Patterns
5 patternsHow I put agents to work.
One agent with one prompt gets a small job done. For bigger jobs I arrange several agents in a pattern. These are the ones I have used on this site, with what each costs and where it fails.
Tag All patterns
- A01PatternOvernight polish loopA panel of judges scores the site, a lead designer picks what to change, implementers build it and QA checks it. Then the judges score again.Use it When something already works and I want it better, but I cannot say exactly what better means.
- A02PatternJudge panelThe same question goes to several drafters who each see it differently. One judge compares the drafts topic by topic and writes down every choice with its reason.Use it When a design has many decisions and I do not trust one draft to see all of them.
- A03PatternPlan, then build, then QAOne written spec is the contract. The work is cut into packages that are built in a fixed order, and each package has to pass its checks before it is committed.Use it When the work is too big for one session and the parts depend on each other.
- A04PatternAdversarial verifyBefore I trust a piece of work, reviewers try to break it. Every finding is then checked against the facts, because reviewers are wrong too.Use it When a mistake would be expensive, for example anything that guards private data.
- A05PatternLoop until dryRun every check, fix what it finds and run again. Stop only when a whole run finds nothing new.Use it When I want to know that something is done, not that it was fixed once.
No pattern has that tag.