I Built an AI Agent That Does X — Here's What Actually Broke

说明与用途

AI 代理失去短期记忆、在超时时陷入无限循环，并调用不存在的函数。解决方案包括分离上下文与记忆、限制重试、验证工具调用，并在必要时使用确定性脚本。

问题

The AI agent repeatedly forgot decisions, entered infinite loops on tool failures, called non‑existent methods, and was less autonomous than expected.

结果

The agent became reliable, avoided infinite loops, correctly handled missing tools, and operated within clear autonomy boundaries.

工作流

• 1. Separate short‑term context from long‑term memory store.
• 2. Add hard retry limits and exponential backoff for tool calls.
• 3. Validate every tool call against a real schema before execution.
• 4. Replace unnecessary agent logic with deterministic triggers and scripts.
• 5. Monitor agent behavior and adjust guardrails as needed.

证据

A few turns in, the agent would make a decision, act on it, and then reconsider the same thing later as if it had never happened.

来源

One of the agent's tools called an internal service that timed out. Instead of surfacing that as an error, my retry logic and the agent's own instinct to "try again" stacked on top of each other and turned into something close to an infinite loop.

来源

The agent confidently called an internal function that was never real.

来源