Don't blame the agent: match the environment to the task, then instrument outcomes and infra failures
bibryam · x · 2026-10-15
bibryam argues a core principle of agent engineering: don't hand an agent a bare machine and blame the agent when it fails. Instead, match the environment to the task, then instrument task outcomes and infrastructure failures separately so you can tell agent mistakes from environment faults.
The attached slide draws a key distinction — job vs. session:
- A CI/CD-style job is a script that runs once, against immutable input, with a bounded result.
- A devbox-style session is an open-ended process that gets inspected, resumed, and redirected over hours or days — and the session must outlive the box.
For anyone building agent infrastructure, it's a crisp framework: classify the task first, then choose the runtime and observability it needs.
More from coding & agent
- Dev builds Mumbai 3D manga background generator with Claude Opus: 25 landmarks, 4 styles — CurieuxExplorer · 2026-10-15
- Santiago AI meetup: LATAM Airlines shares prod architecture, LangChain demos DeepAgents — leslysandra · 2026-10-15
- Why most companies shouldn't be using AI agents yet — and the millions wasted daily on hype — DavidLinthicum · 2026-10-15
- IAPS report: honeypots can catch rogue AI agents before they do real harm — vkrakovna · 2026-10-15
- Dev builds wildfire analysis tool with Claude Opus to track fires and drones — nikola_mr64990 · 2026-10-15
- Solo dev shipped a product in 10 weeks: Claude designed, Codex wrote the backend — theteknosaur · 2026-10-15