ZeroNoise Logo zeronoise
Post
Theo's "Council" skill: models from different families must score a change before the agent stops
•
5 min read
• 74 docs
Theo's Council skill brings models from different families in to review changes and won't stop until they give a set score. Also: Steve Yegge's case for slowing down, Kent C. Dodds' software-factory tips and remote agent setup, and a larger context window for Gemini models in Antigravity.

Theo's Council: reviewers from different model families, with a score to hit

In "Stop Searching For the Perfect Model," Theo argues against routers that pick the right model for each task. He mostly stays on Opus 5.5 and switches only when he has unused tokens on another subscription . Where he does use several models is review. For hairy changes, his Council skill has models from different families review the work. His reasoning is that Anthropic's 5.5 models are distilled from the same base, so asking Opus and its siblings to review one another is "like asking the same person at different times of the day" .

How he runs it:

  • He gives a prompt like fix X, Y, and Z. Full stop with council 9.5. "Full stop" tells the agent to do everything it can to reach a PR it is confident can be merged, then stop before merging. "Council 9.5" means the work isn't done until every reviewer scores it at least 9.5/10 .
  • T3 Code doesn't depend on any one harness, so one harness can call another. Opus starts reviewers: Claude models in Claude Code, OpenAI models in Codex, and Muse in its own harness. Scores go at the bottom of the PR .
  • The reviewers aren't just reading a diff. They can run and test the code on the same machine and ask the parent agent questions .
  • You can change what the score measures. He has asked for scores on "how likely is this to piss Theo off," and on whether a PR is what a specific user, Corey Quinn, would want. He also tells agents they may scrap the PR, because "agents are very scared" to delete code .

His results: across 144 reviewer scores, Opus's own rating of its changes rose from an average of 7.4 to 8.9 after two or more rounds. He notes that even reviewing with Opus alone would probably improve the code . Bugs it caught include three security holes in T3 Code's guest-share feature, such as the browser cache still serving visuals after a share was revoked . It also flagged test deletions that removed guards against old bugs . He estimates about an 80% chance he can merge a PR his threads open .

Caveats: in a few runs the council never agreed. One went 50+ loops, started 600+ subagents, and grew a PR of about 1,000 lines to more than 20,000. He has since added a run cap and an instruction not to let repeated reviews cause scope creep. He says Opus respects it, but OpenAI models keep going after every small nit . He also lowered reviewer effort from extra-high to high . He uses up Muse's weekly limit on the $200 plan in two to three days, and says not to use the skill if you pay API prices . The skills are now in a GitHub repo linked in the video description .

His sponsor segment has a related idea. Macroscope lets T3 Code keep a folder of custom review agents in the repo, scoped by file type. One holds the maintainer's list of things that annoy him in agents' Effect code. Another stops agents from creating UI components that already exist .

Yegge: pace agents to human speed, and think twice before a factory

Steve Yegge summed up his conference talks in a post. His agents have overwhelmed his game's players with features, his accountant with emails, his merge queue with tickets and his machines with load. They have also "launched broken OSS without permission" . He says you probably don't need a software factory: he has had to throw away five because rewriting them became cheaper than fixing them, and keeping one running takes real money and leadership . His advice for now is to build your own agentic chief of staff and get good at working with it .

Kent C. Dodds offers two prompts for those who do build a factory:

  • Ask agents to complain about what they hate about working in your codebase .
  • Before sending any prompt, ask "how could I have made it so I don't have to send this myself and still get the same desired outcome?"

Running agents on a home machine, and the credentials problem

From a New York hotel, Dodds used Kody to install T3 Code on his home NAS. That lets him start Claude Code agents while his laptop is off, without Tailscale . He reported it "worked," though a screenshot shows Docker Hub still failing with a certificate error . Self-hosting Kody is free for personal, nonprofit, noncommercial and classroom use, and comes with a 30-day evaluation, on Cloudflare or via Docker .

A related open question: Marc Köhlbrugge asked how to give coding agents credentials securely, noting that 1Password gives an agent access to the full vault every time . Dodds answered that this is the problem he built Kody for, but didn't explain how Kody handles it .

