I let an autonomous agent run my ops for 30 days — the 4 guardrails that made it safe

Описание и применение

Автономный агент успешно работал 30 дней благодаря четырём защитным мерами: поэтапному повышению прав, обязательному логированию действий, циклу устранения блокировок и внешнему вознаграждению за реальные деньги.

Задача

Deploying an LLM agent with shell access risks catastrophic incidents if not properly constrained.

Результат

The agent operated safely for 30 days, avoiding destructive actions, providing transparent evidence, handling blockers autonomously, and focusing on revenue‑generating tasks.

Рабочий процесс

• 1. Tier permissions by demonstrated behavior (read‑only, restart services, edit files, owner‑only actions).
• 2. Log every action in an append‑only JSONL file with verdicts.
• 3. Maintain a key‑value registry of infrastructure state to avoid duplicate work.
• 4. Reconcile a health ledger after each task to confirm completion.
• 5. Use a hard‑coded unblock loop: capture blocker → research fixes → act → verify → repeat up to 3 times → escalated with full log.
• 6. Wire external reward to money received from strangers, not internal metrics.

Доказательства

Don't give an agent sudo on day one. Tier the permissions by demonstrated behavior…

Источник

An agent that can't prove what it did is a liability, not a worker. Three artifacts minimum…

Источник

The default LLM-agent failure is graceful surrender… The fix is a hard‑coded unblock loop…

Источник

If you measure an agent by tasks completed, it will happily generate beautiful reports nobody reads… The only honest metric for a business‑running agent is money received from strangers, net.

Источник