Coding agents
What is Agentic Coding?
Agentic coding is software development performed as an agent workflow: plan, edit, observe, repair.
The repository and tests are the environment.
Why it matters
Agentic coding is software development performed as an agent workflow: plan, edit, observe, repair. The repository and tests are the environment — the agent interacts with them the way a research agent interacts with the web.
The key insight is that coding is not just text generation. It is a feedback loop where tests and linters provide evidence that the code is correct.
Key takeaways
- 1The repository and tests are the agent's environment.
- 2Agentic coding is a feedback loop: plan → edit → test → fix.
- 3Tests provide evidence; the model provides intent.
Common mistakes
- ✕Generating code without running tests in the loop.
- ✕Treating code generation as a one-shot task instead of an iterative process.
Related terms
Concept neighborhood
Terms linked from Agentic Coding in the glossary graph.
- Agentic Coding
- AI Coding Agent
- Vibe Coding