Aider

Aider is the agent you can see through. You add the files, it proposes the change, and you watch it land, so nothing happens in your repository that you did not read first. That control is the point and the ceiling: it will not wander a monorepo looking for the right module, and it will not run unattended while you get lunch. Switching models mid-session is a flag, not a migration.

Pick it if you like steering and want to pay the provider directly with no middle layer. Pick Claude Code if you want to hand over a task, come back later, and read the result rather than the process.

The risk is that there is no isolation: a wrong lint or test command runs where you are standing, with your credentials in scope. That is the same exposure as running the command yourself: the tool is exactly as dangerous as the commands you allow it.

The mitigation is designed in. Every change is a git commit, so a corrupted edit is one revert and a bad session is a range you can drop. That is the right trade for a tool that shows you everything. What it gets right beyond that: the docs describe every flag, so there are no surprises.

Aider documents its design. 1. Context comes from a repository map built from a tree-sitter parse and ranked by graph centrality, so the files shown to the model are chosen by reference structure, not embedding similarity. 2. Edit formats are chosen per model, which is why the polyglot leaderboard exists: it measures which model follows which format. 3. The SWE-bench Lite 26.3% and full 18.9% figures are self-reported from 2024 with the models of the time and read as history.

The consequence: Aider's numbers are comparable across models and not against other agents. The observation: publishing the harness is worth more than the score, and most vendors here publish neither.

Aider is the labor-of-love case. No revenue signal, no pricing power because the provider takes the whole meter, and a leaderboard the industry cites for free. Sunset risk is real: the release cadence has slowed, and a project with one principal maintainer has a bus factor a diligence memo would flag.

The hedge is that a fork would be trivial, so the code survives even if the name does not. There is no acquirer for a tool with no company; the likely pivot is a handoff to a foundation or a fork. Position: use it freely and do not expect a vendor to call back.

The demo is a terminal that edits three files and writes a sensible commit message. Procurement is trivial and that is the issue: free means no contract, no SSO, no SCIM, no audit trail, and no one to call when it breaks on a Friday. Sixty engineers means sixty keys on our provider account, and a key in a dotfile ends up in a screenshot eventually. Onboarding is a pip install and an afternoon.

Approved with conditions: keys issued centrally and rotated, spend limits per key, and it stays a power-user tool rather than the standard, because the average engineer needs the guardrails it deliberately does not have.

Aider is the reference for ownership. The prompts live in the repo, so I can read what it tells the model and change it. .aider.conf.yml sets model, edit format and lint command, and --model ollama/whatever runs it fully offline on my box with nothing phoning home. It is Python, which means the fix for a bug is a patch I can write before the maintainer wakes up.

The gap is MCP: there is no client, so tool servers are out and I wire tools through lint and test commands instead. If the maintainer walked away I would fork it in an afternoon. That is not respect, that is a recommendation.