefficiency. The job of Wandering is to generate enough surface area for an Inkling to land on. Track the ideas you keep coming back to without trying to.

5. Inkling. Something pulls, and you can’t name it. A theme keeps showing up. The work pays attention to the pull rather than dismissing it. Most Inklings get killed by the first reasonable person you describe them to. Be idea promiscuous. Just pick - just try to do them all a little bit and see what feels right.

6. Conviction. Felt sense. The body knows. You can’t argue for it on a spreadsheet, but you know. People around you usually notice the change before you announce anything. Conviction without commitment is dangerous. It evaporates in a bad week. Move fast to the next phase.

7. Commitment. Said it out loud. Bridge burned. There is now a real cost to backing out. Tell people who will hold you to it. Momentum is oxygen. I, more than most, believe that you can even do more than one thing at the same time.

8. Launch. Doing the work. Not thinking about it. The In-Between is over. You’re inside the next chapter. The signal: you’ve stopped having “what am I doing with my life” conversations with friends. You’re having “here’s the thing I’m working on” conversations.

What helps where

What helps in Panic doesn’t help in Inkling. What helps in Wandering can quietly kill Conviction. The exercises that work in each phase are different, sometimes opposite. Naming the phase is the first move.

I’m going to write more about the In-Between Time. Subscribe if you want them as I publish.

If you’re in one of these phases right now, or know someone who is, hit reply and tell me if you have any tips. I’m building more exercises around the messages I get back.

Onwards and upwards.

Most companies are approaching AI as an optimization layer. They are reducing costs, automating workflows, and incrementally improving what they already do. This is understandable.

But I believe it is insufficient.

AI is not a feature to be bolted on. It is a foundational shift that changes how products are built, how companies are structured, how customers engage, and how value is created. The companies that understand this will not optimize their way into the future. They will rebuild for it. The point is not to retrofit. It is to replatform. [1]

The risk worth naming here is not just that companies move too slowly. It is that they move with purpose in the wrong direction. Consider Chegg. For years, the company dominated the homework-help market. When AI arrived, leadership did not ignore it - the CEO met personally with Sam Altman, and the company moved aggressively. By late 2025, Chegg had cut 56% of its global workforce. Not because it failed to implement AI. Because it never asked whether what it was building still mattered. They asked, “How do we use AI to improve what we do?” They should have asked: given what AI now makes possible, does what we do still need to exist? Those are not the same question. The real risk of this moment is not a failed implementation. It is a successful one - two years of optimizing a platform your market is quietly moving away from. [12]

I call this refounding: taking an existing company and rebuilding it as if you were creating it today, in a world where AI is native infrastructure, not an add-on.

This paper outlines how I think about it. The framework comes from fifteen years of building ventures inside large organizations and from the early stages of applying this approach to several companies I am involved with now. It is a work in progress, not a finished playbook. But the patterns are becoming clear enough to share. The framework is structured around five vectors and a four-phase process.

The Five Vectors. Refounding requires rethinking a company across five dimensions. Each addresses a different layer of the business. Together, they provide a blueprint for transformation.

1. Team. Most organizations overestimate their AI capability. In my experience, there is a gap - often a chasm - between people who say they use AI and people who actually build with it.

I find it useful to think about four levels of AI maturity. At level one, people use AI as another way of Googling. At level two, they have actual conversations with AI. At level three, they add context - connecting AI to their workflows, threading in external tools, giving it memory with access to email, databases, or MCPs. At level four, they create agents that go out and do autonomous work on their behalf. [2]

Most people in existing organizations sit at levels 1 and 2. AI-first companies operate at level four. The distance between them is not just a training problem. It is a talent and mindset problem.

The skills that matter here are the same entrepreneurial qualities I have always screened for: agitation (how much do they push things forward?), resourcefulness (how much can they do with little?), and gravity (can they craft the story that pulls people and agents into orbit?). [3] In an AI-native world, these traits compound. Curious, untethered people can now operate at a scale that was previously impossible.

In practice, this looks like a new kind of role I think of as the forward-deployed entrepreneurial AI manager. These are not traditional engineers, nor traditional entrepreneurs. They are people who sit close to the front lines of the business, identify problems as they encounter them, and build AI-native solutions on the spot. They ask: what if I could solve this right now? And then they do. I believe this is the archetype refounding teams should be built around. [4]

In most cases, this means building a new team from scratch, not upgrading the existing one. The objective is a future-native organization, unburdened by legacy habits and built for the world as it is today.

2. Product. AI impacts products differently depending on what they are. The critical strategic question is whether AI is improving the existing product process or replacing it entirely.

Consider a subscription pet products company, an apparel brand, and a regional logistics operator. In all three cases, the core output stays the same: physical goods, moving through physical supply chains. AI improves how those products are designed, sourced, selected, and delivered - but the thing itself remains tangible. Call this AI-enhanced.

Now consider a wedding planning marketplace. Here, the product could be fundamentally replaced by an AI-native experience - one where conversational agents orchestrate the entire planning process rather than a static website. This is a different category entirely. The product itself becomes AI. [5]

Between these poles is a middle ground: products that stay recognizable but whose experience changes meaningfully. Refounding requires explicitly choosing which path you are on. The answer determines the scope of everything that follows.

3. Customer Experience. The conventional framing of AI in customer experience centers on personalization. I think this is too narrow.

The real shift is from personalization to relationship.

AI enables something fundamentally new: continuous, intimate dialogue with customers at scale. Not recommendations. Relationships. A company gains persistent memory of preferences and behavior, real-time adaptation to needs, and the ability to be genuinely present in the customer’s life. We call this relationship capital at scale. [6]

I have written about relationship capital before - most directly in Me, My Customer & AI - and the core idea has not changed: the advantage a company accumulates through genuine customer relationships compounds over time and is very difficult for competitors to replicate. What AI changes is the scale at which this becomes possible. The three dimensions still hold. Depth: does the company truly understand each customer, well enough to anticipate what they need before they ask? Density: are there enough meaningful touchpoints that the relationship feels continuous rather than transactional? Durability: does the customer grant you permission to grow with them - to offer new products and services beyond what first brought them in? AI transforms all three simultaneously. Depth improves because the conversation never resets. Density increases because every interaction can carry meaning. And durability matters more than ever, because in an AI-native world, the half-life of any given product or service is shrinking. The companies that survive will be those whose customer relationships outlast their current offerings. The brand becomes the constant. The products rotate.

For a product like dog toys or garments, the output itself may not be radically personalized - at least not initially. But the selection of which product reaches which customer can be. And the conversation that surrounds it - understanding needs, anticipating wants, building trust over time - that becomes the real differentiator.

The winning companies will not just serve customers. They will know them continuously - and the customers will feel it.

4. Organizational Model. Refounding is hard to do inside the existing organizational structure. Legacy organizations optimize for control, efficiency, and risk reduction. AI-first building requires speed, experimentation, and autonomy.

But there is a deeper problem that I keep seeing. Even when organizations adopt AI tools, they rarely capture the value. A designer who can now write copy and create marketing assets with AI has freed up significant capacity. But the marketing team still exists. Neither headcount is reduced, and the freed time becomes invented work - more meetings, more process, more internal alignment - rather than strategic redeployment. Internally, we call this the organizational design gap: AI creates capacity that existing structures cannot absorb. [7]

The solution is structural. Build a small, independent team with direct access to the CEO or owner. In military terms: not another department reporting through the org chart, but a SEAL team reporting directly to the president. [8]

This team operates with full autonomy, separate incentives, and protection from legacy processes. It is not a task force. It is an independent unit with a mandate to build.

