Amazon Q Developer

The trait that decides this is where it lives. It answers inside the AWS console, in Slack and in Teams, which means questions about your own account and your own resources get asked in the place you are already looking rather than in a browser tab you have to paste into. For people whose day is half infrastructure, that saves real minutes.

Pick something else if your infrastructure is elsewhere, because the advantage evaporates the moment the account is not AWS. Pick GitHub Copilot when the code review and the pull requests are the centre of gravity.

The risk is a product line changing under its users. The terminal client was open, then it stopped being actively maintained, and the recommended path is now a different, closed client under a different brand. Anyone who scripted against the first one is holding a migration nobody scheduled, and the replacement cannot be read or pinned the way the original could.

Assume the surface you build on may be renamed. What it does right: the free tier states a hard number, fifty agentic requests a month, instead of describing limits as generous and settling them later.

Most agent claims resist evaluation because success is a matter of opinion. A language runtime upgrade is not: either the project builds and its tests run on the new version or it does not, and the same holds for porting a .NET application from Windows to Linux. Framing an agent around tasks with an objective terminal state is a sound design decision and rarer than it should be.

No benchmark accompanies any of it. The general coding loop reads and writes files, produces diffs and runs shell commands, and its verification stage is the user reading the diff.

The strategic job of this product is keeping engineers inside the account, which means it is subsidised by infrastructure margin and never has to justify itself as a standalone business. That is a comfortable position and it produces a particular kind of roadmap: broad, slow, and driven by what the platform needs rather than what the category is doing.

Moat: the enterprise agreement and the identity boundary, neither of which a startup can cross. Likely event: a rebrand under whichever internal name wins. Position: safe to depend on the capability, not on the label.

The finance case is the easy one: $19 per user per month, $1,140 a month at our headcount, on a bill our finance team already processes and under a contract our lawyers signed years ago. The administrative controls we need for a rollout are in the paid tier rather than sold separately, which avoids a second negotiation.

Rollout is a plugin push to four editor families plus a chat integration, so onboarding for a mid-level engineer is an afternoon. Support is a vendor with a support organisation, which is the part startups cannot match. Approved.

The model is fixed and I cannot supply my own, which rules out everything I like doing. No local inference either, so this is a rental in every sense. That is most of my score.

The redeeming detail is configuration. MCP servers are declared in files under ~/.aws/amazonq, with the CLI agents in cli-agents and the editor reading agents/default.json, so the tool wiring lives in my dotfiles rather than in a settings dialog. Grudging respect for that choice: somebody on the team understood that configuration you can commit is configuration you can trust. The rest is a closed box.