GitHub’s recent wave of updates for Copilot and its tooling is centered on making the CLI and agent features more interactive and context-aware. The January 21 changelog entry, GitHub Copilot CLI — plan before you build, steer as you go, outlines several workflow-oriented additions that pair with the broader Copilot ecosystem changes rolled out across early January.
What the Copilot CLI adds
The CLI updates emphasize interactive decision-making during development rather than only post‑hoc suggestions. Key additions include:
- Interactive plan mode — enables creating and reviewing a stepwise plan before code generation, helping to align agent actions with intended design.
- Approve for session — a session-level approval flow that reduces repetitive confirmations during longer agent-driven tasks.
- Configurable reasoning effort — controls for how much internal reasoning the agent should expend, trading latency for depth of analysis when necessary.
- Inline feedback on rejection — allows succinct feedback to be attached to rejected suggestions, improving future responses.
These features complement the recently introduced agentic memory system, which keeps shared memory across development workflows. Together, the CLI controls and shared memory aim to make multi-step, agent-driven workflows more predictable and iterative.
For the full changelog entry and links to each update, see: https://github.blog/changelog/2026-01-21-github-copilot-cli-plan-before-you-build-steer-as-you-go