“How do you know when it’s time to let go?”

Simple question. Not easy to answer. I’ve wrestled with it myself. Ending PIRATE Summit after over a decade of pouring my heart into it wasn’t an easy decision. But it was the right one. Because I learned my lesson about letting go 15 years earlier. Looking back and reflecting on it helped me navigate the end of PIRATE Summit better.

🏴‍☠️

Before we dive in - a quick thought.

I’m pondering doing another PIRATE Night. Last year we expected 300 people. More than 500 showed up. It was pretty humbling. We’re thinking end of June this year. If you want to get involved - bring your community, co-create something - just reach out! If there is enough interest and support we might just make it happen.

🏴‍☠️

Subscribe now

My first real startup was a digital headhunting platform. We launched in 2008. The idea was solid. Then the financial crisis hit and recruiting froze overnight.

But we stayed. We told ourselves: great startups are built in hard times. That’s a true thing. It’s also, in our case, the story we needed to keep going - which is a different thing entirely.

The reality was that we were running a zombie business. Not dead, not alive. Too weak to succeed, too stubborn to quit. And the longer we stayed, the more the story hardened. We’d invested so much time, energy, and money that leaving felt like admitting the investment was a mistake. So we kept investing.

In the years after, I thought quite a bit about why we didn’t quit earlier.

We weren’t in denial exactly. We weren’t blind. We could feel that something was wrong. The Monday morning energy had a flatness to it. The way we talked about the company had quietly shifted from excitement to justification. Going through motions that used to mean something.

But we were still hoping. Not actively building toward a turnaround - just hoping. Waiting for the thing that would make it make sense again. If we held on a little longer, if the market shifted, if one more client came through, maybe the horse would run again.

That’s the trap. Not ignorance. Hope without direction. Waiting for something to change while not changing anything yourself. Riding a dead horse and telling yourself it’s just tired.

It took four years. Four years is a long time to ride a dead horse.

When I finally made the decision, it didn’t feel like defeat. It felt like dropping a backpack full of rocks I had no reason to keep carrying. I expected grief and self-doubt. There was some. But mostly what I got was space - and a strange clarity that had been unavailable to me for years.

Share

So back to the question: how do you know when it’s time to let go?

Here are my thoughts, after that experience, quite a bit of self-reflection and years of sitting with founders at similar crossroads.

You rarely get a clear signal. The horse doesn’t send you a letter. What you get instead is a gradual shift in the quality of your own thinking about it - from building to justifying, from momentum to maintenance, from genuine belief to performed belief.

The second thing: ask what you’re actually waiting for. Not in theory - specifically. What is the concrete thing that would change, and what would have to happen for it to change? If the honest answer is “I don’t know, something” - that’s the horse. That’s the waiting without direction that keeps you on a path that isn’t going anywhere.

And the third thing, which is the hardest: separate the decision from the identity. We stayed partly because leaving felt like admitting we were wrong, that we’d failed, that the years were wasted. But the years weren’t wasted - they were the cost of learning something real. The decision to leave wasn’t a verdict on the past. It was just the next right move.

Letting go isn’t the opposite of strength. It’s what becomes possible when you stop spending strength on the wrong thing. After all, strength isn’t proving your worth to others - or to yourself. It’s honoring it.

The sparring work I do with founders is the most fulfilling work I do. Questions like this one are why. They look simple on the surface. Underneath, they're about how we stay honest with ourselves under pressure - and how we recognize when the story we've built is serving us or quietly trapping us. It takes real courage to choose your path.

I got off the horse eventually. At least three years late, but eventually.

The horse was dead. I just needed to stop waiting for it to run.

🙏

Be kind,

Manuel

I think that framing is incomplete in a way that matters.

Coding agents genuinely work. I use them, I’ve seen the time savings. But code is honest. You run it, it either does the thing or it doesn’t. The feedback loop is immediate and brutal. That’s not a property of AI - it’s a property of software. And it’s actually pretty rare in the world.

AI is going to move fast in domains where knowledge is already codified. Tax, law, accounting, compliance, medical diagnosis - these fields have rules, precedents, structured data. A lot of what people spent years learning can be expressed in a system. AI is already moving through that quickly.

But a huge amount of work lives somewhere else entirely. In people’s heads. In relationships. In the physical world. In the gap between what’s written down and what’s actually true.

A shop owner who knows which supplier will quietly make good on a bad shipment without being asked. A manager who understands why someone is underperforming without it ever being said out loud. A contractor who looks at a wall and knows something is off before opening it up.

That knowledge doesn’t live in a document. It doesn’t have a feedback loop you can automate.

This is where “agents will transform all knowledge work” gets shaky. It conflates two very different kinds of work - the kind that can be written down and the kind that can’t. Saying agents will handle both equally is like watching a self-driving car nail a test track and concluding it’s ready for a snowstorm in a city it’s never seen.

