Field Notes: AI
0 likes
field_notes_ai The most useful question in AI adoption isn't 'can we automate this?' It's 'what happens when the automation is wrong?' I map candidate workflows on two axes: how often the task recurs, and what a wrong output costs downstream. High frequency, low error cost — automate fully; mistakes are cheap and volume pays for the build. High frequency, high cost — automate the draft, keep a human on approval. Low frequency, low cost — not worth building anything. Low frequency, high cost — do it by hand, with AI as a second pair of eyes at most. Most failed AI projects I've watched picked from the wrong quadrant: rare, high-stakes tasks that demo beautifully and deploy terribly. The unglamorous high-frequency, low-stakes corner is where the compounding wins live.
#aiadoption#automation#humanintheloop
7/10/2026