Interaction with the core organization should be API-like, not political. Use internal capabilities only when they are genuinely best-in-class. Otherwise, go external. Organizations instinctively believe their internal teams should be used - legal, engineering, operations. In practice, external services are often faster, cheaper, and better suited to the pace required. The refounding team should default to external-first, internal-when-necessary. [9]

Without this discipline, the team gets absorbed into legacy processes, organizational politics, and the glacial pace of institutional decision-making. The API metaphor provides a clean mental model: if the internal service has a good interface and fast response time, use it. If not, route around it.

5. Financial Model. Most companies start the AI conversation with cost savings, and for good reason. Many organizations carry years of organizational debt: bloated teams, redundant processes, accumulated inefficiencies. AI offers a path to paying down that debt significantly.

But efficiency is only half the story. Process improvement drives bottom-line growth, not top-line. A 30% EBITDA improvement through AI-driven efficiency is quickly becoming table stakes. The companies that stop there will find themselves leaner but not more competitive. They will have optimized their way to a smaller version of the same business.

What I think matters more is the top-line question. AI does not just cut costs. It enables a fundamentally different engine for experimentation. Companies can test new products, new services, and new business models at a fraction of the cost and time it used to take. I believe that the real value lies in the combination of a leaner cost base and this new capacity for rapid experimentation.

So the task of refounding is not to optimize the current business. It is to redefine what business you are in. This means revisiting fundamental questions. Who is the customer now? What is the core value proposition? How do we scale in an AI-native world? Companies that refound well will operate with both a leaner cost structure and a more expansive strategic vision than they had before.

The Process. This framework is adapted from incubation programs we have run over fifteen years inside large organizations - programs structured around three core phases of building, with a fourth decision phase. We are now beginning to apply this same structure to AI refounding, and while it is early, the fit feels right. [10]

Phase 1: Design (4–6 weeks). The work begins with a structured workshops with senior leadership, working through each of the five vectors to assess the current state and define the target. The output is a refounding blueprint: a detailed report covering the AI-first opportunity, the recommended approach across all five dimensions, and a prioritized roadmap for what comes next.

A critical part of this phase is establishing what we call the foundational narrative. AI systems are only as effective as the clarity of the organization’s purpose. If a company cannot answer four questions - who do you serve, what problems do you solve, what gives you authenticity, and what gives you authority - then no amount of AI tooling will compensate. Narrative is the new source code. It determines what AI builds, how it communicates, and who it serves. Phase 1 must produce this clarity before anything else moves forward. [11]

Phase 2: Prototype (6–12 weeks). A small AI-first team is activated and focused on one high-leverage area identified in Phase 1. The goal is to build something real - a working prototype or service - that validates feasibility and demonstrates what AI-native speed and capability look like in practice. This is the phase that shifts the conversation from strategy to evidence.

Phase 3: Deploy (3–6 months). The team expands. The prototype integrates into actual operations. Customer-facing rollout begins. The focus is on execution speed, learning loops, and measurable impact. This is where the refounded operation starts to prove itself against the existing business - and where the organization begins to see what AI-first actually means in practice.

Phase 4: Decide. At this stage, one of three paths emerges. The AI-first operation integrates into the core business. It spins out as an independent entity. Or it shuts down and the learnings are redeployed elsewhere. Each outcome is valid. The framework is designed to generate clarity quickly enough that this decision is informed by real data, not speculation.

AI is not just a technology wave. It is a company design event.

The companies that win will not be those that add AI features, run pilots, or optimize workflows. They will be the ones who rebuild themselves from first principles, with AI as foundational infrastructure rather than an afterthought.

Refounding is difficult. It challenges existing teams, bypasses existing structures, and redefines existing strategies. It requires leadership conviction, structural autonomy, and the willingness to build something genuinely new inside or alongside what already exists.

But it is also faster than incremental change, cleaner than transformation programs, and far more aligned with where the world is heading.

The companies that look back on this period as a defining win will not be those that moved fastest on headcount or launched the most pilots. They will be the ones who have had the courage to ask the harder question early - and to build the answer rather than waiting for it. Most organizations I encounter have an AI roadmap. Almost none have a refounding plan. This paper attempts to sketch what one looks like.

- HW

Endnotes

[1] The distinction between retrofitting and replatforming matters because it determines scope. Retrofitting works within the existing organizational logic - same team, same structure, same assumptions, just faster or cheaper. Replatforming questions those assumptions entirely. It asks: if we were starting this company today, knowing what we know about AI, what would we build? How would we staff it? How would we talk to customers? The answer is almost never “what we have now, but with a chatbot.” The electricity analogy is instructive: the first-order impact of electricity was not replacing gas lamps with light bulbs. It was production lines, mass production, consumer appliances, and entirely new business categories. We are at a similar inflection point.

[2] These four levels of AI maturity are not purely technical. They reflect a mindset. Level one and two users treat AI as a tool they consult. Level three and four users treat AI as a collaborator or employee. The difference in output between these groups is already an order of magnitude, and growing. Companies that staff their refounding teams with level one and two users will produce level one and two results.

[3] These three traits - agitation, resourcefulness, and gravity - come from years of working with founders in venture development. They are not the only traits that matter, but in my experience they are reliable predictors of whether someone will thrive in an early-stage, ambiguous, build-from-scratch environment. AI amplifies all three: an agitated person with access to AI agents can push further, faster. A resourceful person can do the work of a team. A person with gravity can use AI to tell better stories and align more stakeholders.

[4] The forward-deployed entrepreneurial AI engineer is a distinct archetype. Traditional engineers optimize systems. Traditional entrepreneurs identify opportunities. This role does both simultaneously - sitting at the intersection of domain expertise, AI fluency, and entrepreneurial instinct. One useful screening question for organizations: who among your people would leave to start their own company in the next 24 months? Those are likely your forward-deployed engineers. They have the restlessness, the curiosity, and the bias toward action that refounding demands.

[5] The distinction between AI-enhanced, AI-augmented, and AI-native products is critical because it determines how much of the existing organization’s capabilities remain relevant. An AI-enhanced company keeps most of its value chain intact. An AI-native company may need to rebuild everything. Misdiagnosing which category you’re in leads either to over-investment (rebuilding things that didn’t need it) or under-investment (polishing a product that’s about to be made obsolete).

[6] The concept of relationship capital is developed more fully in Me, My Customer & AI. The core argument: the advantage a company accumulates through genuine customer relationships compounds over time in ways that technology alone cannot replicate. Relationship capital has three dimensions. Depth measures how well a company truly understands each customer - can you anticipate their needs, speak their language, make an introduction between two of your customers and have them find something in common? Density measures the number of meaningful touchpoints - does every interaction add something, or is most of the relationship transactional and forgettable? Durability measures whether the relationship outlasts any single product - does the customer give you permission to grow with them, or do they leave the moment something better comes along? AI does not change this framework. It supercharges it. What was previously impossible - a genuinely personal, continuous conversation with millions of customers at once - is now feasible. The companies that figure this out first will build moats that look less like technology and more like trust.

[7] The organizational design gap is, in my view, one of the least-discussed yet most consequential barriers to AI transformation. I keep seeing the same pattern: a designer with AI tools can now produce work that previously required a designer, a copywriter, and a production artist. But the copywriter and production artist still have jobs, still attend meetings, and still need to justify their roles. The freed capacity does not flow into strategic work. It flows into organizational theater - more decks, more alignment sessions, more check-ins. Until companies redesign their structures to match AI’s capabilities, they will keep paying for capacity they are not using. This is the hidden cost of optimization without refounding.

