12 Loop-Engineering Techniques — on the Real Model

Loop engineering is the shift from prompting an agent to designing the recurring cycle it runs unattended: inspect → act → verify → decide, with a real exit condition. Each card below binds a technique to a real use-case from one of the four harness domains — Healthcare, Insurance, Career, Drug Discovery — and runs it through the actual orchestrator (runAgent()) on the local WebLLM model you load above. The loop, tools, guardrails and verification are the production modules; the model is really doing the reasoning.

Load a model (⬇ above) to run on the real LLM, or tick Mock mode to run the real tools + guardrails with no GPU. Then click Run this loop on any card. Techniques are grouped by the same three layers as the harness demo and reuse its modules without modifying them.