“The team is more important than the idea” might be the most repeated advice in startups. YC has been handing out some version of it for twenty years: find a co-founder you genuinely like, come up with an idea together, then pivot until you hit founder-market fit. Don’t fall in love with the idea. Fall in love with the team and the process.

I’ve built two companies on that advice. It worked both times.

This past year, I ran the same playbook a third time. And it didn’t work. Not because we were lazy or unlucky — we looked at more markets than I can count, ran dozens of ideas through validation, partnered with people I respect, spun up real websites, and got several of those ideas to actual revenue. By most measures of effort and execution, the year was a success.

By the measure that matters — did we form a company worth a decade of my life — it wasn’t. And figuring out why has changed how I think about the very beginning of companies. The formation stage. The part the playbook skips.

The tour

Let me give you the honest itinerary.

We started with Tenki, an AI forecasting product. This one, at least, began with genuine passion — I studied economic and political forecasting in college, and I’ve been fascinated by the question of who’s actually right about the future ever since. We aimed at prediction markets first, and for good reason: demand was high and the markets themselves were growing fast. The forecasts were good, too. But here’s the brutal math of that business — it’s not enough to be right about the outcome. You have to be right and the market has to be wrong, at the same time, repeatedly. Beating a liquid market at pricing the future is one of the hardest games there is.

So we followed the logic. If liquid markets are efficient, apply the same forecasting technique to an illiquid one — a market where decisions take longer and prices stay wrong. That reasoning led us straight to patents, and to Trarian: understanding which patents actually matter and what they’re worth. The technology was real and the conversations with buyers were real. But an illiquid market cuts both ways. The same buyers who take a long time to decide are, it turns out, in no particular hurry to buy anything at all.

From there we widened the aperture to the broader legal services market — statutory mail, document automation, a handful of adjacent ideas that each got far enough to have a domain name, a demo, and in some cases paying users. Which is where I noticed something uncomfortable: I was back at square one. Legal services is where my last company lived. A year of rigorous, logical steps had walked me in a circle.

Here’s the thing I want to be clear about: none of these were bad ideas. Some of them are decent businesses right now, for somebody. They just kept failing the same two tests. They weren’t venture scale — the math didn’t pencil to a big outcome no matter how generously I squinted. An idea can be good and still be wrong for you and your team. It took me an embarrassing number of repetitions to internalize that.

The playbook that built two companies

I wasn’t improvising this year. I was running the exact process that worked before.

At ZestyAI, we weren’t insurance people. At Paxton, we weren’t lawyers. In both cases we followed the classic advice: assemble a team you trust, pick a direction, and pivot until the market starts pulling. And in both cases, it worked. We stumbled into positions where the timing was right, an incumbent was napping, and customers wanted what we had. Two companies, two industries where we had no domain expertise on day one, two outcomes I’m proud of.

So the lesson I carried forward seemed obvious: the process is the asset. Team first, idea second, pivot relentlessly. If it worked in insurance and in law, it should work anywhere.

I now think that lesson was incomplete in a way I couldn’t see from the inside.

Filters aren’t generators

Along the way this year, I built myself a serious set of frameworks. I could tell you what a venture-scale outcome requires: a large market or one growing fast enough to become large. Hated, lazy incumbents. A wedge with real pull — customers behaving like it would be weird not to buy. Measured demand, not imagined demand. A data or distribution advantage that compounds. Math that pencils from day one.

I still believe every one of those. They’re good frameworks. If you’re evaluating an opportunity, use them.

These frameworks aren’t wrong. This is how investors validate markets, and they validate markets this way because it works. But here’s what I finally noticed: they apply to companies that are already getting started. Every single one of them is a filter — it takes an idea as input and returns a verdict. Keep or kill. Extraordinarily useful the moment you have something to evaluate, which means they describe the stage that comes immediately after formation. The refinement stage.

None of them generate anything. None of them tell you where to stand on day zero, staring at a blank page. I had built a world-class filtration system and kept pouring lower and lower conviction ideas through it, wondering why nothing came out the other end.

And to be clear — I’m not saying don’t pivot. You absolutely should modify the idea as you go; that never stops. But “pivot until you find founder-market fit” assumes you’re already moving — you have a company, momentum, customers yelling at you, a market shoving you around. Pivoting is steering. Steering advice is useless when the car is parked.