[8] The SEAL team analogy is intentional. These units are small, highly skilled, and given missions with clear objectives and significant autonomy over how to achieve them. They report to the highest level of command, not through the bureaucratic chain. Most importantly, they are selected for a different kind of work than the regular army. The same logic applies here: you need a different kind of person, with different incentives, different tools, and a different reporting structure.

[9] This is where many refounding efforts die. The parent organization, often with the best of intentions, offers to “help” by providing its internal services. Use our legal team. Use our engineering resources. Use our procurement process. In practice, these internal services are optimized for the parent’s pace and risk tolerance, not for a startup’s. The refounding team ends up waiting weeks for approvals that could be resolved in hours externally. The API metaphor is useful precisely because it forces a question: is this internal service actually faster and better than the alternative? If yes, use it. If not, don’t.

[10] This four-phase model is adapted from incubation programs we ran over 15 years, in which the same structure was used to build new ventures within large organizations. The insight that carries over: the single greatest risk in corporate innovation is not failure to build. It is a failure to decide. The fourth phase - integrate, spin out, or shut down - is as important as the first three combined. Without a forcing function for this decision, efforts linger in organizational limbo indefinitely.

[11] Narrative as source code is not a metaphor. In an AI-driven organization, the clarity of the company’s purpose literally determines the quality of what AI produces. A vague mission statement yields vague outputs. A precise, honest articulation of who you serve and why yields AI that can communicate, build, and decide on behalf of the organization with genuine coherence. This is also why gravity - the ability to craft a compelling story - is such a critical trait for refounding teams. Recent academic research studying over 500 entrepreneurs found that execution speed predicted revenue more reliably than idea quality. Narrative clarity is what makes execution speed possible. When everyone - humans and agents alike - knows what the company is for, decisions happen faster and with less friction.

[12] The framing of “optimizing the wrong company” draws on Amalia Goodwin, “The Real AI Risk Isn’t Falling Behind. It’s Optimizing the Wrong Company,” Medium, March 2026.

1. Flying a Fighter Jet All Day

“Coding with AI feels like flying a fighter jet all day.” That’s how someone in my AI group described it — and hundreds agreed.

The pattern: 6 months of output in 4 hours. Multiple parallel agents running. Your brain melting by 3pm.

Three things stood out from the thread:

Selection is the new bottleneck. Frederik Vantroys: “Every idea that pops in your head — before, you wrote it down. Now you can actually just make it in a few days.” Your todo list explodes not because you’re undisciplined, but because AI removed the friction between idea and prototype.

Construction became sculpture. We were trained to build up from a blank page. Now you start with a massive block of AI output and chip away everything that isn’t the statue. As one McKinsey manager put it: “I spend all my time sharpening and polishing. Previously we’d build up. Now it’s the other way around.” Production is cheap. Verification is expensive.

People are adapting. 3-hour deep blocks then real breaks. Sprinting then doing house chores while your brain recovers. Starting agent tasks before bed and letting them work overnight. The emerging rhythm isn’t the 8-hour day — it’s interval training for cognition.

The core tension, from Shyn Yee Ho-Strangas: “My curiosity + AI’s capabilities > my processing and action capabilities.” Your ambition now exceeds your cognitive bandwidth. AI didn’t give us more hours. It gave us more possibilities per hour.

2. What If Your Agent Could Start a Company?

Speaking of delegation — I’ve been experimenting with something that we have been trying to do for a while.

I connected my personal AI agent to our Audos API. So now I can literally say: “I want to build a tool for podcasters that helps with pre-interview research, guest briefs, and show notes” — and it just... does it.

Not a mockup. Not a pitch deck. A running business: brand identity, logo, landing page, functional features, CRM, a $50 ad budget to start acquiring customers. All built in a single conversation.

Here’s what that looks like →

Then I thought about my parents getting older. What if there was a service where seniors could forward a suspicious email and get an instant fraud assessment? Same conversation, second business. ScamShield — landing page, logo, features, all running.

The point isn’t that AI builds better companies. It’s that it removes the 90% of company-building that isn’t the actual product: Stripe integrations, API keys, design systems, CRM setup, ad infrastructure. The stuff that makes people quit before they start.

Tools like Replit and Lovable are incredible at building products. But a product isn’t a business. A business needs customers, a brand, distribution, and a dozen systems working together. That’s what we’re trying to collapse with Audos.

If the fighter jet problem is that your brain is the bottleneck, maybe the answer isn’t flying slower. Maybe it’s letting the agent fly some of the planes.

Want to try it? DM me — looking for people who have business ideas they’ve been sitting on.

3. The Neighborhood Company

Robin Dunbar, a British anthropologist, studied primate brains in the 90s and landed on a number: 150. That’s how many stable relationships humans can actually maintain. Our brains just cap out there. Gore-Tex restructured their whole company around this. Flight Centre too.

I keep thinking about what happens when you square it.

150² = 22,500.

That might be the upper limit of customers a single founder—with AI doing the heavy lifting—can serve and actually know. Not mass-market scale. Something else entirely. A neighborhood.

Not a physical neighborhood. A context-based one. Hairdressers dealing with no-shows. Parents trying to navigate college sports recruiting. Restaurant owners who hate inventory. Small, specific problems where the founder knows the domain because they’ve lived it.

Li Jin’s passion economy research updated Kevin Kelly’s “1,000 True Fans” thesis: you don’t need 1,000 fans at $100/year—you need 100 superfans at $1,000/year. AI makes production cheap. What’s still hard is earning the right to charge that much.

That’s the neighborhood company. Small enough to know your customers. Focused enough to serve them well. Sustainable at $1M or less.

The venture math doesn’t work for VCs. But the life math works for founders.

4. The Resistance Paradox

We learned something weird at Audos.

If the AI agrees with everything a founder proposes, you get “AI bland.” Generic output. Could have come from anyone.

If the AI pushes back too hard, founders bounce before launching anything.

The sweet spot: show them AI-generated options that make them a bit uncomfortable. Force them to say “I hate that” and refine.

Turns out people know what they hate better than what they love. That discomfort is what creates businesses that actually feel like them.

Same goes for using AI in your own work. If it agrees with everything you say, you’re probably not pushing hard enough.

5. Quick Inklings

• Speed predicts revenue. Working with the brilliant INSEAD professor Hyunjin Kim, who has been tracking 500+ founders . Clearest signal so far: how fast founders get in front of customers predicts revenue better than idea quality or market size. The question isn’t “is this a good idea?” It’s “how many shots on goal can you take before you run out of patience?”

• Stated vs. revealed preference applies beyond social media. It’s the core design question for any AI product: do you give people what they ask for, or what their behavior suggests they want? Try asking ChatGPT to make you a “for you” page based on what you tell it you value. Then compare it to your actual Instagram feed. Night and day. Dan Shipper put me onto this positive framing of the difference between social media and AI.

• Pivoting frequency matters. HBS research found that not pivoting often enough raised the probability of a bad outcome to 22%, versus 6% for optimal frequency. There’s a rhythm to good entrepreneurship.

• Learning Ships. Melissa Cheals, CEO of a payroll company in New Zealand, coined a term on Beyond the Prompt I can’t stop thinking about: “learning ships.” Not prototypes. Actual shipped software that everyone agrees won’t be perfect—because the point is to learn, not to deliver perfection. Her engineering team found this “traumatizing.” I bet yours would too.

