Tag7 posts
All posts tagged engineering-lessons.
My demo refund agent returned 200, claimed SUCCESS and refunded twice. How I catch that: tool twins that record every call, and faults that split the reply from the state.
My locked test set ran twice: one field hid 1,151 questions from the headline score. How I fixed the data, proved the model never moved, and published both runs.
RealPath's leakage claim is a pytest: delete everything after the anchor, rebuild, diff. What it caught when I broke the cutoff, and two leaks it still misses.
Before the fix, every solar-return prompt my astrology app could build carried an empty chart and still asked for poetry. How I moved the line between the ephemeris and the LLM into code, and where it isn't there yet.
Membrane AI hands every model stage a masked copy of the diff instead of trusting each one to redact. How it works, what degradation can hide, where it stops.
Four guardrails that turn a vibes-loop into something you'd actually run in production: budget envelope, retry curves, break conditions, and a real fallback chain.
Production lessons from running autonomous agents in long-running loops, fallback patterns that actually work, and the day your agent decides to retry 47 times.