That’s what a cold start actually is. A parked car. At Zesty and Paxton, by the time we were “pivoting,” we were already inside something — there was payroll, there were investors, there were users reacting to what we shipped. The market could talk back to us. This year I tried to pivot from a standing start, over and over, and the market can’t talk back to a hypothesis with a landing page.

A hammer in search of a nail

Somewhere around the fourth or fifth market, I noticed the feeling. I had a great team-shaped thing. I had a great framework-shaped thing. And I was wandering the world looking for an idea-shaped hole to fit them into.

A hammer searching for a nail.

The tell was in how the partnerships worked, too. When you assemble a team around a search rather than a mission, the team is only as durable as the current candidate idea. We’d run hard at prediction markets, and when the market wasn’t there, the team configuration stopped making sense. Run hard at patents — same thing. Nobody did anything wrong. These are people I’d work with again and still count as friends. But a team formed to find something has no reason to exist once the current something dies. A team formed around a mission survives the death of any particular idea, because the idea was never the point.

That’s the inversion I got backwards all year. I treated “team first, idea second” as a formation strategy. It isn’t. It’s a description of what matters once you’re already formed and moving.

Start with the work

So here’s where I’ve landed, and I’ll admit it sounds almost stupidly simple after twelve months of sophisticated flailing.

You have to start with the work itself. The problem you’d be pulling on anyway, paid or not. The thing where doing the work is the reward, and a company is just the vehicle that lets you do more of it. Then — and only then — you build the team around that work, and the company around that team. Mission, then team, then company. The frameworks come out of the drawer after that, to keep you honest about whether the mission has a venture-scale expression or a lifestyle-business expression or no business expression at all.

I had the order reversed. Team, then search, then hopefully a mission emerges. It never emerged, because missions don’t emerge from search. Search is what you do when you don’t have one.

The frustrating part is that this matches my own history if I read it honestly. Zesty and Paxton look like triumphs of process in the retelling. In the living, they were much closer to: we were doing work we found genuinely interesting, in rooms where interesting things were happening, and we grabbed what the moment offered. The process story is what I told myself afterward. Survivorship narration.

I don’t think the past year was wasted. I learned what the frameworks are for and what they’re not for. I got sharper at killing ideas quickly, which is a skill worth its weight. And I found the edge of a playbook I’d trusted for a decade, which you can only find by running past it.

But I’m done searching for nails. The next chapter starts with a mission — the work I’d do regardless — and builds outward from there. I don’t know exactly what it is yet. Identifying it is the work now, and for the first time in a year, that feels like the right problem to be working on.

Thanks for being with me on this journey. I’ll keep you posted.

No essay this week. I’m taking a breather.

One of the things I keep relearning is that the weeks off matter as much as the weeks on. Stepping back is usually when the good ideas show up anyway – rarely at the desk, usually on a trail.

But I didn’t want to leave you empty-handed. Here’s a video I genuinely enjoyed: Mark Manson’s “I’m 41, if You’re In Your 30s, Watch This…”

It’s Manson looking back at his 30s from the other side – what turned out to matter, what didn’t, and what he’d tell his younger self. I’m in the middle of my own reset right now, thinking hard about what the next decade should look like. So the timing hit. Maybe it will for you too.

When you build something in Claude and hit “publish,” it becomes a public URL. And public URLs get indexed by Google. Which means there’s a record of everything anyone has ever chosen to share from inside Claude, sitting in plain sight — every quiz app, every strategy memo, every tiny CRM someone duct-taped together for their corner store.

Read more

Bend, OR, where I live, spent last week inside a smoke plume. Air quality over 200 downtown — the level where health agencies tell everyone, not just sensitive groups, to stay in. Mirror Pond gone gray. Crews chasing dozens of lightning starts east of town. Thirteen smoke seasons on the West Coast now, and every one of them I end up asking the same dumb…

Read more

Note: A quick ask before the article. I’m currently digging into two opportunities: one in trademark filing and brand monitoring, and another in AI-powered regulatory and policy tracking. If you’ve spent time in either world or even as part of running your business — filed trademarks, fought counterfeiters, tracked legislation for a living, anything clo…

Read more

First, a favor.