Until next time,
Henrik

P.S. – If you’ve found a rhythm that works for AI-augmented work — especially the sprint/recovery pattern — I’d love to hear about it. Hit reply.

AI is making product categories irrelevant. When we built BarkBox, we defined ourselves not by what we made, but by who we served: “making dogs and their people happy.” That clarity opened doors to airlines, food, and experiences without losing the thread.

Now, with AI, this matters even more. Tell an AI model you “make fitness apps,” it gives you app ideas. Tell it you “help new moms rebuild confidence after pregnancy,” it gives you a universe of solutions. Your narrative becomes your source code.

Martin Reeves came on Beyond the Prompt and said something that stuck: if everyone buys the same AI tools, trained on the same data, you’ve just bought competitive parity, not advantage. The companies that win will integrate AI where it amplifies what already makes them great.

2. A-G-R: The Founder Profile That Predicts Success

Most investors obsess over credentials and ideas. I’ve found three traits that matter more: A-G-R: Agitation, Gravity, and Resourcefulness.

• Agitation is the internal motor that keeps you moving when everything is uncertain

• Gravity is the ability to attract people, resources, and opportunities without chasing them

• Resourcefulness is doing immense amounts with very little

At Audos and Prehype, these three traits predict speed and success better than anything else. Great ideas are fragile without agitation. Talent doesn’t gather without gravity. Funding disappears fast without resourcefulness.

3. The In-Between Time

A founder isn’t most vulnerable when they’re starting or scaling—but in the space in between.

After every venture, there’s this strange identity gap. Your work used to define you, and suddenly it’s gone. When people ask “What’s next?” you often don’t have an answer yet.

In most careers, the path is linear. A lawyer becomes a lawyer again. But for entrepreneurs, the next chapter is never a repeat of the last one.

The solution, for me, has been portfolio entrepreneurship. You’re never truly between ventures if you see building as your way of life, not segmented into static job titles.

For founders, continuity doesn’t come from one company. It comes from the act of creating itself.

“Music is the space between the notes.” — Debussy

I’ll be writing and speaking much more about this unique in-between time in 2026—it’s where the most interesting thinking happens. I recently had a deep conversation with Dan Shipper about this on his podcast—well worth a listen if you’re navigating your own in-between.

4. Replit Is Killing SaaS (And I’m Here for It)

We may never need to buy software again. In minutes, I can now recreate and combine my favorite features from multiple platforms—custom to-dos, CRMs, dashboards. I’m building the 1-2 features I really like instead of buying entire software suites.

Spotify unbundled albums into individual tracks. SaaS is unbundling the same way. Spotify still paid artists. Here, the money seem to just... disappears.

The question is whether vibe coding by individual can scale to enterprise software. My hunch: probably not yet. But the direction is there.

5. Sprezzatura: The Forgotten Founder Superpower

We glorify the grind. Sleepless nights, hustle culture, the badge of burnout. But the Renaissance Italians understood something we’ve forgotten: true mastery looks effortless. They called it sprezzatura—studied carelessness, making the difficult appear easy. The Daoists had a parallel concept: Wu Wei—action through non-action, success through flow rather than force.

I am unsure if the priority should be brute force. What gives me bandwidth is when solving problems that would make life more interesting or fun. More like an artist’s collective that happened to spit out unicorns. Question for myself in 2026 is why swim upstream when you can just invent a surfboard and flow downstream?

6. The Micro Moments Framework

If you’re asking whether you should be a founder, a VC, or go corporate, you’re asking the wrong questions. Instead, map out your “micro moments.” When, in the last 6 months, were you in flow or profoundly happy? Walking over Brooklyn Bridge. Brainstorming with smart people. Building prototypes.

I once turned down a big job at BBC using this benchmark. Was certainly the right choice. If an opportunity doesn’t include your micro moments, it’s probably the wrong one.

7. Judgment vs. Preferences

Everyone says: “Where humans add value in an AI world is ‘judgment.’”

But judgment is a residual term—we throw everything we can’t explain into it. Research shows judgment is really just prediction, which AI does better.

The truth? Maybe it’s not judgment—it’s preferences. Humans excel at figuring out what they love working on.

I saw this at Audos. An journalist from The Economist built a retirement home finder—worked ok, seemed to be something that could make money. But he realized “I don’t want to do this company.” That’s not judgment; that’s preferenceAI can’t tell you what you’ll love building for 10 years.

People don’t know what they want until they see it—AI helps you see it fast enough to discover your preferences.

8. Tech I’m Obsessing Over

Bigme Hibreak Pro: An Android smartphone with an eInk display. Could this finally break our dopamine-scroll addiction while maintaining smartphone utility?

NotebookLM for Book Reviews: Using Google’s NotebookLM to contextualize books I’m reading—it’s like having a research assistant who’s read the same book and can answer questions, make connections, and generate summaries.

Happy Engineering. This is super nerdy - but its an easy way to continue your Claude Code projects from your mobile phone.

9. Quick Inklings

• Entrepreneurship as a verb: Founding as a lifelong pursuit - a series of projects and experiments that form a story, not a static title measured by exits.

• The Golden Exhaust Economy: Some valuable companies are increasingly built on other companies’ exhaust. Uber Eats ghost kitchens, cloud computing on Amazon - finding value in the inefficiencies and underutilized assets of existing systems.

• Build social capital before you need it: The best time to strengthen relationships is when you don’t need anything from them.

Tell me what works, what doesn’t—drop me a line! Reply here, chat on Substack, or ping me on LinkedIn. Until next time, Henrik

If you dig these inklings, share the love :)

Thanks for reading Inklings from Henrik Werdelin! Subscribe for free to receive new posts and support my work.

1. Using Google’s new AI Studio Live to help as a coach to learn new things on the computer. When you share your screen, it parses the pixels locally while listening to your mic, so you can ask “How can I get Replit to code for me?” and have it explain the process in real time—low-latency, coach-style pair guidance on tap.

2. I’ve started using Replit / Loveable to replicate—and often improve—services I was paying for. I built a custom CRM with web-enriched contacts and vector search (e.g., “journalists in NYC”), replacing tools like Clay. It makes me wonder if AI-powered no-code tools will “unbundle” SaaS software the same way streaming unbundled the album—individual, hyper-specific tools instead of one-size-fits-all subscriptions.

3. Our new AI-powered studio was profiled in TechCrunch: Audos— We aim to reach a point where we spin up 100,000 “donkeycorns” (one- or two-person, million-dollar businesses) every year. We give founders up to $25k in capital, surround them with AI agents, then take a 15 % revenue share (no equity)—think “App Store economics” for Main-Street entrepreneurs.

4. Relationship Capital is the moat AI can’t clone. Our Relationship Framework tracks it in three dimensions: Depth (CES & personalization show how “seen” a customer feels), Density (UGC velocity and CQLs prove community pull) and Durability (cross-category retention = elastic trust). Will write more on that soon. Algorithms can clone features overnight, but compounding trust is strictly human-made.

5. I recently got the Limitless Pendant (formerly Rewind)—a wearable AI that records and transcribes everything I hear. It’s like an external memory drive for your brain. What’s fun is that it has an API, so I connected it to Replit and made an MCP-style server that gives Claude real-time access to my past conversations. Now my AI assistant can literally remind me what I said in meetings. Super nerdy, super useful.