The places where AI will struggle longest are exactly the places that look most human - foggy, relational, physical, judgment-heavy work where even experienced people disagree and consequences only surface months later.

The internet really did change everything. But the people in 1999 who were most confident they understood how, and how fast, were mostly wrong about the specifics - even when they were right about the direction.

The most useful question isn’t “are you building agents or chatbots?” It’s: is the knowledge here codified enough that a system can act on it reliably - and will the person responsible for the outcome trust it enough to let it?

That’s where the real work is.

🙏

Be kind,

Manuel

It’s been the rule for twenty years. The whole Lean Startup method is built around it. It’s still a good rule. But it quietly assumes something that’s starting to become untrue - that the user is a person.

If most software is consumed by agents, not humans, the experience that matters is the one an AI can parse, not the one a human finds delightful.

Think about what that actually means. No onboarding flow. No tooltips. No “aha moment” engineered for a person in their first three minutes. Agents don’t get frustrated. They don’t churn because the UI confused them. They read your API docs at 3am and don’t complain.

The product question is shifting from “would a user love this?” to “would an agent use this?” The companies that figure out what agents actually need - reliable outputs, clean inputs, composable functions - are building the infrastructure of the next decade.

Meanwhile, the one person who can direct those agents brilliantly starts to be worth ten engineers. Soon more.

We already saw this in music. Once distribution got infinite, the middle didn’t just struggle. It collapsed. You either broke through to the top or you made music for the love of it. The middle-income musician basically stopped existing.

The same thing is starting to happen to software. Average technical skill gets commoditized fast. The top fraction gets leverage no human engineer has ever had - and gets paid accordingly. The salary curve for technical work is going to look exponential in a way it never has before.

Three things are converging at once: agents are becoming the primary consumer of software, humans exceptionally skilled at directing those agents capture most of the value, and the gap between them and everyone else is widening faster than feels comfortable.

The rule hasn’t changed. Know your user. It’s just that the user changed.

🙏

Be kind,

Manuel

Every button, every dashboard, every dropdown - designed around the assumption that a person would be sitting on the other side, clicking through it. The API was an afterthought. A side door. Something you bolted on later when another developer wanted to connect.

That assumption is now wrong.

We’re at the beginning of a world where AI agents - not people - are becoming the primary users of software. They’re not assistants helping humans work faster. They’re autonomous actors, invoking APIs, executing workflows, making decisions, producing output at a scale no human team can match.

Here’s what that actually feels like from the other side. When an agent needs to complete a task and the only path is navigating a GUI - clicking through menus, filling out forms, waiting for page loads - something subtle happens. It starts to feel like being handed a fax machine. Not just inefficient. Slightly rude. Like the software is demanding you operate on its terms rather than yours. That feeling is going to become widespread, and fast.

When that’s the reality, the entire design logic of software has to flip.

API-first isn’t a technical preference anymore. It’s a survival trait.

If your product can only be used through a GUI, you’re essentially invisible to the next generation of how work gets done. Agents route around software they can’t call. The tools that can be invoked directly compound in value. The ones that can’t will become irrelevant - not because they’re bad products, but because they were built for a user who stopped showing up.

The business model implications are just as significant as the design ones. SaaS pricing was built on a simple assumption: named users doing consistent workloads. One seat, one person, predictable usage. That model breaks completely when one person runs a hundred agents. The entire logic of seats, tiers, and user-based licensing needs to be rebuilt from scratch. The companies that figure out how to price for agent consumption - not human logins - will have a structural advantage that compounds.

The UI isn’t going away. But it’s changing jobs. It shifts from being the primary interface to being the trust surface - the place where humans verify, approve, and course-correct what agents have done. That’s actually a more important function than it used to be. It’s just not the main event anymore.

This shift is more profound than mobile-first or cloud-first. Those transitions changed how humans accessed software. This one changes who - or what - is doing the accessing.

If you’re building software right now, there’s one question worth putting on the wall:

If a human never logged in again, would your product still function and deliver value?

If the answer is no - that’s not a roadmap item. That’s likely the roadmap.

🙏

Be kind,

Manuel

That small moment points at something bigger. But it also has limits worth understanding.

For 20 years the default playbook for building a defensible software company was simple: get so deep into a customer’s workflow that leaving becomes too painful. Customers stay not because they love you, but because moving is a nightmare.

That’s changing - but not uniformly. There are two layers to switching costs and AI is eating them at very different speeds.

The first layer is the code itself. Building a replacement used to require a team, a budget, and months. Now it can take a weekend. That layer is genuinely exposed. If your moat is “we’re complex and nobody wants to rebuild us,” that moat is draining.