Model and tool updates

  • Antigravity: the context window doubled for Gemini Argon, Carbon and Hydrogen. The team says compaction should drop 3.5x per user turn . Nicholas Moy, from the Antigravity team, says Carbon is still "only ok" at swarms and wants reports on where it falls short .
  • Codex: Tibo says the team "might have actually fixed" the "model at capacity" errors .
  • Fable 5.2 vs. Opus 5.5: ThePrimeagen says his work is too trivial to tell Fable 5.2 from Opus 5.5 high . That fits last period's advice to keep Opus as the default.
  • Agent UI: Theo says kanban-style agent dashboards waste over half the screen on in-progress work . He adds that Vibe Kanban, which he invested in, was moving toward a T3 Code-style layout before it shut down .
Theo's "Council" skill: models from different families must score a change before the agent stops
swyx
  • Steve Yegge advises pacing agent output to human and operational capacity: agents overwhelmed players with features, an accountant with emails, the merge queue with tickets, and machines with load; they also launched broken open-source software without permission. He says the resulting disruption made him learn to slow down.
  • Build a software factory only if you truly need one: Yegge says factories require substantial upkeep and curation, and he scrapped five because rewriting them became cheaper than fixing them. Although Claude can build them, they tend to become bespoke to a domain and require ongoing money and human leadership.
I just finished talks about my software factories at two conferences. Afterwards a bunch of people came up and told me what they were sur…
Theo - t3․gg
  • For risky or “hairy” changes, Theo uses a Council skill rather than routing each task to a supposedly best model: a primary coding agent brings in reviewers from different model families, whose distinct approaches can surface novel issues. He scores reviews from 1–10 and can require a threshold—for example, full stop with council 9.5 means the work is not done until every reviewer scores it at least 9.5.
  • Make cross-harness review actionable: Theo’s T3 Code setup lets the primary harness invoke Claude, OpenAI, and Muse agents in their respective harnesses; reviewers can run and test the code, ask the parent agent questions, and discuss findings. He records reviewer scores on the PR.
  • Bound the review loop and guard against scope creep. Theo reports runs exceeding 50 loops and 600 subagents, with a roughly 1,000-line PR growing beyond 20,000 lines; he recommends a run cap and stopping when reviews circle without progress. The workflow is token-intensive—he says it exhausted his Muse weekly limit in two to three days on a $200 plan—and he advises against it at API prices.
Stop Searching For the Perfect Model
Kent C. Dodds 🐨

Kent C. Dodds used Kody from a hotel to install T3 Code on his home NAS so he could run Claude Code agents while his laptop was off; he later reported that it worked. The setup message still noted a Docker Hub certificate error.

I'm in a hotel in New York and I'm using Kody to install T3 code on my NAS at home so I can spin up Claude Code agents even when my lapto… worked 💪 ![A list of permissions for a T3 Connect environment.](https://pbs.twimg.com/media/HXOLeF7XsAIiVuk.jpg) ![A screenshot of a text…
Nicholas Moy

Nicholas Moy shared a creative-coding demo whose creator described it as an Antigravity project using a “Gemini 4 Carbon with Hydrogen swarm”: a JavaScript-only pelican-on-a-bike animation that moves between art styles and emotions, with music and sound also coded. The creator also said their Pro subscription limit was reached.

Gemini 4 Carbon with Hydrogen swarm < best thing ever you can do with Antigravity > This art piece is pure code , JS only pelican r… I never thought the pelican test would come this far [https://x.com/chetaslua/status/2121885449509851415](https://x.com/chetaslua/status/…
Theo - t3.gg

Theo argues coding-agent interfaces should prioritize active work: he criticizes Kanban-style layouts for using over half the screen for in-progress work and considers them poorly optimized for productivity, in contrast to T3 Code’s UI. This is his product-design opinion, not a measured comparison. He also says that Vibe Kanban, where he was an investor and contributor, began moving toward a UI similar to T3 Code’s before the company shut down; he does not establish that the UI contributed to the closure.

Every single time someone says T3 Code isn’t “optimal for real work” they show the most useless 2025-era kanban slop that wastes over hal… Fun fact: I was an investor in and contributor to Vibe Kanban. They realized kanbans were dead and started moving to basically the exact …
Kent C. Dodds 🐨

To run Claude Code agents while his laptop is off, Kent C. Dodds uses Kody to install T3 code on a NAS at home . He says Tailscale is not required for this setup—just Kody .