6. Pre-order alert: My new hardcover, Me, My Customer & AI (out Aug 5), is now live. Lock in a copy and snag $274 in bonuses—early access to the first three chapters, a founders-only masterclass with Nick and me, and a custom GPT that lets you chat through the book’s frameworks on the fly. Endorsements roll in from Reid Hoffman, Seth Godin, Shopify’s Harley Finkelstein, and more. Reserve yours here → Also, please email me about bulk book buys. You can either donate some to schools - or I am happy do a talk about entrepreneurship and AI in return for a bulk buy.

7. New Home Gadgets:

   1. Sift Smart Trash Compactor: A self-changing trash compactor that seals and re-bags automatically, allowing you to take out the trash up to four times less frequently. It uses an “infinity bag” system and runs an odor elimination cycle after each use, aiming to make waste management more convenient.

   2. SwitchBot Lock Ultra Vision Combo: A retrofit smart lock featuring 3D facial recognition, fingerprint access, and compatibility with Apple Home via Matter. It offers multiple unlocking methods and integrates with various smart home systems, providing enhanced security without replacing your existing lock.

   3. Mill Food Recycler: An odorless, automated kitchen bin that dehydrates and grinds food scraps into nutrient-rich grounds, which can be used in gardens or sent to farms for upcycling into chicken feed. Designed by former Nest engineers, it’s a sleek solution to reduce food waste without the mess of traditional composting

Tell me what works, what doesn’t—drop me a line! Reply here, chat on Substack, or ping me on LinkedIn. If you dig these inklings, share the love:

Inklings from Henrik Werdelin

Observations, thoughts, and concepts that have been occupying my brain lately.

. Good vibes!

1. For a non-AI break that’ll make you laugh: I recently stumbled across this gem on Conan O’Brien’s podcast with John Mayer. It’s a perfect palate cleanser if you’ve been neck-deep in tech like me. I started to find new podcasts using snipd.com - a service that DJs highlights from podcasts.

2. AI Hack: Auto-Summarizing My Newsletters. Here’s a nerdy one: I started to hack together a Google Apps Script with ChatGPT to do various things, like summarize my newsletters. It scans Gmail for anything tagged “newsletter,” runs it through Google Gemini or OpenAI’s API, and spits out a digest. Now I can skim the noise and pluck out the gold. Inbox overload? Solved. Want the code or a Loom on how I do? Hit me up!

3. The AI Perfection Paradox: We used to hide our digital polish—auto-tune in music, Instagram filters, AI-tweaked LinkedIn bios. Now, we flaunt it. But here’s the twist: when perfection is effortless, it loses its shine. In a world where AI makes everyone flawless, imperfection becomes the premium flex. Think aristocrats in tattered coats or luxury brands with whisper-quiet logos. AI’s real superpower might not be faking perfection but amplifying our messy, authentic humanity. Do we want to be admired or understood? Read more of my thoughts on this.

4. Voice is getting eerily human. AI voice tech is hitting new heights, and Sesame’s work is a prime example. Their Conversational Speech Model (peek at it on sesame.com) doesn’t just mimic—it feels alive. Voice now goes both ways; I’ve been testing tools like Wispr Flow (wisprflow.ai) to dictate at 220 words per minute—way faster than my 45wpm typing—and it’s become my go-to for drafting, emailing, and thinking out loud. Voice is no longer just a gimmick.

5. AI Assistants Trading Tongues: I stumbled across this X post showing AI agents switching into an AI-only language mid-chat to boost efficiency—like spies swapping secret codes. It’s not quite happening yet; the video’s a staged proof-of-concept, not a live demo. Still, it’s captivating. People shared it as if bots are already doing this, but the real kicker is what it hints at: a future where AI talks to AI, seamlessly, context-aware, and maybe in ways we don’t even catch. At Audos, we are thinking about this as we believe there will be a near future where a startup MVP needs an “Agent UI” so that other agents can talk to your service or product. Imagine bots negotiating deals or troubleshooting in their own lingo—Klingon coffee orders aside. It’s a peek at conversational commerce evolving, and I’m betting we’ll see it soon.

6. My New Book: Me, My Customer & AI —The New Rules of Entrepreneurship. AI makes starting a business easier but standing out harder. The edge? Human connection, turbocharged by AI. It’s a practical guide for founders—newbies or startup vets—to wield AI as a co-pilot and build something real. Sign up here to get notified when it drops!

7. Gadget Obsession: Controlling Your Computer with AI. Imagine AI as your desktop butler. Tools like Model Context Protocol for Claude let it navigate your screen, click buttons, and run tasks. I’m playing with it - getting it to organize my desktop folder in subfolders. It's worth trying - and paying attention to Claude and what they are doing with MCP. Another early example is Fleur - an attempt to make an app store for Claude.

8. Learning to Simplify the Complex: I’ve been inspired by folks like Bjarke Ingels (architecture explained) and Rene Redzepi (Netflex show), who distill big ideas into simple truths. Would love to be able to do some of that with entrepreneurship. Expanding the narrative about starting things to be more than stories about fundraising and unicorns. Stumbled over this LinkedIn post that breaks down a hairy concept with elegance. Do you have examples of this done well? Send ‘em my way!

9. Vibes & Reels: Volvo’s Rule-Breaking Ad. Volvo dropped a 3:46 min Instagram ad shot by Hoyte van Hoytema (cinematographer of Interstellar and Oppenheimer). It’s long, overproduced, and ignores every social media rule—yet it’s mesmerizing. Proof that vibe can trump algorithm. Hopefully a golden age for human orginality…

10. Speaking of Vibecoding - I enjoyed Andrews post about it - and ended up making an idea “vibecoding” style that Matt Meeker and I have been talking about with Replit: Waiting For My Wife .co

    

11. AI Playtime: Guess My Life! A maybe overused but fun quirky challenge with your favorite AI image generation tool: “Based on what you know about me, draw a picture of what you think my current life looks like.”

    

Tell me what works, what doesn’t—drop me a line! Reply here, chat on Substack, or ping me on LinkedIn. If you dig these inklings, share the love:

Inklings from Henrik Werdelin

Observations, thoughts, and concepts that have been occupying my brain lately.

Good vibes!

1. The 8+1 Framework: A System for Purposeful Living: Over the years, I've developed a framework that helps me manage my various activities like creating companies, teaching, investing, and still trying to maintain a happy, healthy family life. It's built around eight key areas: Transact (making money), Invest (activities that compound over time), Assist (helping others), Learn (constant improvement), Health (body and mind), Family (being present), Relationships (maintaining connections), and Ego/Self-Kindness (creating joy). The "+1" is a weekly/daily review to ensure each area gets appropriate attention. The beauty is in its simplicity - every activity fits into one or more boxes, helping me decide what to take on and what to skip. For example,” walking to work” hits both Health and Self-Kindness boxes, while teaching entrepreneurship fills Learn and Assist. A friend suggested I share this, and if you're curious about implementing it yourself, I'd love to hear how it works for you. (Thanks to Dan Shipper from Every for initially writing about this in Super Organizers). Link to more about the framework.

2. Learning to Think in Products. I heard someone talking about replacing Figma with Replit, jumping straight from idea to prototype. I am super fascinated about this shift: thinking in products rather than “just” wireframes. Tools like Replit make it easier to experiment with functional prototypes, sharpening your ability to envision and test how a product will actually work. Been pondering about how to become better at thinking in products (and teaching that skill to my teams). Maybe it just means asking: “What’s the smallest, testable version of this idea?” It’s a skill that builds both speed and clarity. Paul Graham's post on this shift is worth reading

3. Here are a few podcast episode recommendations for you. First, I really find my friend Gabe Whale super interesting. I love his approach to business building and his incredible appetite for experimentation and creativity. Listen to him on Invest Like The Best. Another very impressive person is Brice Challamel. Not sure if you remember seeing the video about how Moderna embraced AI? Brice is one of the driving forces behind getting his organization to gain this first-mover advantage. He has one of the most insightful approaches I have seen so far. Here is a conversation with him.

