Lessons from Eight Years Building Startups

I've been trying to figure out how to write this for a while. The short version is: I spent eight years building startups, most of them failed, one got acquired, and I'm still sorting through what all of it meant. The long version is messier.

Most founder stories you read online are the highlight reel. The raise, the launch, the exit. I want to tell the other version too, because that's the one I actually lived most of the time.

Shuffl

In 2017 I started building Shuffl.ai as a side project while working at Amazon. It was a platform that connected people across companies for conversations and networking. Nights and weekends. No grand plan. Just something I thought should exist.

It grew. Over time we hit a million connections across 3,000+ organizations. When the pandemic hit and everyone went remote, traffic spiked hard. That felt like a sign.

Leaving Amazon

I quit in early 2020, right as COVID was starting. I'd never left a stable paycheck before. I remember the math making sense (savings, traction, window of opportunity) but the feeling being completely different. Just this low-grade panic that wouldn't go away. What if it doesn't work. What if I can't go back.

I brought on a co-founder around the same time. We tried to raise venture funding and ran into issues I still can't talk about publicly. We never closed a round. The product had users but no revenue, and I had to sit with something I didn't want to accept: people using your product is not the same as people paying for it.

We never actually shut Shuffl down, which is its own weird thing. Stopped active development but kept it running. It's still going. People still use it. There's no clean ending to point to. It's just this thing that works but never became a business. I don't totally know what to do with that.

What Shuffl cost me

The business stuff is one thing. The personal cost is harder to talk about.

I stopped showing up for friends. I'd tell myself I'd reconnect when things stabilized, but "when things stabilize" is a target that keeps moving when you're running a startup that isn't making money. Family would ask how it was going. I'd give the short version. The real version (I might run out of money, I don't know what I'm doing, I'm scared) stayed inside.

My sense of self got wrapped up in the company. When Shuffl couldn't monetize, it didn't feel like the product was failing. It felt like I was failing. I know that's not how it works. But knowing it and feeling it are different things. I had stretches where getting out of bed was hard. I wasn't in crisis exactly, but I wasn't OK either. I should've been talking to someone. I wasn't.

Burned through most of what I'd saved at Amazon. Every month without revenue was another month closer to the edge. I thought about getting a regular job constantly. Some mix of pride and hope kept me going longer than it should have.

Sleep went bad. I'd lie awake running through every possible scenario. Exercise disappeared. I gained weight. The things that would've actually helped were the first things I dropped. I'd repeat that pattern a few more times before I learned.

The sabbatical

After Shuffl wound down I took five or six months off. Sabbatical feels like too fancy a word. I just stopped.

It was summer in Seattle. I sailed a lot. Took a road trip to Denver with my buddy and my dog. I went out on a boat into the Pacific Ocean to hunt for the SS Pacific, this steamship that sank in 1875 carrying what might be millions in gold. A friend had been searching for the wreck for decades and I tagged along on one of the fall 2021 expeditions. We spent days scanning the ocean floor with sonar off the Olympic Peninsula. They actually found promising images of the wreck site that season. Completely surreal experience.

I also reconnected with someone I'd met years earlier at Amazon. A GeekWire article about Shuffl popped up on LinkedIn, she reached out, and we started dating. She's now my wife.

I remember telling myself: I don't need money. I just want to enjoy life.

That's a funny thing to read back now. After everything that came next, my relationship with money is completely different. I've felt what it's like when every month is a countdown and I never want to be there again. The sabbatical me and the current me would disagree about a lot.

B612

Toward the end of the sabbatical and into 2022, I worked on something totally different. B612 was an immersive experience project called Rose. A physical space that used IoT hardware, sound, and light to help people unwind. I built the hardware and software integration, including a mobile app that controlled the whole environment.

Weird departure from everything else I'd done. Honestly it was good for my head. Not everything needs to be a venture-scale startup. Sometimes you just need to build something different and not think about TAM for a while.

Madrona Venture Labs

After B612 I joined Madrona Venture Labs as a Technical Expert in Residence. Got to see the venture side of things, how companies get funded, how investors think about markets and teams. It gave me structure and a paycheck, which I needed badly. And it made me want to build again, which I probably should've been more careful about.

co-founder.ai (fifty nos)

ChatGPT launched late 2022 and I couldn't stop thinking about it. I started co-founder.ai to help founders evaluate ideas and build startups using AI. This time I went solo. No co-founder. I wanted to own the whole thing.

