The spec said 1.5 seconds
I wrote a 1.5 second latency target into a spec and reasoned from it for weeks. Measured, the real figure was 60 to 110 seconds.
Architecture
10 notes filed here
State you assumed would still be there
An auth token held in memory and a SQLite file on disk both worked locally and disappeared in production, because my laptop provides durability the serverless platform doesn't.
The model that wouldn't take instructions
Chief's daily brief runs on a small local model. Getting it to behave meant moving a bullet-count limit out of the prompt and into code, then measuring what a repeat penalty setting actually cost in accuracy.
Four boundaries an AI assistant's guardrails have to cover
Validating user input is the most familiar guardrail and the smallest part of the problem. The model's own output, what it hands to tools, and what gets rendered afterwards all need the same fail-closed treatment.
What a promotion actually asserts
For an AI feature with guardrails, releasing it carefully requires two disciplines: coordinating multi-service promotions with evidence at each step, and keeping the validation environment faithful enough that human feedback can be trusted.
What AI features force you to specify
When you add an AI layer to a retrieval feature, data decisions that developers used to make by instinct become explicit contract requirements. Live demos are often the first place you find out which ones are missing.