Why I shut down my first startup on day 800 — and started over

The hardest part wasn't the wind-down of Company A. The wind-down was clean. The hard part was the six weeks after, when I had to decide whether I had a second startup in me or whether I was just trying to redeem the first one.

My partner — who'd lived through the entire 26 months of Company A — sat me down at our kitchen table on a Saturday and asked, very directly: "Are you starting another company because you have a real new idea, or because you can't stand having lost?" I didn't answer for almost a minute. The answer was both. Some percentage of the energy was redemption. Some percentage was actually about a problem I'd identified.

§The twenty conversations

I told her honestly that if I couldn't separate those, I shouldn't start. We agreed I'd take six weeks, talk to twenty platform engineers about drift detection specifically, and if at the end of those six weeks the new thesis still felt real and not borrowed, I'd start. She'd support it.

I did the twenty conversations. Sixteen of them were paying for drift detection in some form already. Four had built it internally and would've paid to delete their internal tool. The thesis held. I later applied the same twenty-conversation discipline before recommending it to the founder of Greenmark when she was deciding whether to deploy her seed round.

The first six months of Tarmac were also financially scary in a way Company A never was. I was self-funding. I'd already drained one set of savings on Company A's wind-down and tax tail. By month seven of Tarmac I had three months of personal runway left. Hilde joined the week before I would have had to take consulting work to keep the lights on.

The pricing model is what unlocked Tarmac. I'd priced Company A on seats, which forced procurement engagement and elongated sales. I priced Tarmac per AWS account. Every platform engineer manages somewhere between 5 and 200 AWS accounts. The metric was natural, the buyer was the user, and the price scaled with the customer's surface area without scaling with their headcount.

The second unlock was a public credibility move. In June 2025 I published a long technical post-mortem of Company A's failure on my own blog. It was brutally honest about specific mistakes I'd made: the wrong ICP, the wrong pricing model, the wrong cofounder split. The post got 28,000 reads in a week, partly because Hacker News kept it on the front page for fourteen hours. It became, weirdly, our best inbound channel for Tarmac. Platform engineers who'd read it would email saying: "If you're being this honest about your last company, I'm willing to try yours."

Three of our top-ten customers as of April 2026 came from that post.

§What the second company did differently

Tarmac v2 hit $400k ARR in seven months. v1 had taken twenty-eight months to hit half that. The mechanic was not skill — it was the absence of three specific mistakes I had memorized from v1. We charged from week one. We refused logo-grab discounts under €500/month. We shipped without a co-founder for the first nine months, which let me rebuild the product muscle I'd lost while managing a team of nineteen.

The first hire at v2 was a senior infrastructure engineer named Lior, who'd shipped the HashiCorp Boundary integration at his previous company. He joined as employee #1 with a 4% grant — half what my v1 co-founder had — and has out-shipped the four-engineer team at v1 by every metric I can measure. We moved from AWS to a hybrid Hetzner plus Cloudflare R2 stack in month four, which cut our infrastructure spend by 71% and bought us another twelve months of runway without raising. The discipline of treating near-death as a forcing function — instead of a story to tell investors — overlaps with what the GPU-spend account-saving story from this cohort calls "almost-churned as a positive cohort signal."

1. 1
   Wind down clean. Refund what you can. Be honest with your team early.

   Investors and engineers have long memories. The way you end one company determines whether anyone wants to back the next one.

2. 2
   The tool you wished existed is rarely the tool customers are paying for.

   Build the thing they already line-itemed in their budget, just better.

3. 3
   Pricing per resource beats pricing per seat in devops.

   Seats trigger procurement. Resources scale with usage. Use the metric your buyer can charge to a credit card.

4. 4
   Separate redemption energy from product conviction.

   They feel the same in your chest. They are not the same in the market. Take six weeks. Talk to twenty users.

5. 5
   A public post-mortem is the cheapest credibility you'll ever buy.

   Be specific about what you got wrong. Vague apologies don't compound; specific ones do.

What this looked like in practice.

The thing nobody tells you about *shutting down your first company* is that the grief is real and lasts about ninety days. Then it converts into an asymmetric advantage that compounds for the rest of your career. You stop being optimistic about ambiguous data. You start asking buyers for money on call one.

The thing nobody tells you about *charging from day one* is that it is not a pricing decision — it is a customer-quality filter. Free tiers attract advice-givers. Paid tiers attract operators.

The thing nobody tells you about *founding solo* is that the loneliness is overstated and the speed is understated. I shipped more in the first six months alone than in eighteen months with a co-founder, because there were no internal alignment meetings to run.

The thing nobody tells you about *moving off AWS* is that the savings are 30% of the story. The bigger 70% is regaining mental simplicity. Hetzner plus Cloudflare R2 plus Tailscale is comprehensible by one engineer; AWS at scale is not.

The thing nobody tells you about *second-time founding* is that your investors expect the same playbook as your first. They will be surprised when you charge earlier, hire later, and stay smaller. Pick investors who have already seen a successful second act — they will not blink.