Going solo was clarifying. Faster decisions, no alignment meetings. But it also showed me everything I'd been leaning on at Shuffl. My co-founder had carried more than I realized, parts of the product, parts of the business, a lot of the emotional load. Doing it alone meant sitting with all of that myself.

I say "solo" but my partner was deeply in it with me. We wrote the pitch deck together. She pressure-tested every version of the story, sat with me through every revision. She could see things from the outside in a way I'd completely lost. That mattered more than I can probably say.

Raised a small angel and pre-seed round, started building. Never got users. We pivoted something like 10 times. Built things, demoed them, got polite feedback, threw it out, started over. Nothing stuck.

The core problem, looking back, is that I was building in a domain I didn't deeply understand. I didn't have 10 years as an investor or a serial founder. I was guessing at what people needed. I was doing all the "right" things (customer interviews, experiments, iteration) in a space where I didn't have any real intuition.

And honestly? There's no clean tool that makes founding a company easier. It's too emotional, too personal, too tangled up in relationships and timing and gut feelings. The customers could tell. They were polite but they didn't convert.

I pitched over 50 investors for a real round. Every single one said no. Fifty nos. You keep rewriting the deck (the one you and your partner stayed up until 2am on) and walking into the next meeting telling yourself it'll be different. It's not different. That does something to you that's hard to describe.

The timing was wrong. The domain wasn't mine. The AI ecosystem was too immature: fragmented SDKs, no clear infrastructure, models changing every few months. Most people were happy with their $20 ChatGPT subscription. I shut it down.

The second time was different, and also the same

Shutting down a company is bad. Shutting it down for the second time is a specific flavor of bad. You know the drill, which makes it slightly easier and slightly worse at the same time. I had to tell investors, my partner, myself. Again.

My partner had heard the whole Shuffl saga. She wasn't there for it, but she knew what I'd been through. Now she was watching it happen in real time. She never flinched. Every time I wanted to quit she said: "This isn't your last startup."

I was more careful with myself the second time around. I'd learned that the cost compounds when you ignore it. I still had rough stretches. Still had nights wondering if I was cut out for any of this. But I had better boundaries. Talked to her more honestly. Paid attention to sleep and whether I was moving my body. Small stuff. It helped.

BAML (the detour that changed everything)

I'd been using BAML while building co-founder.ai. It's a DSL for structured LLM outputs, replaces raw prompt strings with typed contracts. The infrastructure problem pulled me in and I eventually joined the team.

Such a relief to not carry the whole thing for once. Small team, hard problem, and I could just build. No fundraising. No existential dread about runway.

It also gave me a front-row seat to what developers actually needed. Not better prompts. Observability. Debugging. A way to get traces out of their stack without getting locked into one vendor's dashboard. Everyone was duct-taping logs together or buying into platforms that didn't quite fit. Nobody had a single instrumentation layer that worked across SDKs and providers and fanned out to whatever backend you chose. That gap was where the Untrace idea came from. Segment for LLM apps.

The idea sat in my head for about six months before I did anything about it. I kept telling myself I didn't have the energy. I didn't. But the idea wouldn't leave me alone.

BAML was hard too, honestly. Still burnt out from co-founder.ai. Good work, but not rest. The accumulated weight of years of startups doesn't disappear because you join a team. It just gets quieter.

Then my wife got pregnant. And the math changed. A baby coming is maybe the worst time to start a company. It was also clearly now or never.

Untrace (the one where they said yes)

Untrace was the thing I'd wished existed the whole time. An open-source SDK that captures LLM traces and fans them out to any observability platform you want. One instrumentation layer. No lock-in.

Same idea as Segment's analytics.js. Instrument once, events fan out to Mixpanel, Amplitude, wherever. We applied that to LLMs. One SDK, doesn't care which provider or model, routes traces to Langfuse, Datadog, your own eval harness, wherever you want. Built on OpenTelemetry Gen AI semantic conventions. Privacy-first. Open source. No mandatory cloud.

I quit BAML with my wife three months from her due date. Terrifying doesn't really cover it. Still burnt out, not much money, body tired, confidence not great. I'd believed in ideas before and watched them all die. But I did it anyway. I'm still not sure it was a rational decision. More like a compulsion.

