Which agent can write and run code for me from the terminal or a ticket?

+ Handles long multi-step tasks across a whole repo with strong tool use and MCP support.
− Token spend on the API plan can be high and Pro-plan rate limits reset only every five hours.

+ Tab autocomplete predicts multi-line edits across files better than any competitor.
− Usage-based pricing changes and request limits on Pro have repeatedly frustrated heavy users.

+ Deepest GitHub integration: assign an issue to Copilot and it opens a PR.
− Completions and agent lag behind Cursor and Claude Code in quality despite more models being offered.

+ Cloud mode runs many tasks in parallel sandboxes and opens PRs for each.
− Native Windows support in the CLI arrived late and sandboxing can block legitimate commands.

+ MCP-first design means any MCP server becomes a Goose capability with no glue code.
− Less tuned for coding than Claude Code, so results vary widely by model.

+ Free tier gives a 1M-token context window and around 1,000 requests per day.
− Less reliable at multi-step edits than Claude Code and quietly falls back to Flash under load.

+ Runs fully unattended in its own VM with browser, shell, and editor, and reports back on Slack.
− Struggles with ambiguous tasks and the price per compute unit adds up on long sessions.

+ Inherits Claude Code's file, shell, and subagent tooling, so agents are production-grade on day one.
− Tied to Claude models and the Claude Code runtime rather than a provider-agnostic design.

+ Plan mode lets you approve an approach before the agent touches files or runs commands.
− Reading whole files into context burns API tokens fast on big projects.

+ Repo map and git-native workflow make it cheap and precise with any model, including local ones.
− It is a chat-driven editor rather than an autonomous agent, so you steer every step.

+ Works out of the box with a built-in provider so no API key setup is needed.
− Less mature than the projects it forked and still catching up on polish.

+ Top-ranked open agent on SWE-bench that you can run on your own infrastructure.
− Docker-based local setup is heavy and the UI is rougher than commercial agents.

+ Custom modes (Architect, Code, Debug) and Orchestrator subtasks keep context small per step.
− Rapid releases mean frequent regressions and settings sprawl.

+ Threads are shareable, so a team can inspect exactly how an agent produced a change.
− No model picker; Amp chooses the model and spends your credits accordingly.

+ Uses the IDE's own semantic index and refactorings, so edits respect project structure.
− Quota-based AI credits run out quickly on the lower tiers.

+ Cascade keeps context of what you just did in the editor and terminal without being told.
− Ownership turmoil in 2025 (OpenAI deal collapse, Google, then Cognition) caused model-access gaps.

+ Context retrieval holds up on monorepos with hundreds of thousands of files.
− Pricing moved to credits and sits well above Copilot for individual developers.

+ Spec-driven workflow produces requirements and design docs before code, which keeps agents on track.
− Waitlists and tight free-tier limits made it hard to actually use after launch.

+ Runs in the background and sends an audio changelog summarising each PR.
− Limited daily task quota and no way to intervene mid-task once it starts.