bloom0
Cheat sheet

AI agent patterns

An agent is just an LLM in a loop with tools. These are the patterns real 2026 agent products are built from.

THINKACT (use a tool)OBSERVE the resultrepeat until done

Autonomy is a dial

Between "chatbot" and "fully autonomous" is a spectrum. Start every build at ask permission before risky actions, and earn autonomy with tests + guardrails.

Errors compound

If each step is 95% reliable, a 20-step task finishes correctly only ~36% of the time (0.9520). So: shorten chains, recover instead of cascading, and checkpoint progress.

The /loop pattern

Run a task on a schedule or until a goal is met — "keep improving X until it passes." Cap the steps so it can't run away.

The LLM Council

Ask several AI "seats" (e.g. a builder, a skeptic, a safety reviewer) the same question, then synthesize. Diverse views catch what one model misses.

Handoff contracts

When one agent delegates to another, write a contract: objective, output format, boundaries, budget, what to do when blocked. Vague delegation is the #1 failure.

Guardrails

  • Default-deny — allow only what's needed.
  • Spend caps + step limits.
  • Kill switch — a way to stop it now.
  • Never write memory from untrusted text.

Sniff test before you trust an agent

Does it observe real results (not just plan)? Is there a step cap? Does a failed step retry or re-plan instead of poisoning the run? Can you stop it and see what it did? If yes to all — you can let it off the leash a little.