Chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer; workflows-vs-agents and durable execution.
The decision that quietly determines your token bill, your p99, and whether anyone can debug the 2am page: did the LLM follow your code path, or write its own?
Five LEGO bricks — chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer — that snap together into every production LLM pipeline, with the cost and failure mode of each baked in.
Make a multi-step LLM pipeline that can be killed mid-flight, restarted hours later, and resume exactly where it left off — without double-charging a card or re-calling a $3 model.
The unglamorous machinery — constrained decoding, relevance classifiers, and input/output guards — that turns a pile of LLM calls into a workflow you can put on-call for.