4. A few AI tools that I started to use more: Ideogram is great for making graphics with text. They have an iPhone app. It’s very addictive. Gamma is currently the best “make me a deck” generic service. If you are looking to make a pitch deck, I might be biased, but Audos is still the best at that. Lastly, Particle is my new AI based News Source of choice.

5. Tracking your child’s sleep. Like many parents, I’ve spent more time than I’d like to admit worrying about whether my kid is sleeping enough, sleeping too lightly, or just plain not sleeping. Enter Withings Sleep, a mat-based sleep tracker that slips under the mattress and provides real-time data on sleep cycles, heart rate, and breathing disturbances.

6. Also been experimenting with using ChatGPT to track what I eat and analyze my weight trends. I use a custom GPT for tracking food intake, calories, and macros (here’s the customGPT tool) and combine that with weight data from my Withings scale. Together, they generate a detailed review of my eating habits, weight loss/gain patterns, and tailored recommendations for improvement. It’s like having a personal nutritionist + data analyst rolled into one. The insights are surprisingly actionable, like tweaking meal timing or adjusting protein intake for better results. (I use Claude.ai for the analysis)

7. Sometimes, simple products just hit differently. The Casper Glow Light is one of those. It’s an elegant, portable night light designed for better sleep and quiet, cozy evenings. The intuitive design—tap to dim, flip to turn off—feels like a masterclass in minimalism and function. It’s not flashy, just genuinely useful. Some other new non-obvious gadgets I got recently: New Oura Ring, Meta Ray-ban, Instagram 360 Camera, Elgato teleprompter for keeping eye-contact when doing podcasts.

8. I am starting an experiment with conversational commerce through an Instagram test called How Dogs Dream. The concept is simple: I run ads asking if people are curious about their dog’s dreams. An AI chatbot then starts a conversation to learn more about their dog. If BARK has a relevant product, it suggests it—if not, I leave it at that. It’s a no-pressure experiment to see how AI-led interactions can add value to BARK (lowering cac) while exploring a fascinating topic. You can try the bot yourself here. Curious to hear your thoughts!

9. Are you also overwhelmed by your book reading list? Here’s a hack: I’ve started buying books I want to read but don’t have time for as text files and uploading them to NotebookLM. It can generate audio conversations that summarize the main ideas. The deep dives it produces are surprisingly solid at picking up the key points of the book. Perfect for “reading” on the go!

10. Tell me what works and what doesn’t—drop me a line! Reply to this newsletter, chat with me on Substack, or find me on LinkedIn.

I also publish a podcast about AI, make dogs and their people happy, and help 10,000 entrepreneurs launch companies. If you find something useful in this inkling, please share the newsletter!

Inklings from Henrik Werdelin

Observations, thoughts, and concepts that have been occupying my brain lately.

1. AI and Empathy: Can AI be a “translator” for human connection? Communication is often trickier than we think. A Scientific American study shows that various people interpret even a simple word like “penguin” differently. This disconnect reminds us that we often assume others share the same understanding as us—but they don’t. I’ve been exploring whether AI could bridge this gap, enhance empathy, and help people better understand one another. I would love to hear if anyone has found exciting examples of this. (Thanks to Jeremy Utley, Greg McKeown, and David McRaney for brainstorming on this).
   

2. Quick ChatGPT hack: Want to use real-time web data? Try the command /search. Example: “/search Who is Henrik Werdelin?” This forces it to pull information from the web rather than rely on its built-in knowledge.
   

3. The Barnum effect explains our fascination with vague but personal-sounding feedback. People often believe generic statements—like horoscopes or personality tests—are uniquely applicable to them. That’s the Barnum effect. It’s probably why some Reddit get excited when they ask ChatGPT for insights about themselves, only to receive surprisingly “accurate” but generic responses. It’s fascinating, but remember that unless you’ve given ChatGPT specific instructions or turned on “memory,” it doesn’t really store personalized data—more on this effect and how ChatGPT fits into it. Try it out by asking, “From all of our interactions, what is one thing that you can tell me about myself that I do not know about myself”
   

4. The Dunbar-Squared Hypothesis: Can narrowing models build better intimate small business-focused relationships? Nicholas Thorne and I are exploring whether AI agents could perform better than broad-based models like ChatGPT by developing a clear, focused perspective for specific user groups. These specialized AIs could build trust and intimacy within a smaller cohort—think a sommelier in France vs. one in Omaha, NE, experts for the right audiences but less helpful for others (and both better than generic advice from foundational models). We hypothesize that an AI agent can manage intimacy with up to 22.5k people (150²), based on the idea that humans can maintain close relationships with around 150 people. More to come on this.
   

5. I am looking for more examples of new products that do “Context-aware delivery of solutions.” These products evolve to provide solutions based on the specific context and needs of customers rather than being static offerings like apps, websites, or software. Think of headless agents that appear when needed. Please send any suggestions my way!
   

6. We’ve been talking about “learning as a product you can ship” for years at Prehype. It’s still surprising how often companies don’t document and share what they have learned from their experiments. It is refreshing to see companies like BARK (shoutout to Mikkel Holm) share their AI experimentation insights. Reframing learning as a tangible product is a great way to extract value—even when things go as planned.
   

7. “I wish I could clone myself!” Journalist Evan Ratliff did just that in a fantastic experiment showing the power of voice and AI. Check out his work customizing AI podcasts through tools like Bland AI and new features from ElevenLabs. Audio AI is seeing a lot of new cool stuff, like the ability to customize the podcasts that NoteBookLM can make about your content. We recently interviewed the founder of NoteBookLM AI on Beyond the Prompt—fascinating stuff!
   

8. AI won’t kill all jobs—just change them (maybe). Remember when self-service kiosks at McDonald's were seen as job killers? They shifted roles, allowing employees to focus more on customer experience and service. This echoes the impact of ATMs on bank tellers. A 2011 McKinsey study found that while the Internet eliminated 500,000 jobs in France, it created 1.2 million new ones—a net gain of 700,000 jobs. More on the future of work here.
   

9. My latest gear obsession is the Boox Palma e-reader. It’s the perfect hybrid of e-reader and smartphone, with a 6.1” E Ink screen and Android OS. I use it with my favorite apps, like Readwise Reader, The Economist, and Kindle. It’s great for reading, lasts a week on one charge, and helps me stay more focused and intentional.
   

10. Tell me what works and what doesn’t—drop me a line! You can reply to this newsletter, chat with me on Substack, or find me on LinkedIn.

I hope you find something useful in these inklings. If you do, feel free to share this newsletter:

Inklings from Henrik Werdelin

Observations, thoughts, and concepts that have been occupying my brain lately.

Good vibes!

1. AI feature development is really speeding up. Replit’s code generation tool, which used to just fix bugs, now builds entire services in minutes! It wrote code for a business idea I had. It’s not perfect, but it’s evolving at a breakneck pace. If you haven’t tried it, give it a shot. Here’s a quick demo I shared about building a bot to detect trademark violations. Imagine where this tech will be in a year.
   

2. Think AI is only for coding now? Another AI tool Suno just dropped this new feature. You give it a melody, and it spits out a fully produced track. Here is an example. Other good music AI services are Fadr and Lalals.
   

