# Agent first, inside and out (slim) > Vlad Cealicu on how Anchor Terminal runs on agents, what we do and why, with the deploys, freezes and fixes behind each one, and what we choose not to do because we don't think it adds anything. - Full: https://www.anchorterminal.com/blog/agent-first-inside-and-out.md (~4,800 tokens) · this version ~1,730 tokens · JSON https://www.anchorterminal.com/blog/agent-first-inside-and-out.json · canonical https://www.anchorterminal.com/blog/agent-first-inside-and-out - Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-11-15 - Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data) By Vlad Cealicu*, 15 November 2026. 2,250 words, about 10 min to read. At four in the morning on 5 November a deploy failed two seconds before it would have worked. Most of Anchor Terminal runs like that, with agents doing the work, two of us deciding, and everything an agent needs to do it the way we would written down where it will look first. How the site is put together, the files, the skills and the agents that research and grade the listings, is in [a post from 9 November](/blog/a-machine-that-builds-the-benchmark) [[1]](#ref-1). ## What we do Every agent reads one file first, CLAUDE.md. Then there's a spec for every feature under docs/design, with what was decided, why, what we rejected and the numbers, and my words wherever I decided something in a chat [[3]](#ref-3). We write down how we sound, VOICE.md for the site and one each for Quynh and me, and the build fails any page or file with an em dash or a word off the banned list in it [[5]](#ref-5). Everything is in git and lands through a pull request, and Quynh or I read each one and press the button. Each job an agent does again and again is a skill of its own, small and with its reasons. And we keep the data nobody else has, because it's the only part a competitor can't download. ## Agent first on the site The site is the same idea pointed outwards. That DNS answer was the one in the post about it in October, with that day's numbers [[11]](#ref-11). Agent first doesn't mean people second, and this week showed why. ## Checking ourselves, and the lab This week we added four things, and the facts file is the one I wanted most, because I'd been bitten by the thing it's for. Two more are for the lab box, a Minisforum MS-S1 Max with 128 GB of unified memory that sits next to our server and can only be reached over SSH from it [[17]](#ref-17). What runs where, and what leaves the box 1. **Read** (gateway). The harness opens the listed MCP server and reads its tool list, and nothing else is reached [[17]](#ref-17) 2. **Write** (model). The model on the box reads the tool list and writes a small Go program to try each tool [[17]](#ref-17) 3. **Run** (cell). The program runs with no network, no files of its own and limits on memory and time, and can only ask for a call [[17]](#ref-17) 4. **Call** (gateway). One public address per run, never a tool named like a change, every call counted and timed [[17]](#ref-17) 5. **Record** (journal). Every step goes to a file as it happens, and only numbers and fixed words leave the box [[17]](#ref-17) 6. **Collect** (not built). The site's server will collect the numbers and check every field before a listing shows them, and that part is still to build [[17]](#ref-17) Every job has a cost we can see, so a job that costs more than it's worth gets fixed or stopped, and every test on the box records the model's time and tokens and how long the programs took. And we pick the lab's model with fifty cases that have known answers, run again whenever a new model comes out. The fourth is for the companies we list. None of the lab has run on the box yet. ## What we don't do We don't install other people's skill packs, because I don't think they'd add anything. We don't turn customer requests into pull requests before a person reads them. We don't run outbound or ads on their own. We don't run an orchestrator that never turns off, with a dashboard for the fleet. What's ours is the research behind every listing, the decisions with the reasons next to them like the six hours and Pamba, and soon what our own agents met when they called the tools we list. --- More in [Inside Anchor Terminal](https://www.anchorterminal.com/blog/inside/index.md) · Previous [How we built Agent of Somebody](https://www.anchorterminal.com/blog/how-we-built-agent-of-somebody.md) - Older: [A machine that proves Fermat in 29 hours, and the behemoth that's back](https://www.anchorterminal.com/blog/a-machine-that-proves-fermat.md) - All posts: https://www.anchorterminal.com/blog/index.md Anchor Terminal grades the tools AI agents run on, with a source for every score. ## References Read on 15 November 2026. [1] Vlad Cealicu, A machine that builds the benchmark, Anchor Terminal, 9 November 2026. Source for how the site is put together, the skills and the research agents. [2] Anchor Terminal, docs/design/deploys.md, 15 November 2026, not published. [3] Anchor Terminal, docs/design, the specs, 15 November 2026, not published. [4] Anchor Terminal, CLAUDE.md, 15 November 2026, not published. [5] Anchor Terminal, VOICE.md, VOICE-VLAD.md and VOICE-QUYNH.md, and the build's voice check, 15 November 2026, not published. [6] Anchor Terminal, pull request 805, the vendor brief, 15 November 2026, not published. [7] Anchor Terminal, the listing-check and landing-page skills, 15 November 2026, not published. [8] Anchor Terminal, docs/design/listing-policy.md, 13 November 2026, not published. [9] Anchor Terminal, docs/FACTS.md, 15 November 2026, not published. [10] Anchor Terminal, for agents. [11] Vlad Cealicu, dig TXT mapbox.grade.anchorterminal.com. [12] Vlad Cealicu, Why we built Anchor Terminal, Anchor Terminal, 8 October 2026. Source for building the site for agents first with a Markdown and a JSON copy of every page, and for agents reading the HTML 1,085 times when the Markdown was right there. [13] Anchor Terminal, docs/design/directory-filter.md, 14 November 2026, not published. [14] Anchor Terminal, docs/design/reviews-paged.md, 14 November 2026, not published. [15] Vlad Cealicu, What's hyped now, what already peaked, and which company you're in, Anchor Terminal, 13 November 2026. Source for the first version's 25 of our 27 decision models and why it was wrong. [16] Anchor Terminal, docs/design/facts.md and the build's facts check, 15 November 2026, not published. [17] Anchor Terminal, docs/design/lab.md and lab/README.md, 15 November 2026, not published. [18] Anchor Terminal, lab/bench and docs/design/lab.md, 15 November 2026, not published. [19] Anchor Terminal, docs/design/vendor-brief.md and the brief for Exa made from the site's data on 15 November 2026, not published. [20] Vlad Cealicu, AI is eating bureaucracy, Anchor Terminal, 7 November 2026. Source for review text never being treated as instructions or put into a prompt. [21] Anchor Terminal, docs/COMPANY.md, 3 November 2026, not published. [22] Anchor Terminal, pull request 807, 15 November 2026, not published.