If you’re in the US and what I describe below sounds interesting to you – even a little – please reply to this email or DM me. Tell me your use case. What documents are you generating, what deal or hire or handshake are they protecting, and why would something like this be useful to you? I’m exploring whether to build it, and ten honest r…

Read more

There’s an xkcd comic where someone asks for the salt at dinner. Twenty minutes later, they’re still waiting, because the person across the table is busy “developing a system to pass you arbitrary condiments.” It’ll save time in the long run, he insists.

Read more

The most recent YC batch ended on a warning, and I haven’t been able to shake it.

It came from Garry Tan. Run the post-mortem on every startup that ever died, he said, and the number one cause, by a wide margin, is that they didn’t build something people want. They built the wrong thing. Or they never really built anything at all. That’s the median start…

Read more

There’s a specific kind of tired that comes from selling something nobody is in a hurry to own. You write the email. You get on the call. People nod. They say nice things. “This is really cool.” “We’ll definitely need this at some point.” And then nothing happens, over and over, and you start to wonder if the problem is your deck, or your pricing, or yo…

Read more

A few years ago the hardest part of building a software company was building the software. That difficulty was the moat. You burned two years and a few million dollars getting a product to actually work, and the pain of that was the thing keeping the next ten people from doing exactly what you’d done.

That moat is draining fast.

Read more

We’re launching Trarian’s second product this week at IPBC Global in San Diego: an infringement search that tells you who is likely practicing a patent, shows you the evidence, and rates how strong that evidence is. If you’re at the conference, come find us. If you’re not, here’s what we built.

Give it a patent number. It comes back with a list of compan…

Read more

In 1995, Newsweek ran a piece by an astrophysicist explaining why the internet was mostly hype — no database would replace your newspaper, no CD-ROM would teach like a real teacher. It’s famous now for being wrong. But here’s the part everyone skips: he was also right. Most of the companies that made the web feel inevitable in 1999 were dead by 2002. Th…

Read more

Most founders build one deck. They swap the title slide, maybe soften a number or two, and then fire the same thing at a customer on Monday and an investor on Thursday. Then they wonder why both rooms went quiet.

I get the instinct. You spent three weeks on the thing. It’s beautiful. Of course you want to reuse it. But a customer and an investor are not …

Read more

I’ve been using Claude Design to build out my new company Trarian’s website, design system, wordmark, LinkedIn banner, and two pitch decks — one for investors, one for customers. Pretty much everything we show the outside world. It’s the research-preview product Anthropic Labs launched in April, and it’s been more useful than I expected.

In my past start…

Read more

We’re excited to share Trarian’s first product: a 1–10 score that tells you how likely a US patent is to be invalidated. It’s trained on real court and patent-office decisions, it’s built for the people who buy, insure, and finance patents, and it does something that today’s most capable AI models on their own still can’t: it gives you the same answer e…

Read more

If you follow me on LinkedIn, you may have noticed a new title.

I’m now Co-founder and CTO of Trarian — a new venture I’m building with my friend Jordan Birnholtz. Trarian is the AI-native underwriting and pricing layer for the $1 trillion+ patent asset class. A FICO score, a Moody’s rating, an appraisal report for a market that hasn’t had one. We’re goi…

Read more

Two and a half years into the generative AI cycle, a pattern is starting to clear up. Claude and ChatGPT are emerging as platforms.

Not the way Windows is a platform. Not even quite the way iOS is a platform. The analogy in the title is imperfect and I’ll come back to that — Gemini is genuinely good and getting better, open weights models are real, and o…

Read more

Most AI founders are fighting over the same budget.

There’s a second one sitting right next to it that’s roughly six times bigger, and way fewer founders are fighting for it.

Julien Bek at Sequoia published a piece last month called Services: The New So…

Read more

People ask me all the time what I’m using.

Usually they want the one-tool answer. “Just tell me if I should use Cursor or Claude Code.” I get it. Decision fatigue is real, and AI tools are multiplying like JavaScript frameworks, new shiny thing every week, half of them gone in six months.

Read more

In 1881, Edison opened the first central power station on Pearl Street in Manhattan. Within a few years, you could buy an electric motor and bolt it onto your factory floor. And that’s exactly what manufacturers did – they ripped out the steam engine and dropped in an electric motor in its place, running the same system of shafts and belts that distribu…

Read more