Key findings
- All 15 accounts that discuss model choice praise it, from local Qwen and gpt-oss models to hosted GPT-5, Claude, and Gemini.
- All 9 accounts that discuss context control credit choosing the files the model sees, and 6 report lower token use or cost than with agentic tools.
- Engineers split on autonomy: 10 accounts prefer Aider's developer-driven control, and 13 moved to or prefer agentic tools, 10 of them Claude Code.
- Release pace is the newest concern. Every account that discusses it dates from September 29, 2025 or later, and 10 of the 11 raise it as a problem.
Engineers praise Aider for two choices: the model and which files it sees. After Claude Code and other agents arrived and Aider's releases slowed, debate shifted to autonomy and release pace. Score fell from +42 in the first half of 2025 to +13 in 2026; pin a version and test it against an agent on the same tasks before standardizing.
Who wrote these accounts
GILD read 85 public accounts by 81 authors, published from January 7, 2025 to October 8, 2026. They include 55 Hacker News comments, 13 GitHub issues and 6 issue comments on the Aider repository, and 7 posts on personal sites, Substack, and DEV Community. Others are one XDA Developers review, one Spotify engineering blog post, one Six Feet Up company news post, and one AIMultiple benchmark. Each names its author and links its page, and GILD recorded the version, provider, or model wherever the author states one.
75 accounts come from individual engineers. Ten speak for a team or an organization: Spotify's engineering blog, two engineering leaders, two professional teams, one of them AIMultiple's benchmark team, and five very small teams. Their reports are in the team section below.
What engineers value
The praise holds from January 2025 to October 2026. Engineers choose the model per task, often a reasoning model as architect beside a cheaper editor model, and some run local models where code cannot leave the machine. They add only the files that matter, and several report spending less than with agentic tools. One keeps context under 25,000 tokens at about 3 US cents a prompt on Gemini.
All 6 accounts that discuss git integration like it, because a commit per edit lets them jump back when the model loses the thread. Engineers who dislike automatic commits turn them off and review diffs in their editor.
- Six accounts get better edits by starting in ask or architect mode before code mode.
- Eight name small, scoped changes as the condition for good results.
- One freelancer counted a record 2,900 commits in 2025 after adopting Aider, against about 2,000 a year before 2023.
Where the debate moved
Through mid-2025 the complaints were about adding files by hand and weak results on cheaper models. Once Claude Code and similar agents arrived, autonomy became the argument. Among accounts that take a side, 6 of 8 in the first half of 2025 preferred agentic tools. In the second half, 6 of 9 preferred Aider's control. In 2026, 4 of 6 had moved to an agent.
From September 2025, engineers also asked whether anyone still maintains Aider. The GitHub issues "No new releases since early Aug" and "What is the intended future of Aider?" drew replies from users who had moved to a community fork or to Claude Code. Others stayed and said the slower pace suits a tool they already know.
| Period | Accounts | Positive | Mixed | Negative | Score |
|---|---|---|---|---|---|
| Jan to Jun 2025 | 36 | 19 | 13 | 4 | |
| Jul to Dec 2025 | 25 | 10 | 10 | 5 | |
| 2026 to Oct 8 | 23 | 7 | 12 | 4 |
GILD analysis of the 84 public accounts that give an opinion, January 7, 2025 to October 8, 2026
What teams report
Spotify's platform engineers tried Aider and Goose for a background agent that migrates code across the company's repositories. Both went far on a prompt alone, but at that scale neither reliably produced mergeable pull requests, so Spotify built its own agent loop. An ML team uses Aider daily and forked it to work from Jira tickets as AI use spread across its company. One engineer wants to wrap Aider as an IDE plugin because less experienced colleagues find the CLI install too hard. Two teams running Aider headless found that it exits with code 0 after fatal API errors, so their automation logs failures as successes.
One team turned auto-commits off because Aider's commits skip its git hooks. A small team picked Aider for the control it gave, then moved to Claude Code, which finishes the task and opens a pull request, for less money. An engineering manager whose team maintains nearly fifty systems used Aider, then Crush, to rebuild context on old, undocumented code, and found the bigger diffs put more review on him and his teammates.
What to test in a pilot
Run the same tasks in Aider and in the agentic tool you would otherwise buy, on the same model, and count accepted changes and review time for each. Pin the Aider version, enforce git hooks with --git-commit-verify, and check exit codes and test results yourself in every headless run. Write down beforehand which gap in releases would make you re-evaluate the tool.
Methodology
GILD opened every account on October 12, 2026. It recorded the author, publisher, publication date, overall sentiment, evidence type, and each position taken, in the same format as the Claude Code Signals. A verbatim excerpt records whether each account speaks for one engineer or a team. Every position rests on a verbatim excerpt checked against the saved page. Hacker News comments came through the Hacker News API and GitHub issues through the GitHub API, and only accounts of the author's own use count. Comments by Aider's maintainer, Aider's own testimonials, news write-ups, and content-farm reviews are excluded. Medium posts are left out because Medium blocks automated reads, and so are accounts dated before January 1, 2025. GILD added five accounts on October 25, 2026 from a company engineering blog, an independent benchmark, a personal blog, and GitHub issues, each with a verbatim quote. A finding reaches the Signals page only when at least 5 accounts from 3 authors discuss it. The counts describe these 85 accounts only.
Products in this piece
Sources
- Official documentation
- 6
- Public source
- 20
- Aider
- Aider docs: Git integration
- Aider docs: Chat modes
- Aider docs: Tips
- Aider releases on GitHub
- aider-chat on PyPI
- Andrew Keenan Richardson: Claude Code vs Aider (Mar 20, 2025)
- Harper Reed: My LLM codegen workflow atm (Feb 16, 2025)
- Harper Reed: Basic Claude Code (May 8, 2025)
- Max Charas and Marc Bruggmann, Spotify Engineering: Background Coding Agents: Context Engineering (Honk, Part 2) (Nov 24, 2025)
- Sedat Dogan and Şevval Alper, AIMultiple: AI Coding Benchmark: Claude Code vs Cursor (Aug 21, 2026)
- Anurag Singh, XDA Developers: I stopped forcing every coding job through Claude and started using this instead (May 22, 2026)
- Hacker News, BeetleB: branches, a 25k-token context cap, and Aider as cheaper than Claude Code (Mar 9, 2025)
- Hacker News, BeetleB: about 3 cents a prompt on Gemini under 25k tokens (Jun 12, 2025)
- Hacker News, Karrot_Kream: plan, code, and review in Aider at a fraction of Codex's cost (Jun 12, 2025)
- Hacker News, chr15m: commits per year before and after Aider (Feb 24, 2026)
- Hacker News, tharkun__: the repo map overflowing on a 100-service monorepo (Sep 9, 2025)
- Hacker News, flyinglizard: still loyal, but a single maintainer (Sep 29, 2025)
- GitHub issue #3871: an ML team using Aider daily extends it to Jira tickets (Apr 23, 2025)
- GitHub issue #4215: wrapping Aider as an IDE plugin for colleagues (Jun 12, 2025)
- GitHub issue #5552: Aider exits with code 0 on fatal API connection errors (Aug 11, 2026)
- GitHub issue #3583: Aider using no-verify by default for commits (Mar 20, 2025)
- GitHub issue #4584: No new releases since early Aug (Oct 13, 2025)
- GitHub issue #4648: What is the intended future of Aider? (Nov 15, 2025)
- GitHub issue #4751: Thoughts and questions about the future direction of aider (Jan 7, 2026)
- Hotovo: AiderDesk AI orchestration platform, a case study of Hotovo's own engineers (undated, opened Oct 25, 2026)