Built an MVP fast, took it to AI Tinkers (Seattle AI meetup), showed it off. Got real interest. Started talking to investors that same week.

And then something happened that had never happened to me before. They said yes.

At co-founder.ai I'd pitched 50+ investors. Every one said no. Had to beg for meetings. With Untrace it was the opposite. Five-minute conversations and they were in. We were raising $4M in under a week. I didn't even know what this felt like. When you have the right thing at the right time, everything is different. That contrast taught me more than any advice ever has.

I met Chris Clark through Aviel Ginzburg. Aviel was at Simply Measured, my first startup job before Amazon, where he was co-founder. After Simply Measured got acquired he went on to co-found Foundations, a Seattle startup workspace. When I showed him Untrace he said I had to meet Chris. Chris was building OpenRouter, the unified API for models. Untrace was unified observability. Same audience, same philosophy.

Taking the acquisition was one of the hardest decisions of my life. Things were actually working for once. Investors saying yes. Product had pull. Part of me wanted to keep going, raise the round, see how far it could go. But I'd done the startup grind twice already. The baby was coming. I even looked at YC. In the end, OpenRouter was right. Untrace became part of OpenRouter and I joined as Head of Engineering.

What the acquisition felt like

I want to mention this because the rest of the post is pretty heavy and I don't want to skip past the ending.

It felt like relief. Not celebration. I sat in my car after the deal closed and just sat there for a while. Didn't feel like I'd won anything. Felt like I'd survived.

My wife cried. Not sad. She'd been through co-founder.ai with me, through BAML, through Untrace. Every pivot, every rejection. Every time I doubted myself she said: "This isn't your last startup." She was right. I haven't thanked her enough.

From the outside the acquisition probably looks fast. It wasn't. It was the tail end of a path that went Shuffl to co-founder.ai to BAML to Untrace to a conversation with Chris. Years of dead ends and slowly getting better at knowing what to build.

On being a "founder"

Took me years to figure this out: being a founder is just a job. That's it.

Going on Stripe Atlas and registering a C corp doesn't mean you know how to lead, or hire, or fire, or make payroll. Most of the time, especially early on, you're just working on a side project full time. A person with a laptop and an idea. The title carries way more identity weight than it deserves.

I spent too long letting "founder" mean something about who I was. When I started treating it as what it is (a job: show up, do the work, be honest about what's working and what isn't) things got a lot clearer.

The LinkedIn thing

I felt like an imposter for most of this. Probably still do a little.

Hard not to when your feed is all success theater. Raises, launches, "humbled and grateful." Especially YC people. YC teaches you to be loud, own the narrative. Works for them. But when you're on the outside watching people who seem to have it figured out while you can't close a round, it does a number on you.

90% of startups fail. Even YC companies see about half fail long-term. The people posting are the survivors. The other 90% quietly moved on. I spent years comparing myself to the loudest voices in the room. Bad benchmark.

On competition

Used to be terrified of it. Someone else building something similar felt like proof the market was taken.

Wrong. Competition means the market is real. Customers exist, they're buying, there's budget. You don't need to be #1. Some people love Notion, some love Obsidian. Both massive. Find your slice.

What even is success

My answer has changed a bunch of times.

First it was money. Then traction. Then investors saying yes. All of those are traps if that's where you stop.

Where I'm at now: did you build something real, did it matter to someone, and are you still standing? That's the bar. The OpenRouter acquisition felt like success not because of the deal, but because the years of getting it wrong finally turned into something that lasted. I'm still here. Marriage intact. Health recovering. I have a kid.

I don't know if I have a cleaner answer than that.

Some things I think are true

I'm wary of the "lessons" format because it never really captures it. But here are some things I believe after eight years of this:

Traction without monetization is a warning sign. Usage is not revenue. I learned that one the hard way.

Timing matters as much as how good you are. Maybe more. I built good things at the wrong time and it didn't matter.

If you're pitching 50 investors and getting 50 nos, the problem probably isn't your pitch.

The company is not you. When it fails, you didn't fail. That takes a really long time to actually believe.

The cost is real. Your health, your relationships, your savings. It compounds when you ignore it.

Sometimes the right move is to quit, get a job, recover, and try again later. That's not failure. That's staying in the game.

What looks like an overnight outcome is usually years of dead ends and slowly learning. If you're in the middle of it right now and it feels heavy, that's because it is. You're not behind. You're just in it.