GitHub Copilot

Copilot is the tool most people already have, and agent mode has caught up enough to matter for everyday tasks: a bug from an issue, a test for a function, a rename across a module. The trait that decides it: the coding agent lives where your issues and pull requests already are, so assigning a task is the same motion as assigning it to a colleague.

Pick it if your code is on GitHub and you want zero setup and one bill. Pick Claude Code or Cursor when you need the strongest agent for a hard refactor, and keep Copilot for the routine work in between.

The risk is split. The cloud coding agent is isolated in an ephemeral Actions environment, but it is metered as premium requests, so the safer path is the one that upsells you. Local agent mode edits and runs commands on your machine outside that sandbox, so a wrong command lands on your laptop.

The consequence is that teams on the base plan will run the unsafe mode by default and discover the safe one when the bill arrives. Ship the isolation to the local mode and this verdict changes. What it does right: the cloud agent returns a pull request, so nothing lands on a branch without a human review.

Two documented halves. 1. Local agent mode runs in the editor with direct filesystem access, and context is the open workspace. 2. The cloud coding agent runs in an ephemeral GitHub Actions environment and hands back a pull request, so the repository's own CI is the verification step, run by the same workflow that gates every human change.

No benchmark is published for either half, so capability is a feature list. The second design is the sound one: verification is borrowed from infrastructure the team already trusts rather than invented. The observation: the strongest part runs where the tests already run.

Copilot is the incumbent and incumbency is the moat. It ships inside the editor and the repository host where the world's code sits, on accounts procurement approved years ago, and every competitor on this board has to displace it before selling anything. Moving the meter to premium requests without losing the customer is pricing power in its purest form.

The risk is not runway, it is complacency: an incumbent that bundles can ship a weaker agent for years. Likely acquirer: none needed, Microsoft owns the whole stack. Position: long by default, the way you are long the index.

The demo is an issue that becomes a pull request while nobody watches. Procurement is mostly done: the GitHub account exists, Business and Enterprise carry policy controls and IP indemnity, and the audit log is the one we already read. Onboarding is a license assignment.

Sixty Pro seats is $7,200 a year, and that is not the number; premium-request credits are, because the agent modes draw on them and nobody budgets a meter that starts after the seat. CI fit is native, the cloud agent runs in Actions. Approved, with an organization-level cap on the meter and a monthly report of who spent it.

Copilot surprised me. BYOK is real, local models work in the CLI, and MCP config is the standard servers JSON block with command and args, so my existing servers drop in. That is more than the other incumbents give me, and I did not expect it from this one.

The ceiling is that I cannot read the agent's prompts, cannot change the loop, and nothing forks if Microsoft loses interest, so the good parts are settings on a closed binary. Grudging respect for the BYOK. It is a rental with good windows, and I still would not sign a long lease.