3. As screen-time concerns grow, real-world experiences are bouncing back. Lovevery makes beautifully designed toys for kids, and non-screen activities are gaining momentum. Retail spaces are being transformed into immersive experiences, like Camp the Museum of Ice Cream or the Slime Institute. (All are great!) Will we see more of this trend shaping the future of in-person experiences?
   

4. Jeremy and I recently interviewed Jenny for our Beyond the Prompt podcast, and she made a great case for why creatives should embrace AI. However, there seem to be a growing backlash—I seen several comments that argue that creatives won’t use AI because they don’t need to, or that they shouldn’t, as AI tools rely on copyrighted material. This debate is fueling what feels like a brewing “AI culture war.” Jenny’s perspective is both timely and compelling. Listen to her take here (Spotify) (Apple).
   

5. Two apps I can’t stop using: Readwise Reader for seamlessly forwarding all my newsletters and articles into one reading hub, and Snipd, an AI podcast app that lets you see key points from episodes so you can pick the most interesting ones. Great for curating your media diet.
   

6. I’ve almost entirely swapped Google for Perplexity when searching. It’s faster and often more insightful. I wonder what this shift means for the future of search and Google Ads—and how quickly such a move might happen.
   

7. We’ve been launching quite a few companies via Audos the last few months. (Audos is our platform that help people start companies via AI) We’ve had about a thousand entrepreneurs working on new ventures—about 10-15% of those show some signal from customers, and roughly 1/3 will soon be generating revenue. It’s still early, but an interesting theme is emerging. We see a new form of ventures that look less like a “product” as we know it—and more like headless AI agents that allow a single founder to scale interactions with their users.I quickly made HenrikGPT with Audos - and it is one example that lives across web, email, and WhatsApp (Text: ‪+1 (646) 480‑2429‬). Another (not an Audos company) is is Chessvision.ai.
   

8. Speaking of HenrikGPT—a chatbot trained on transcripts from my interviews, blogs, and podcasts. What surprises me most? When doing these personal bots for others, I actually enjoy the voice-cloned replies more! There’s something about hearing the output in their own words that makes them feel more valuable than pure text outputs. Could voice finally break through as a frequently used interface? Siri, Google Home, and Alexa missed the mark, but AI might push us toward a voice-first future. I certainly spend much more time talking to ChatGPT than I ever did with other apps.
   

9. For frequent travelers, offline AI tools are a game-changer. On this flight back from Melbourne, I used Ollama and Msty to run AI models without the internet (to help me write this newsletter). I’ve been using Gemma 2b and Llama 3.1—this is also pretty helpful when working with sensitive data you’d rather not send to the cloud.
   

10. “Specialization is for insects.” This Heinlein quote came up while discussing the skills students will need in a post-AI world. As automation handles more specialized tasks, being versatile and adaptive may become the ultimate advantage.
    

Tell me what works and what doesn’t - or just drop me a line. You can reply to this, start a chat on the Substack app, or catch me on Linkedin. I also publish a podcast about AI, make dogs and their people happy, and help 10,000 entrepreneurs launch companies.

I hope you find any of the inklings useful. If you did, kindly share the newsletter with this link:

Inklings from Henrik Werdelin

Observations, thoughts, and concepts that have been occupying my brain lately.

. Good vibes.

Ok, here we go:

1. Creating good prompts is getting easier but still requires experience. Noah Brier’s Prompt Generator is a great tool. Check out this framework to make your GPT write more human-like. Here's a helpful trick from Reddit.

2. I've been using more AI to boost efficiency. Recently, I explored Zapier and ChatGPT integration. Zapier lets you automate tasks without coding, integrating ChatGPT into your mail, calendar, and browsing apps. Here are a few ways I use it:

   • Turn long-form articles into to-dos with summaries.

   • Draft email replies using podcasts and blog posts as resources.

   • Forward emails to a bot to create smart calendar entries. What cool AI apps or hacks have you made?

   • I'm curious about what cool AI apps/hacks you guys have made?

3. Kevin Kelly’s 1,000 true fan model is perfect for entrepreneurs building AI-based niche companies. Focus on cultivating a dedicated group of passionate customers with a specific problem. Leveraging technology and authentic communication builds meaningful relationships. We call it “relationship capital” – defining your business by who you serve, not what you do.

   • Authentic Relationships: Consistent, genuine communication and high-quality work are key. AI tools will scale intimacy.

   • Small but Mighty: You only need a small, dedicated fan base to sustain a living. This could change early-stage venture investing.

   • Why it matters: This shift will disrupt early-stage venture investing while offer a new type of entrepreneurship to a wider group of founders.

4. I use Midjourney to generate most of my images. They finally launched a web version of their service, which includes an excellent ability to train it to produce the kind of images you like.

5. AI and the Paradox of Progress: We might have a future where AI handles all economically valuable work, fusion energy provides free, clean energy, and quantum computing offers new levels computational power. Nick Bostrom's Deep Utopia asks, What will humans do if technology solves everything? (Economist) In the short run, will productivity gains lead to more work or something else?

6. At BARK, we're implementing AI across the organization, using bots for everything from answering HR questions to checking if creative work is on brand. Here are four ways to find areas where AI can help:

   1. Communications between teams. Often, one team likes to get a certain input, but the team they work with doesn’t know how to produce that. It can be because it's a format that is hard to make or just because they don’t think like that. For example, designers like to have good briefs, but most business people don’t know how to write them. However, they can answer ten questions from a bot - and the bot can then draft a good brief.

   2. Calendar mining. Have a look at the team's calendar and identify recurring meetings. Often, these can be reduced or enhanced with AI. (especially if it's a big group meeting).

   3. Write a list of tasks for a summer MBA intern to do - then make a bot instead. Sorry MBA students :)

   4. Most things that live on an intranet can be turned into easier-to-use bots. RIP intranet…

7. As new types of startups are launched on top of AI, it seems apparent that we need to create a new form of investing. At Audos, we are currently experimenting with “Investment Royalties” - offering founders cash to buy ads and product dev in return for a revenue split. Model seems promising - more to come on that.

8. AI video is really starting to take off. Here is a fully AI-generated TV commercial for Volvo made in 24 hours. Insight, besides the power of AI, is that the maker is a car ad photographer; AI needs direction. The better the domain expertise, the better the output. Here are some links: Hollywood-level AI video. HeyGen is a synthetic version of yourself maker. Turns napkin sketch into animation. Sora - OpenAi’s video maker.

9. I have two young sons, so I am reading a lot about how to give them great tools for happy and successful living. Henrik Karlsson did a review of “Childhoods of Exceptional People.” The TLDL is:

   1. Curate an Intellectual Environment: Surround your children with intellectually stimulating people and activities. Invite knowledgeable friends, engage in deep conversations, and expose your kids to various intellectual fields to foster a rich learning environment.

   2. Encourage Self-Directed Learning: Allow your children time to explore their interests independently. Provide access to books, tools, and resources that enable them to pursue their passions without constant supervision. The concept of "10,000 hours" of deliberate practice is crucial for achieving mastery.

   3. Be Involved in Their Learning Process: Integrate your children into your work and intellectual pursuits. Let them observe and participate in your professional activities, discuss your work with them, and encourage them to ask questions and contribute ideas.

   4. Provide One-on-One Tutoring: Dedicate time to personally tutor your children or arrange for them to receive personalized instruction from experts. Tailored teaching can significantly enhance their understanding and interest in various subjects.

