Patterns cookbook

Production patterns for AI agents

The same problems show up in every real agent app. Retries that do not double-charge users, circuit breakers, parallel research, cheap routers, hard human gates. Each pattern below is a self-contained YAML recipe with the trade-offs spelled out.

Reliability

reliability

Retry with backoff

Soak transient failures with exponential backoff before they reach the user.

reliability

Circuit breaker

Stop hammering a failing service and route to a graceful fallback.

Performance

performance

Fan out, join

Spawn N specialists in parallel, wait for all, fold the results.

performance

Semantic router

Pick the cheapest specialist that can answer, instead of one big model.

performance

Summarise and feed back

Compress past turns when the context window starts to bite.

Safety

safety

Human in the loop

Pause for explicit approval before any irreversible action.

Cost

cost

Rate limit with fallback

Cap the agent's external calls and degrade gracefully when the cap fires.

Data

data

Audit everything

Every tool call, every credential read, hash-chained for replay.

Newsletter

Get the next post in your inbox.

Engineering notes from the Digitorn team. No marketing, no launch announcements, no "10 prompts that will change your life". Just the things we write that we'd want to read.