The engineer's field guide to optimizing coding agents.
A structured curriculum for getting more performance per dollar and per token out of a production coding agent: architecture, retrieval, context engineering, inference economics, and evaluation.
Five tracks, one capstone.
Coding-agent architecture
How real agent harnesses work, from mini-SWE-agent to production SDKs.
Retrieval & code localization
BM25, embeddings, AST, dependency graphs, RepoMap, and reranking.
Context engineering
Selection, compression, compaction, and memory policies that raise token utility.
Inference economics
Cache economics, model routing, and cascades: cost per task, not per token.
Evaluation & optimization
Outcome vs process metrics, and proving gains without regressions.
The context optimizer
A token-efficient context optimizer for a SWE-bench-style coding agent.
Who it's for
Engineers and technical leaders running a coding agent in production who want to cut cost and latency without losing task completion.
Prefer we do it with you? Request an agent review →
Get the field guide
Enter your work email above and we'll send the PDF plus the companion Insights series.
Get the guide