About

I invest at the seed stage in AI developer tools and infrastructure. I back founders building the tools and platforms developers rely on. I founded Chef, the infrastructure automation company, and helped start the DevOps movement. At Amazon I built three connected practices as one body of work: modern Incident Management, and what we now call Site Reliability Engineering and Chaos Engineering.

What I do now

I have invested in and advised over sixty companies, including PagerDuty, Fastly, LaunchDarkly, Tailscale, Sanity, and Continue. I focus on the seed stage, and on companies across sectors where AI is becoming the operating system.

Most of what I do as an investor comes from being a founder myself. I have shipped v1 to skeptical buyers, grown communities from nothing, and fought through the years when the market had not caught up to the idea. I serve on the boards of Continue, Memgraph, Sanity, and Mobot. I help founders with what I know firsthand: open-source go-to-market, product positioning, and turning communities into superpowers.

What I look for

Founders with taste, solving a problem they know firsthand. The products I get excited about change how you think about a problem the first time you use them. You try it once and wonder how you ever worked without it.

I care about developer experience and open-source go-to-market. When founders do this well from the beginning, they compound. It must start authentically. It can’t be faked. At Chef, community became our superpower. Marketing spend does not replicate that.

How I got here

I was building ISPs in high school and working full-time hours on early wireless modems before most people were using the internet. I had an early IPO in 1999 and took a deliberate detour to become a firefighter and EMT.

When I joined Amazon in 2001, I became “Master of Disaster,” responsible for the availability of every property bearing the Amazon brand. That scope took me into most of Amazon’s teams and systems, helping engineers build and operate more reliably.

I built the GameDay practice, deliberately breaking production systems so teams could practice before real disasters hit. Adapting the Incident Command System I learned as a firefighter, I built three connected practices as one body of work: modern Incident Management, and what we now call Site Reliability Engineering and Chaos Engineering. The culture behind them became the foundation of the DevOps movement.

After Amazon, Steve Souders and I founded the O’Reilly Velocity Conference. It became the gathering point for people who were independently discovering the same thing: the wall between development and operations was the biggest bottleneck in shipping reliable software. That community became the DevOps movement.

Then Adam Jacob, Barry Steinglass, Nathan Haneysmith, and I cofounded Chef. We wanted to bring powerful infrastructure automation to the masses and build a core configuration utility for the internet. The custom tools built by Google, Amazon, and a few others were closely guarded secrets. We opened them up to everyone else. Chef let engineering teams define infrastructure as code, writing recipes to configure entire fleets of servers instead of managing them by hand. It was adopted by Facebook, Google, Apple, and IBM. We grew Chef to $75M in revenue and sold it a few years later. I am proud of what we built. Chef shipped to tens of thousands of organizations and reshaped how engineers thought about infrastructure.

I went on to cofound Orion Labs, a real-time voice AI platform for frontline teams who need information fast, hands-free, under pressure. The same kind of people I had served alongside as a firefighter.

Where this is going

The last twenty years changed the way we build and deploy software. The next ten years will change everything.

Every major abstraction in computing created a new generation of tools to manage the complexity it introduced. AI is no different. I invest in the companies building those tools. The ones that matter automate complex workflows, make AI reliable at scale, and ship to production.

The teams shipping AI agents to production are running into the same operational problems we ran into shipping web services twenty years ago. Agents need context, guardrails, observability, and the same developer experience standards we expect for humans. The systems that make that work look like infrastructure, not models. I back the companies building it.