How I built developer tools after leaving Google: 18 months in

The first six months were brutal in a way that surprised me. At Google a hard problem was "design a system that scales to a billion requests." Outside Google a hard problem is "the Stripe webhook stopped firing and I don't know why and there is nobody to ask."

I underestimated identity loss. I'd been a staff engineer at Google. People had asked my opinion in design reviews. Three months into Pelt I was emailing strangers asking them to please try our product, and most of them ignored me. My partner once gently asked me, after I'd snapped at her over dinner, whether I was sure I shouldn't go back.

§The recruiter email

The scariest moment was January 2025. We had nine paying users, my savings were down to about four months, and a recruiter from Anthropic emailed about a staff role with a comp number that would have netted more in two years than Pelt was likely to net in five.

I sat with that email for a week. I didn't reply. The Greenmark founder's framing — that you optimize for survival of one company, not portfolio outcomes — became the line I kept repeating to myself. Han, when I told her about the email, said one thing: "If you take it I'll respect that, but I left Google because I was tired of building tools that twelve people would use. So have you."

The inflection was a tweet. In June 2025 a fairly well-followed engineer at a Series B fintech tweeted that Pelt had cut their onboarding-for-new-hire time from "three days to forty minutes." The tweet did 400k impressions on X.

In the seven days after that tweet we did $48k in new ARR — more than the previous five months combined. I learned, painfully, that we hadn't built our infrastructure to handle the load. Cloudflare Workers were fine. The Hetzner provisioning queue was not. We had a four-hour outage on a Tuesday morning that I will never forget.

What saved us was that we'd been brutally honest in the status page about exactly what was wrong and what we were doing to fix it. Nobody churned. Two of the customers who'd hit the outage actually upgraded plans afterwards because, as one of them told me, "you sounded like an engineer, not a comms team." The same week I read a thoughtful post on what status pages actually communicate which crystallized what we'd done: a status page is the most honest sales artifact a SaaS company produces.

That week taught me the rest of the year's strategy: ship the fix, narrate the fix, never lie about the fix.

§The eighteen-month inflection

Six weeks after the Datadog integration landed, MRR jumped from $24k to $61k. The mechanic was simple: developers who installed Pelt's CLI saw their first useful trace in under ninety seconds, versus the OpenTelemetry Collector plus Jaeger plus Grafana Tempo stack that most of them had been wrestling with for a quarter. Time-to-first-value collapsed and so did our trial-to-paid drop-off — from 71% abandoning at the auth step to 38%.

I made my first hire that month: a developer advocate named Yuki who'd worked on the Vercel edge runtime team. She didn't build features. She rewrote our docs in a weekend and recorded forty-seven YouTube walkthroughs over the next quarter. Our top-of-funnel from organic search tripled. By month fifteen we crossed $140k MRR with two engineers and Yuki. I'd taken a salary for the first time since leaving Google — $8,000 a month — and stopped checking my old vesting schedule.

The other quiet shift: I closed the Discord server and moved community to a small Linear public roadmap. The signal got noticeably better.

1. 1
   Ex-Big-Tech is not a moat.

   The credential gets you the first ten conversations. After that, only the product matters. Customers don't care where you used to work; they care whether their build is faster today.

2. 2
   The Stripe webhook problem is the actual job.

   Big Tech protects you from the long tail of integration weirdness. Plan for the integration weirdness to be 60% of your year-one engineering hours.

3. 3
   Hetzner over hyperscalers, until you can't.

   We saved enough on infrastructure to fund three months of runway. Most devtool startups overspend on cloud out of habit.

4. 4
   Status pages are sales pages.

   How you communicate during an outage is the most honest signal a buyer ever gets about you. Be specific, technical, and unembarrassed.

5. 5
   The recruiter email will come.

   Decide in advance what you'll do with it. The minute you start negotiating with yourself in the moment, you've already half-quit.

What this looked like in practice.

The thing nobody tells you about *leaving big tech* is that you lose the cultural shorthand. At Google, "RPC" meant Stubby. At Pelt, "RPC" means whatever the customer's stack settled on this quarter. Re-learning vocabulary slowed me down for nine months.

The thing nobody tells you about *time-to-first-value* is that ninety seconds is a different product than nine minutes. They look similar in a demo and feel completely different in production. We instrumented every step of onboarding with PostHog and re-ran the funnel every Friday for a year. Most fixes were one-line copy changes.

The thing nobody tells you about *hiring a DevRel before a fifth engineer* is that documentation is a lead generator at a quality far beyond ads. Yuki's first walkthrough video drove more demos than our entire prior quarter of paid acquisition combined. I see a similar pattern in the open-source database story from the same cohort — distribution comes from craft, not spend.

The thing nobody tells you about *closing the Discord* is that the loud users are not the paying users. Our highest-LTV accounts never posted. Once we shut the server they emailed us instead, and the email volume was lower and higher quality.

The thing nobody tells you about *taking a salary again* is that it changes the kind of risks you are willing to take. I shipped fewer reckless features after I started paying rent from the company.