R023 · AI education › Learning paths for AI engineers

Building a coding agent from scratch: an open harness-engineering course

Read the original

agent-harnesseducationlearning-resourcesevalscontext-engineeringagent-architecture

TL;DR

Paul Iusztin and Decoding AI released a free, Apache-2.0 course and codebase that builds a Claude Code-style coding agent, called decode, in Python. Its thesis: the agent loop is about 20 lines; the engineering is the harness around it. Eight articles and six videos walk through that harness, from the loop to sandboxing, context engineering, subagents, remote headless runs, evals and record-and-replay.

Architecture

The course builds one headless harness and attaches two interfaces to it.

The repo carries ADRs for each design choice, a glossary with one name per concept, and unit and integration tests that mirror the source tree, with capstone tests per milestone.

Lessons

#LessonWhat you build
1Building a coding agent from scratchThe full harness sketch: headless core, six modules, two modes, evals
2The bare-bones agent loopNine tools, three providers, a steerable TUI
3From a raw shell to a sandboxTools run in Docker or Modal, never on the host
4Context engineeringMemory, skills, LSP diagnostics, compaction at 80%
5Subagents are context engineeringParallel Explore subagents; build, plan and reviewer personas
6Swarm of remote agentsHeadless runs on Modal from CLI, webhook or cron
7Agent evals 101A 19-task benchmark and a regression suite
8Failures into a regression suiteRecorded runs, failure cohorts, replay to verify fixes

Tradeoffs and failure modes

The structure is strong; the claims around it are the sponsor's and the author's.

Implications

For anyone teaching agent engineering, this is a ready-made syllabus whose order matches how production harnesses are built: loop, then isolation, then context, then scale, then evals.

Checklist before adopting it in a course or team:

Sources

Connections

See R023 on the knowledge map

Part of the Research notes on nowucca.com. Figures and claims are the source's own unless the summary says otherwise.