Profiling with Cursor 2.0: The Missing Layer in AI Code Generation
Performance is the one thing AI can't fake. How continuous profiling is becoming as essential as syntax highlighting for validating AI-generated code.
I write about building products, making AI actually useful, and the weird stuff I learn along the way. Currently building AI products at Grafana Labs. Previously built Pyroscope and sold it to Grafana. I like making a difficult thing fit within even tighter constraints, shipping terrible first versions, and figuring out how things actually work instead of how they're supposed to work.

The best way to learn something is to build it badly, then fix it. I write about the mistakes, the weird hacks that actually worked, and big ideas that make me sound a lil crazy.
Accepting that the first version will be the worst removes failure anxiety and gets you started
Sometimes the best investment isn't in what you learn, but in who you learn with
Some upfront thinking saves hours later, but always bias toward action and iterating quickly
Stuff I've built, mistakes I've made, and things I've learned the hard way
Performance is the one thing AI can't fake. How continuous profiling is becoming as essential as syntax highlighting for validating AI-generated code.

How I solved the LLM consistency problem while grading 200+ sales pitches using round-robin tournaments. A unique approach to getting deterministic results from AI.
Built the same crossword app twice. Bad prompt took 6+ hours, good prompt took 40 minutes. Prompt engineering is actually a real skill.

The whole story of getting acquired. Careful early planning, negotiations in fancy vegas hotels, leverage dynamics, and why it was actually the smart choice (not giving up).

Game theory thinking saved us at every step. Valuation psychology, investor motivations, acquisition dynamics. Understanding what people actually want.

Building Pyroscope meant solving a hard computer science problem. How do you profile code continuously without killing performance?

Side projects became my secret weapon for learning and getting unstuck. The constraint of one weekend forces you to ship something.
Built an AR lens in 10 minutes just to see if I could. Turns out Snapchat Lens Studio is way more powerful than I expected.

Honest take on Hack Reactor after working in tech for two years. The real value wasn't what I expected.