Blog

Field notes from shipping product with AI

What I learned building production AI and running product teams. Specifics, not takes.

September 11, 20264 min read

The old apps still had owners when I killed the restart

I combined three healthcare apps so a patient could finish in one session. The tickets on the apps I was about to shut down did not stop. They looked like progress.

September 23, 20266 min read

AI Product Manager is a work process, not a personality

We have turned "AI PM" into a vibe: prompt-fluent, tool-hopping, always current on the leaderboard. That is not a job. An occupation has a work process, a standard for done, and a way to tell a junior from someone who can own the miss.

September 16, 20266 min read

Generation got cheap. Proof did not.

Everyone can produce the feature now. Almost nobody can prove it is still true after the agent touched it. That gap is the actual product job in 2026, and it is why most AI still does not show up in the P&L.

September 9, 20265 min read

They already had voice notes. They still typed.

I sat with therapists whose EHR already has voice SOAP. They do not use it. They jot during care and type the note later. The product they wanted was not a smarter microphone. It was a cleaner pass over a seed they already wrote.

September 2, 20265 min read

I still write production code. That is a leadership choice.

I sit in the codebase on purpose. It is how I keep judgment current, and how I kill AI features that would have survived a deck.

September 1, 20265 min read

Velocity is not a strategy

I have sat in reviews where the burndown was clean, the release notes were full, and revenue had not moved. The team was proud. The dashboard was not.

August 29, 20265 min read

Product leadership that scores capture will invent a dongle

Three apps lived because three teams did. The member started over each time. Unifying them did not make anyone’s surface look bigger. It made one job finish.

August 26, 20266 min read

An agent without an owner is a product with no PM

Every agent in front of a real person makes decisions. Booking, withholding, handing off. Those are product decisions. Most teams treat them as a model problem, or nobody's problem.

July 25, 20265 min read

Designing AI that users actually trust

Normal software makes a promise: the same input gives the same output. AI breaks that promise, and it changes what design has to do. Here is what built trust in a Voice AI that answers real calls, and what quietly destroyed it.

Product Management

Trust

Read it

July 23, 20266 min read

Wrappers die in production: the case for orchestration

A demo is one model call. Production is a system. The wrapper does not survive contact with real users. Here is what orchestration actually takes, and the metrics that decide whether it works.

July 18, 20265 min read

The human moat: what product managers still own in the AI era

AI already took part of the product manager's job, and I am glad it did. It took the busywork. It did not touch the part that was always the actual job: judgment, taste, and choosing the problem worth solving.

June 30, 20266 min read

What actually mattered putting Voice AI into production

I spent two years building Voice AI that answers real phone calls for real businesses. The model was never the hard part. Here is what actually decided whether it worked in production.