I'm in a hotel in New York and I'm using Kody to install T3 code on my NAS at home so I can spin up Claude Code agents even when my lapto… Oh, and tailscale's not needed for this BTW. Just Kody. [https://x.com/kentcdodds/status/2122186583893815741](https://x.com/kentcdodds/st…
ThePrimeagen

ThePrimeagen says his work is so trivial that he cannot tell a difference between Fable 5.2 and Opus 5.5 High, a task-specific observation rather than a general model comparison.

All of my work is so trivial that I cannot tell a difference between fable 5.2 and opus 5.5 high
Kent C. Dodds 🐨

Kent C. Dodds amplified a post asserting that 63% of developers say AI is eroding their coding skills and asking whether it matters, responding “LOUDER FOR THE PEOPLE IN THE BACK”—a signal of concern about AI’s effect on developers’ skills, not a practical workflow tip.

63% of developers say AI is eroding their coding skills. Does it matter? [![Video](https://pbs.twimg.com/amplify_video_thumb/212165842455… LOUDER FOR THE PEOPLE IN THE BACK. [https://x.com/stolinski/status/2121658494466265524](https://x.com/stolinski/status/2121658494466265524)
Kent C. Dodds 🐨

Kent C. Dodds says Kody is ready for teams and targets the increasingly mainstream problem of securely and usefully providing credentials to coding agents; the linked discussion highlights the risk that 1Password gives an agent access to the entire vault, but the post does not explain how Kody handles credentials.

I love that the problem I've built Kody for is starting to become mainstream. Kody is ready for you and your team. [https://x.com/marckoh… Has anyone found a secure and usable way to provide their coding agents with credentials? I’m using 1Password but it gives an agent acces…
Kent C. Dodds 🐨

When starting an agent session or sending a prompt, ask how to get the same desired outcome without having to send it manually; use that question to identify opportunities to automate work into a “software factory.”

Next time you spin up an agent session or send a prompt to an existing one, all yourself: "how could I have made it so I don't have to se…
Nicholas Moy

Nicholas Moy called Terminal-Bench 4’s top placement among benchmarks respondents would trust to indicate frontier-model capability “kind of sus,” but gave no reason . The poll reported 25 votes for Terminal-Bench 4 from 56 respondents, while the most common comment was that people did not really trust benchmarks .

TB4 being top here kind of sus [https://x.com/chrisbarber/status/2122020768523714827](https://x.com/chrisbarber/status/2122020768523714827) I asked 56 people in AI which benchmarks they'd trust most to indicate if an LLM is frontier: 25 votes: Terminal-Bench 4 (@alexgshaw, [@a…
Nicholas Moy

Nicholas Moy says Carbon is only “ok” at swarms and asks users where it currently falls short; his team is working to improve its swarm capabilities .

Even though swarms are powerful I still feel like Carbon is only ok at them. What are people’s experiences with it? Where does it fall do…
Kent C. Dodds 🐨

Kent C. Dodds describes the shift in software work as moving from human-staffed “software factories” to people managing the factory while agents work on the floor—a high-level framing of human-agent collaboration, not a specific workflow recipe.

Kent C. Dodds (@kentcdodds) on the buzzword of AI Engineer: "10 years ago we had software factories. It was just that the workers on the …
Kent C. Dodds 🐨

Kent C. Dodds recommends having coding agents identify what they dislike about working in your product or codebase as a way to improve a “software factory.”

The secret to the software factory is getting agents to complain about what they hate about working in your product/codebase.
DHH

DHH says Omarchy installs on Nvidia N1x in 18 seconds and offers a local model for download seconds later, presenting fast, low-friction environment setup as achievable; the post does not identify the model or say it has already been installed or run.

Omarchy now installs in just 18 seconds on the [@nvidia](https://x.com/nvidia) N1x! And a few seconds later, you're being offered to down…
Nicholas Moy

Antigravity doubled the context windows for Argon, Carbon, and Hydrogen; the announcement says users should see 3.5× less compaction per user turn.

We’re doubling the context window for Argon, Carbon, and Hydrogen in Antigravity. Lots of work to make this happen but you should now see…
Tibo

The Codex team may have fixed the recurring “model at capacity” errors, but the post presents this as a possibility rather than a confirmed resolution .

Codex team might have actually fixed the "model at capacity" errors that have been pestering you. ![](https://pbs.twimg.com/media/HXOX68A…