What belongs to the harness, and what can't a prompt alone handle?
Review note: The answer refers to the original author's CV or target job offer — adapt it to your own experience.
Answer out loud or in your head first, then compare.
Answer
The harness is the executing program around the model: it assembles the input, handles tools, saves state and enforces limits. A prompt describes the expected behaviour, but it alone won't provide a database transaction, restricted file access or correct handling of a restart.
In a coding agent, the repo, the terminal and the way changes are checked are part of the environment. When evaluating such an agent, we test the model together with its tools and configuration. The same model can behave very differently in two environments. Your own skill extends a work procedure, but it doesn't automatically mean you built a whole agent runtime. That's an important distinction when discussing your experience.
Why it matters: Develops the simple definition from I-P01 into the limits of responsibility.
Possible follow-ups: Which component should end the work when the budget runs out?
How did it go? Be honest — “Not yet” puts the question on your review list.
Start with
- I-P01 · What is an agent harness? — Harness, MCP and coding agents
- I-P02 · What is MCP? — Harness, MCP and coding agents
- I-P04 · How does a coding agent differ from a chat that suggests code? — Harness, MCP and coding agents
Further reading
- Anthropic: Effective harnesses for long-running agents(opens in a new tab)26.11.2025. Working in stages, progress artifacts and resuming between sessions.
- Anthropic: Harness design for long-running application development(opens in a new tab)24.03.2026. Planner/generator/evaluator, handoffs and assessing the agent's work.