How DevConsensus works

1. Collect

• We search Hacker News comments month by month for the names of AI coding tools and models.
• We keep comments that name a tool or a coding model and contain an opinion or a personal experience. This edition covers 20,064 comments from Nov 1, 2025 to Nov 2, 2026.

2. Extract

• A language model reads each comment and records which tools or models it judges, whether the judgement is positive, negative or mixed, whether the author says they use it, which topics it covers (cost, speed, code quality and so on), and any explicit switch from one tool to another.
• Opinions about non-coding uses of a model are dropped. Spot checks of random samples are used to tune the extraction.

3. Score

• Approval = (positive + half of mixed) ÷ all opinions that took a side. Neutral mentions are not counted.
• Ranking uses the lower bound of a 90% confidence interval on approval, so tools with few opinions cannot jump ahead of well-tested ones.
• A tool or model needs at least 15 opinions to get a page.

4. Limits

• Hacker News readers skew towards experienced, terminal-friendly engineers, so tools built for them tend to score well.
• Older opinions may describe versions that have since changed. Check the 90-day column and the trend chart.
• The comments we show are paraphrased summaries. Every one links to the original comment on Hacker News.

Referral links

• Some “Visit” links may become referral links. Rankings are computed before any link is attached and are never changed for a partner.