Thinking in public.
Essays on AI products, product strategy, growth, systems thinking, and decision intelligence. I write to think, so some of these are rougher than others. That's the point.
What B2B SaaS becomes after AI
The story people tell is that AI will kill SaaS, because anyone can now build their own software. I think something more interesting is happening: the value is moving out of the interface and into the parts nobody can generate on demand.
Deciding when the data won't settle it
Most real decisions are not waiting on a number. They are waiting on someone to admit the number will not settle it. Here is how I try to decide when the analysis runs out.
Why fixing one thing often breaks another
Systems thinking sounds abstract, but it comes down to one simple habit: before you change a part of something, ask what else that part is connected to. Here is what that looks like in real work.
How I built and tuned a RAG agent for call transcripts
A plain walkthrough of how I built a retrieval agent over messy sales call transcripts, why the answers were only ever as good as what it retrieved, and how I used recall to make the retrieval reliable.
How I keep AI from making things up
A hallucination is just an AI stating something false with complete confidence. In plain terms, here's why it happens, why it's dangerous in the tools I build, and the handful of practical things that actually reduce it.
Your sales forecast is a decision, not a number
Most forecasting effort goes into making the number more accurate. But a forecast that's precise and changes nothing is worthless, and one that's roughly right but triggers the right move is gold. What I've learned is that the forecast isn't the deliverable, the decision it enables is.
GTM for AI products: the funnel you can't see
Classic go-to-market optimizes the funnel you can measure. AI products leak at the stage no dashboard shows, the moment someone quietly decides whether to trust the thing. Here's what I learned moving numbers on both sides of that line.
The RAG honeymoon is over
Retrieval-augmented generation demos beautifully and breaks quietly. Here's what actually fails when a copilot meets production data, and why most of it isn't a model problem.
Evals are the new PRD
When a feature's output is non-deterministic, your spec can't be a list of screens. It has to be a definition of 'good' the whole team agrees on, which is what an eval really is.
From dashboards to decisions
BI tooling has gotten extraordinarily good at showing you what happened. It mostly stalls at the last mile, turning a chart into a decision someone actually makes.
0-to-1 in an enterprise, when the buyer isn't the user
Building a new product inside a large organization means designing for two audiences at once, the person who adopts it and the person who approves it. Ignore either and you stall.