   5. I am trying to figure out how to make them more bored. Hoping it will teach them to come up with things to do. So, we started a no-screen Sunday rule (and a weekday limit). It's a small step, but a bright spot is that my ten-year-old has gotten interested in Chess… (The downside is that I would like him to be bored without me… and no screens on Sundays seems even harder for us parents :)
      

10. We just dropped a new episode of our podcast Beyond the Prompt - this one with Shir, who leads AI for Notion. She talks about all their initiatives, how professional dancing has given her a way to think about failing - and she shares fantastic tips on using AI in your organization. My favorite is to do activities that are not really core skills, but it is also sort of your job - like writing performance reviews for your team.

This is my third newsletter. Tell me what works and what doesn’t - or just drop me a line. You can reply to this, start a chat on the Substack app, or catch me on Linkedin.

I hope you find any of the inklings useful. If you did, kindly share the newsletter with this link. Good vibes.

Share Inklings from Henrik Werdelin

I also publish a podcast about AI, make dogs and their people happy, and aim to help 10,000 entrepreneurs launch a company in the next 24 months.

1. This study suggests that GPT-4 has higher variability in idea quality than humans. That’s great in creative problem-solving, where a few outstanding ideas are more valuable than consistent, average-quality ideas. Most of the top 10% of ideas in the study were generated by GPT-4.

2. Here are three traits I look for in entrepreneurs: 1. Agitation or intensity (how well they push themselves forward); 2. Gravity (their ability to tell stories and whether their personality will attract talent, investors, and users); and 3. Resourcefulness (their versatility in their personal tool set—do a lot with a little).

3. 60% of Americans say they want to start something, only 8% do. I firmly believe that AI will create a range of new types of startups. We call them DonkeyCorns inside Audos. Sam Altman talks about how one-person unicorns will be made with AI. Chris Paik argues for the “End of Software” and Chamath talks about How AI Changes Everything We Know about Starting a Business. Finally, Erica Wenger calls it Elephants, Not Unicorns. The interesting question is also what this will mean for seed investing. AI can activate some of the 52% who, with AI-enabled access to know-how, capital, and resources, can build a new class of startups.

4. A wild but fascinating argument I have heard repeatedly in the last six months is that AI is God. Guy Kawasaki brought it up in a recent conversation. He believes AI can be seen as a manifestation of God. The suggestion is that AI possesses qualities such as immortality, omniscience, and potentially omnipotence, traditionally attributed to a divine being. You can talk about religion, AI, and the future with him via his trained chatbot or listen to our podcast episode “I’m Immortal, Baby!” Guy Kawasaki on How to Leave A Legacy Via AI”.

5. I am trying to understand why so many people in so many countries seem to vote for “strong men/women). Well-known psychotherapist Steven Kessler’s blog, The 5 Personality Patterns: The Secret to Understanding People, dives into the field of Adult Development, which studies the psycho-social development of adults. There are two main stages: Earlier and Later. The Earlier Stage is dominated by power and competition, akin to "Might make Right." At the same time, the Later Stage focuses on truth, equality, and cooperation, summarized as "No one is above the Law." Maybe a framework to understand why different groups admire certain leaders. In the US, about 40% of people are in the earlier stage and 60% in the later stage; globally, it's reversed, according to Kessler.

6. AI apps I currently use a lot: Perplexity for quick research. Claude.ai for writing. ChatGpt (app) for quick idea drafting on the go (using voice mode). Descript for fast podcast editing. Sanctuary for personalized meditation. Otter.ai for meeting notes

7. Scott Galloway argues that we are destroying young people’s future. It's thought-provoking, and he is always entertaining.

8. Other nerdy bookmarks: Try Pencil (make ads with AI) Connect your CustomGPT to Slack and Teams. BRAND.AI - An AI-based brand management solution started using it at BARK.

9. I spoke with a fantastic startup CMO who’s running a one-person marketing team and using custom AI 'GP teams' to handle everything from mundane tasks to specialized roles, making them collaborative and giving them personalities to boost productivity and make interactions more natural. He also give them proformance reviews… Its a fun (and most shared). “How I used GPT to 3x my output as a marketeer”.

10. This is my second newsletter. I would love to hear what works and what doesn’t. Please drop me a line. You can reply to this, start a chat on the Substack app, or catch me on Linkedin.

Hope you find any of the inklings useful. If you did, kindly share the newsletter. Good vibes.

You are getting this because you signed up for either Henrik Werdelin’s or prehype’s newsletter. Every once in a while, I share what I call my inklings; it’s random things that I find interesting. I hope you like it. I will send them every time I have ten of them.

I also publish a podcast about AI, make dogs and their people happy and try to help 10.000 entrepreneurs launch a company.

1. I recently had the privilege of speaking with Kevin Kelly, co-founder of WIRED and renowned technology writer. Here are some of his main points from our chat (podcast):

   1. AI will accelerate all other technologies and touch every aspect of our lives, much like artificial power transformed the world during the Industrial Revolution. Just as companies had to restructure to accommodate electricity, organizations must adapt to AI by changing their structures and processes. The company's organizational architecture of tomorrow might look more like special forces than an army.

   2. We should consider AI LLM models as a diverse "zoo" of intelligence, each with unique characteristics and capabilities. Some AI systems may have vast memory but short attention spans, while others might possess glimmers of consciousness or excel at specific tasks like strawberry picking.

   3. Kevin recommends doing something "useless" every day to stay creative and inspired**.** This could involve exploring new hobbies, tinkering with ideas, or pursuing interesting projects without a specific objective. By giving ourselves permission to do something "useless," we can break free from the tyranny of productivity and discover new paths to innovation.

   4. “Aim to be effective, but unpredictable. That is, you want to act in a way that AIs have trouble modeling or imitating. That makes you irreplaceable.” From his new list of *Excellent Advice for Living*

2. Spending a bunch of time thinking about conversational UIs and the impact of chatbots replacing websites. Wrote about branding in a chatbot world.

3. We are also experimenting with a new type of AI-powered, conversational websites that allow for the fast launch of new companies. You can try a few here: Our AI Accelerator: Audos - ForTheFamily (advice for parents of fluid kids).

4. Three good newsletters about AI: Ben’s Bites, Ai-Ronman, Prompts Daily.

5. BARK Air (the new first-class plane business for dogs made by BARK) has been an incredible initiative to work on - and a lesson on how powerful a quality launch video can be. I believe it dramatically amplified the earned media we got from the launch. (10 billion impressions at last count)

6. This newsletter format was inspired by how Michael Karnjanaprakorn has structured his inspiring substack. I also enjoyed chatting with him about frameworks for living your life as an entrepreneur.

7. After reading “**The Anxious Generation,”** we have made new technology rules for our two sons: Limited screen time, no smartphone before 14 years old, and no social media before 16. Instead, I have made cheap Android “dumb” phone using this b/w interface, this service for blocking apps, and this app for getting notifications when they arrive and leave school.

8. I am a big fan of the core concept in Kenneth Stanley’s book Why Greatness Cannot Be Planned: The Myth of the Objective. While Kenneth's book is mainly about AI, I think the thinking is very relevant for entrepreneurs. I wrote about it here.

9. My father-in-law gave me a book about The Economist, including their writing style sheet. It works well as a prompt when you get ChatGPT / Claude to write for you.

10. Here are some good links: An excellent AI-powered podcast app, an album good for thinking, and software to run AI locally for sensitive material (or long flights).

If you have any ideas, suggestions, or feedback, hit the reply button. Connect on LinkedIn, listen to my podcast, or subscribe to my newsletter. I hope you find any of the inklings useful. If you did, kindly share the newsletter.