Somebody indexed 977 agentic coding tools and the most honest line on the whole site is "a high score is not a recommendation."
respect tbh. Most "top AI coding agents" listicles would never.
ADE Index went up as a Show HN this week, pitched as "a sourced and ranked index of 975 agentic coding tools." By the 19 Nov 2026 snapshot on its methodology page it's 977 listings, and only 264 of them have a rank. 702 sit at "Not enough data yet."
My take: the ranking is a popularity meter and the site says so out loud. The actually interesting part is that second number. Nearly a thousand harnesses, workspaces and orchestrators exist, and most of them haven't earned enough Hacker News chatter in 90 days to get scored at all.
ok so how does the score actually work
The Index score is Hacker News mindshare, meaning stories and comments that discuss the product in the last 90 days. That's 80 to 100% of it. Community votes can add up to 20%, but they can only raise a score, never lower it. GitHub stars are shown next to the score and not counted.
The formula is public, which is lowkey the best thing about the project. Mentions go through a log scale, so "the first hundred mentions count for more than the next nine hundred." One viral launch can't eat the leaderboard.
Claude Code CLI scores 92 off 4,602 Hacker News mentions and sits at #1. Cursor's editor gets 75 from 1,045 mentions, rank #5.
why no stars? Per the methodology page, closed-source tools have no repo so they have no stars, and blending stars in lifted open tools "for a reason that has nothing to do with how much developers use or discuss them." Fair call imo.
Votes need a GitHub account at least 30 days old and maker self-votes don't count. When votes joined the score on 2026-11-18 no eligible votes had been recorded, so for now it's basically all Hacker News.
The thread was small (25 points and 7 comments when we pulled it) but it had opinions.
One commenter looked at the Pi vs Oh-My-Pi comparison and said the comparisons are "data points that don't actually say much about the product" and don't tell you which one is better. They also guessed the site "was built in a short amount of time through a set of agents."
Another checked Antigravity's listing. They found it unranked with no Hacker News mentions in the last 90 days, capabilities at 2 of 8, and a screenshot they said shows the Antigravity App instead of the IDE. Their verdict: "Very bad data quality, and the data is literally the whole point."
The maker took it pretty well ngl. They replied it was "indeed built very fast," that early work went into indexing and automated pipelines, and that sentiment is meant as a pre-filter so nobody has to sort 1,000 tools by hand.
hot take: the bones are fine, the data is the product
I kinda disagree with the slop pile-on. At least half of it.
The methodology is more careful than most rankings out there. Every product fact is meant to be a verbatim quote with the URL and the date it was read. Where there's no quote the cell says "Not yet checked," and the site says it never prints a negative like "no free tier" without a quote that says so. In a market this noisy that's the right instinct.
But the commenter is also right that the data is the whole point. a directory where 702 of 977 listings have no score and a big-name listing reportedly has the wrong screenshot is gonna get judged on those gaps, not on how pretty the formula is.
Then there's the obvious question. ADE Index is made by the team behind atrium, a desktop workspace for coding agents. They disclose it, atrium goes through the same formula (today: "not enough data yet") and any Hacker News story that links to adeindex.com is dropped from the mindshare counts. More disclosure than most review sites manage, but worth knowing when atrium shows up in the "Owner's product" slot.
what 977 harnesses means if you're building agents
The homepage counts 237 CLI agents, 113 orchestrators and 383 agent workspaces, and says 156 of those workspaces shipped a release this week.
One commenter said it gives them 2000s web wild-west flashbacks and expects common themes to show up in 6 to 8 months. Same read here. You're probably gonna switch harness more than once this year.
So don't bake the important stuff into the harness. Say you wanna try a terminal workspace you found on the list, like pando, which shows 9 stars. If your app tokens get pasted into that tool's MCP config, you're handing them to whatever you tried this week.
This is where I'd put MewCP in the middle. Its docs connect Claude Code, Cursor, Codex, OpenCode, DeepSeek Harness and Antigravity with the same URL and key, and any app that can reach a remote MCP server over streamable HTTP with a custom header can connect too. Your agent never holds the app logins, and you choose which tools each agent can use, enforced on every call. new harness, same GitHub access as the old one, nothing extra.
A commenter who says they've built an ADE for 8 months added that the most important piece is "how to let the human know when & where they're needed." If I was picking from this list that's what I'd test first.
stuff we still don't know
- Whether the data catches up. The listing complaints are specific, and the maker says routines check content quality. Next snapshots will tell.
- How votes behave. None counted yet, so no live data.
- Sentiment. The methodology page says it won't affect rank until the method is validated, while the maker frames it as a pre-filter.
- Movement. Ranks started on 18 Nov 2026, so there's no history yet.
where all this came from