22 sessions on pytest with a coding agent: the agent failed at least once in 14, and 6 finished without a human stepping in.
Takeaway: Next time I'd open with the parallel-work caveat up front — telling it about the concurrent chart-engine branch early saved a lot of back-and-forth about which files were safe to touch.
Takeaway: Next time I'd flag upfront that the PR should stay scoped to bar charts only, so the agent doesn't fold waterfall-chart commits into the same squashed history it has to unpick later.
Takeaway: Next time I'd start by telling the agent up front that this tool is agent-only and used solely for fresh generation, not editing - that scoping cut a lot of wasted exploration.
Takeaway: I'd start by asking for the duplication cleanup and the pyright strictness fix in the same pass, since touching sizing/renderer/verification together was always going to surface type issues in cli.py too.
Takeaway: Next time I'd start by asking it to reproduce the bug locally with a runnable demo script before touching routes.py, since the first demo attempt crashed and wasted a cycle.
Takeaway: Next time I'd ask for the eligibility audit and the improvement pass in one go, since the audit alone confirmed we already qualified and the real value was in the polish work.
Takeaway: I'd start the next prompt by pointing straight at the audio component's event bindings since that's usually where Gradio state gets stale, rather than describing the repro steps alone.
Takeaway: Next time I'd just ask for the mode buttons themselves to change color from the start, since that's the solution I actually wanted — the badge approach was wasted work.
Takeaway: I'd start by asking for a real browser-based inspection (Playwright/DOM dump) from the very first CSS complaint instead of letting the agent guess at selectors screenshot by screenshot - that back-and-forth ate most of the session.
Takeaway: Next time I'd ask upfront for a deploy script instead of letting the agent hand-run git/HF CLI commands each time — that's what actually stabilized the workflow.
Takeaway: Next time I'd start by telling the agent explicitly not to commit anything without asking, since it committed twice before I stepped in to stop it.
Takeaway: Next time I'd start by asking for a ground-truth labeling workflow and benchmark harness before building any alignment logic, since that's what actually let us tell good methods from bad ones.
Takeaway: Next time I'd specify upfront that word clips must not bleed into adjacent words, since that was the real bug hiding behind 'alignment isn't better' — the fix was clamping word end to the next word's start plus much smaller padding for aligned clips versus the proportional fallback.
Takeaway: Next time I'd start the first prompt by stating explicitly that this is a kids' reading app needing realistic per-word audio with strict latency and reliability constraints, so we don't cycle through browser-TTS-vs-VoxCPM-vs-Modal three separate times.
Takeaway: Next time I'd give the exact test fixture (silent WAV) up front so the agent designs the endpoint to handle it from the start instead of hitting a gated-model load in production.
Takeaway: Next time I'd open with the real complaint upfront — that cluster labels were dominated by generic/common words like 'US' and were unusable — instead of first asking how they were chosen, since that led to a lot of exploration before we got to the actual fix.
Takeaway: Next time I'd open with both asks together — what goes into the embedding and why there's a 100-project cap — since the agent had to dig through the same files twice.