Based on official documentation
The provider documents these capabilities. The steps below are an editorial way to evaluate them; we have not independently reproduced the result.
The provider documents these capabilities. The steps below are an editorial way to evaluate them; we have not independently reproduced the result.
What you’re working toward
Work with a coding agent on a bounded task, then inspect the diff and run the project’s actual checks.
Who it’s for: Makers and software developers. Experience: intermediate.
A practical way to start
- Define a small task with a concrete expected result and a way to test it.
- Start the coding agent in the correct repository and give it the project instructions.
- Review requested permissions and inspect the changes it produces.
- Run relevant tests and review the diff before merging or deploying.
Know the limits
- An agent-produced patch is not evidence of correctness.
- Access, quotas and product behavior vary between terminal and cloud experiences.
Cost and alternatives
Check the tool profiles and current provider plans before starting. Model usage, hosting and optional connectors may have separate costs. Alternative tools need their own configuration and evaluation.
Compare the approaches
Sources and evidence
- 1. OpenAI — official source →Observed Oct 16, 2026 · official
Officially documented capability; the steps here are an editorial evaluation approach, not a claimed independent execution.
- 2. Anthropic — official source →Observed Oct 16, 2026 · official
Officially documented capability; the steps here are an editorial evaluation approach, not a claimed independent execution.
Something changed or needs a closer look? Read our corrections policy.