The five layers of an agent harness, mapped: Microsoft wires the middle three
adnan_hashmi · x · 2026-11-09
A breakdown of the five layers of a harness agent in Microsoft Agent Framework:
- Chat client (you): the model; Framework 1.0 ships first-party connectors for Foundry, Azure OpenAI, OpenAI, Anthropic, Amazon Bedrock, Google Gemini and Ollama (the last four still pre-release in Python).
- Chat pipeline (harness): function invocation with iteration limits, message injection, history saved after every model call, optional compaction.
- Context providers (harness): instructions, tools, todo lists, plan/execute modes, file memory, optional skills and background agents.
- Middleware (harness): tool approval with standing rules and auto-approval callbacks, OpenTelemetry, optional bounded looping.
- Application (you): streaming output, progress display, approvals; Microsoft only ships console samples.
One createharnessagent call wires the middle three layers; you bring the first and the last.
More from coding & agent
- Open-source Rewind for X extension logs every post you scroll past so refreshes stop losing good tweets — dotey · 2026-11-10
- Meta's LLM-generated failing tests catch 74 privacy SEVs, cutting incidents 20% — Meta_Engineers · 2026-11-10
- From Data Scientist to AI Engineer: The Evolving Skill Stack Explained — mdancho84 · 2026-11-10
- Hark Labs' model + agent + VM experience beats Codex, says early reviewer — Scobleizer · 2026-11-10
- Dev rebuilds Pokémon Red in 2.5D from his own ROM cartridge using an agentic workflow — NathanWilbanks_ · 2026-11-10
- Same prompt shootout: Claude vs ChatGPT vs Figma Make on an e-commerce landing page — Tegadesigns · 2026-11-10