About a month ago I was laid off from VideoAmp, four months into my first executive role. I’m ok enough to write about it now.
In 18 years, it was the first time I’d been let go. My first reaction was fear: I’m my family’s main provider. My second was doubt about whether I was prepared for where the industry is headed. Product owners on one side, end-to-end developers on the other, and no engineering management in between.
Looking back, every stop has been a piece of this: verifying ad performance at scale at Google, building products and developer tooling from scratch at Celonis, defining agentic workflows at VideoAmp. And underneath it all, a decade of side projects built to my own requirements in the evenings.
This month those side projects became the test cases, and the evenings became the day job. Bakeoff is a private eval that takes real bugs from my own codebases, from starculture to a job-search agent, with a Mint replacement next, and runs them through the latest OpenAI and Anthropic models in their own harnesses, plus GLM in pi. Every run is scored against live test suites. Three weeks in, it’s about 5,500 lines of design docs and about as much Go.
A teaser: the newest frontier models cost 80x the cheapest entrant for the same fixes, and took longer to deliver them. More results to come.
The last few years of AI have changed software engineering more than the cloud or mobile eras did. As in those eras, some tech organizations are outrunning their competitors by using a handful of techniques spread mostly by word of mouth. Which of them will work on your business’s tasks? Nobody has clean data, because public benchmarks don’t reflect your business and are already in the training set.
I’ll be reaching out to people in my network to hear about agentic pain points: where agents fail on your codebase, what you measure, what you wish you could. If you can spare 30 minutes, reply or message me.
Appendix
A few things that didn’t fit in the post.
The month also included my brother and me pulling a dog out of 45mph traffic on Lake Shore Drive. Ask me about it. The dog is fine and (thanks for asking) so am I.
The codebases feeding Bakeoff are recent side projects: starculture and warmpath, a job-search agent that is, for obvious reasons, getting a workout right now. Next up is Minty, the budgeting app I built after Mint shut down.
To everyone I worked with at Google, Celonis, and VideoAmp: thank you. Several of you talked me through this month, and I’d love to catch up with more of you, 30 minutes or otherwise.