Four lessons from a year of running AI agents on boring tasks

Описание и применение

Опыт эксплуатации AI‑агентов в рутинных задачах показал, что ключ к надёжности – чёткие задания, аудит‑трейл, изоляция и автоматическая проверка результатов.

Задача

Operating AI agents reliably over time requires more than good prompts; it needs clear briefs, audit trails, isolation, and verification.

Результат

Stable, auditable AI agent operations with clear briefs, isolated environments, and automated verification.

Рабочий процесс

• Define a task file with goal, constraints, stop conditions, and output location.
• Run the agent using the brief; it reads its own ledger at start.
• Agent logs each action to an append‑only CSV/markdown ledger.
• After completion, run automated verification (curl, row count, diff).
• If verification fails, investigate using the ledger and git history.
• Isolate each agent with separate browser profile, tokens, and working directory.
• Clean up the sandbox after each run with a simple rm -rf command.
• Repeat weekly, adjusting briefs as needed.
• Maintain audit trail for human review and compliance.

Доказательства

Every reliable agent run in my setup starts from a task file: goal, constraints, stop conditions, and where to write results.

Источник

A plain append‑only ledger — CSV or markdown, one line per action — outperformed all of them.

Источник

Separate browser profile, separate tokens, separate working directory.

Источник

Every task in my system ends with an independent check: curl the page, count the rows, diff the output.

Источник