What changed my mind.

Positions I held — then revised after desks, latency, and liability argued back.

What: Revised beliefs from shipping
Who: Builders updating priors with evidence
Why: Mind-change is a knowledge signal
Next: Decisions · Lessons · Playbook

Exact match beats clever fuzzy — for trust

I used to believe similarity ranking would feel “intelligent” and win desks.

I believe now exact-first FTS with normalized queries wins live calls. Cleverness that adds latency or ambiguity destroys trust faster than missing a fuzzy hit.

Decision · Lesson

AI isn’t the product

I used to believe shipping a model feature was the milestone.

I believe now the product is the workflow that survives Monday — review gates, provenance, merge rules. Models are components inside that workflow.

Playbook · AI review pattern

Workflow beats automation theater

I used to believe more automation always meant more value.

I believe now automating a broken or unclear step creates faster liability. Map the desk first; automate only where confirmation and recovery are designed.

Playbook · How I build

Simplicity scales further than clever stacks

I used to believe ambitious architecture signaled readiness for growth.

I believe now Cloudflare-first simplicity (Workers, D1, R2 when they fit) scales operations and understanding. Complexity should earn its keep against desk risk — not résumé optics.

Pattern · Decision