Field Notes
Practical writing for product leaders and boutique advisors – on deciding what to build, answering the hard questions, and running a sharper practice.
Your Features Shipped. Your Metrics Didn't Move. The Gap Has a Name.
Your team ships on schedule and the numbers stay flat. That's the build trap — and cheap AI building made it easier to fall into and costlier to ignore.
You Don't Choose a Direction Once
You can't set a heading once and walk — the trees pull you off course. Product direction is a bet you re-make, not a decision you make once.
The Highest-Value Answer in Market Validation Is 'No'
Validation isn't a green light. For one client, its most valuable output was a no — before a dollar went into a market that wasn't there.
When Your Research Tool Tells You What You Want to Hear
An AI research tool handed me a confident number that was exactly what I wanted to hear — and wrong. The verification gap is the step cheap answers skip.
I Ran Three Research Projects Through Tavily and Exa. Exa Won All Three.
I ran the same deep-research plan through Tavily and Exa three times, blind-judged and single-variable. Exa won all three — cheaper, faster, more complete.
An 80% Team Beats a 95% Team
Ship at 80% and a learning team compounds past a team that polishes to 95% – because shipping is the loop, and the cost of over-polishing is invisible.
Research Agents Get Less Accurate the More They Read: 42% Worse From 2 Tool Calls to 150
Deep research agents keep link validity above 92% at every depth while fact-check accuracy falls 42%, so here's the order to verify a report in.
You've Researched Enough When Three People Who Live the Problem Agree It's Real
Analysis paralysis isn't cured by more confidence. The stopping rule: when three or four people who live the problem agree your hypothesis is real.
Your Metric Isn't a Target. It's a Hypothesis You Forgot to Test
Features ship and the metric stays flat because teams treat the metric as a target to hit, not the hypothesis it was built to test.