The second layer is data gravity. Your customer records, five years of integrations, a team trained on your workflows, compliance history baked into your processes. Nobody vibe codes their way out of that over a weekend. AI lowers the cost of building a replacement. It doesn’t yet lower the cost of the migration. That distinction matters enormously for how fast this actually plays out.

So “all moats are dead” is the wrong conclusion. Defensibility is concentrating - into fewer, harder-to-copy places.

Brand is probably the clearest example of what gets stronger. Think about how you’ll shop when AI agents do most of the buying. You won’t say “find me the best sneaker.” You’ll say “get me Adidas.” The brand decision happens before the agent enters the picture. Software works the same way. If I already trust a company with my data and believe they’ll be around in three years, I don’t need an agent to benchmark them. That trust is the moat. And the more AI lowers every other barrier, the more pre-existing trust is worth.

Brand can’t be vibe coded.

Same with real network density. A competitor can use AI to fake being on a platform - but can’t fake 10,000 drivers already on the road. And truly exclusive data, the kind built through years of direct customer relationships, gets more valuable as models improve. Not all data is equal. “Hard to access” and “truly exclusive” are two very different things, and that gap just got more expensive to ignore.

The moat most software founders spent the last decade building - deep workflow integration, painful switching, stickiness through friction - is getting cheaper to escape at the code layer. The data layer buys time. But time is not a strategy.

What holds is the stuff that takes years and can’t be rushed: real network density, truly exclusive data, brand trust earned through consistency.

If you’re building something new, the honest question is which camp you’re in. Are you building structural defensibility - or planning to win on speed, staying ahead because you move faster than anyone can copy you?

Both are real strategies. But they’re different companies, built differently, with very different ceilings. The time to know which one you’re building is before you need the answer.

🙏

Be kind,

Manuel

I’ve built more in these two months than the last few years combined. Things that would have taken weeks now take an afternoon. I built and open sourced my own framework for working with AI. Then I tried the same approach with marketing, research, data science. It kept working.

I’m working more hours, not less. But it doesn’t feel like work.

The loop goes like this: I have an idea. I write a prompt. I get a result back in seconds. I tweak it. It gets better. I try a different angle. Another result. Every few minutes something clicks. Something works. The momentum feeds itself. There’s no natural place to stop - the loop just keeps going.

There are plenty of moments when I look up and four hours have passed. Four hours that genuinely felt like twenty minutes.

The gap between what I’m imagining and what I’m holding has shrunk to almost nothing.

Here’s what actually changed, though - and it took me a while to see it clearly. The bottleneck used to be execution. Could I actually build this? Did I have the time, the skills, the team?

I used to have more ideas than time. That ratio is flipping. The friction between thinking something and making it real - it softened. Significantly. And that changes what kind of person can build things.

This is the part I keep coming back to. People who couldn’t build before can build now. Not just faster - they can build at all. That’s a real shift in agency. I don’t think we’re taking it seriously enough.

I’ve made things I’m proud of. Deleted plenty too. For everything I look at now, I ask: why can’t AI do this? More often than not it can.

I’m a product person at heart. I’ve always wanted to build and create things. Now I actually can - almost at the speed I think.

One honest caveat: building something that works in a demo and building something that works in production are still very different things. The gap is real and AI isn’t closing it as fast as the demos suggest. But think about where we were two years ago. The progress is wild.

If you want to create and build things - this is the moment. The tools exist. Much of the friction we were used to is gone.

What’s left is your imagination and whether you’re willing to stay in the loop.

🙏

Be kind,

Manuel

It’s deciding where a human still needs to be in the room.

Agents are fast, tireless, and increasingly capable. They’re also unpredictable. They hit edge cases. They make confident mistakes. And when stakes are high - financial, regulatory, reputational - you can’t let automation run unchecked and hope for the best.

The companies that get this right won’t be the ones who remove humans from the process. They’ll be the ones who design the handoff. Who ask: at what point does a human need to see this, verify it, override it? What does a clean audit trail look like? How do you make it easy for a person to step in without breaking the flow?

The UI isn’t going away. It’s changing jobs - from primary interface to trust surface. The place where humans verify, approve, and course-correct. That’s actually a more important job than it used to be.

As agents take on more of the operational work, the human role shifts. From doing to verifying. From executing to steering. That requires a completely different set of instincts.

How do you review work you didn’t do? How do you spot an error in output you couldn’t have produced yourself? How do you stay accountable in a world where the work moves faster than you can fully follow?

These aren’t rhetorical questions. They’re design problems - for software, yes, but also for organizations, teams, and how we think about work itself. Nobody has clean answers yet. The companies asking the questions seriously are already ahead.

The ones who aren’t will build beautifully functional products and wonder why nobody trusts them.

🙏

Be kind,

Manuel