Book a free call
TrainingProductsTeamSupportContactLabBlog Book a free call
Lab build · November 15, 2026

AI Agent Lab: Critic

Use an AI Critic Agent to Fix the Problems the AI Builder Missed

We built two versions of an interactive 3D world using the same prompt and the same model (Claude Fable 5.2). The goal was a late-night trading desk you can fly into and tour, all the way down to the computer's motherboard and graphics card. One AI Builder just checked its own work and called it done in about an hour. We asked the other AI Builder to use new, blind AI Critic Agents to review every round without any prior knowledge, and after 26 rounds over almost five hours, we were amazed at the difference!

Compare two "finished" 3D worlds built by:AI Builder Only vs. AI Builder with Critic Agents

  1. Pick a shot
  2. Drag over either picture (both magnifiers move together)
  3. See what the Judge Agent found
Shot
AI Builder Only11 self-checks
2.6× zoom
AI Builder with Critic Agents26 Critic Agent rounds
2.6× zoom

AI Builder Only

AI Builder with Critic Agents

Six shots from each final 3D world, scored by the Judge Agent

We asked a new Critic Agent to judge both finished builds. We call it the Judge Agent. It was never told which build it was looking at, and three AI models each ran three independent reviews of each build. The wide shots of the office favored the AI Builder Only version. Inside the computer, where the fine detail lives, the AI Builder with Critic Agents won in every shot.

How the Critic Agent loop works

Two agents with two jobs. The AI Builder writes and fixes. The Critic Agent only looks, and a brand new Critic Agent starts every round.

  1. The AI Builder renders

    Claude Code builds the world and renders six fixed shots, close-ups and frames from the tour.

  2. A new Critic Agent looks

    It sees only the pictures. No code, no notes, no earlier reviews.

  3. It ranks what's wrong

    Every problem gets a label (Critical, Major, Moderate or Minor) and a place in the frame.

  4. The AI Builder fixes, repeat

    26 rounds, until what was left were judgment calls for a person to make.

Why the Critic Agents work blind

We replayed 23 rounds of the build with two kinds of Critic Agent looking at the same pictures. One saw only the pictures. The other also read the AI Builder's notes on what it had just fixed.

3.7major problems per review from a blind Critic Agent
2.6from a Critic Agent that read the AI Builder's notes

The Critic Agent that read the notes listed more problems overall (about 19 per review against 13), yet rated fewer of them major. Once it knew the AI Builder had "fixed" something, it went easy on it, so a loop run that way would likely have called the world done much sooner.

See both final worlds in motion

AI Builder with Critic Agents: fly it yourselfinteractive

Checking your graphics card…

AI Builder with Critic Agents: final guided tourvideo
AI Builder Only: final guided tourvideo

What this means for your AI work

An AI Builder calls its own work done early.

It stopped after 11 self-checks. From across the room the result looked finished. Up close it wasn't.

A new Critic Agent sees what the AI Builder stopped seeing.

With no backstory and only the result to judge, Critic Agents kept finding real problems for 26 rounds.

A person makes the final call.

Critic Agents also push changes a person might not want. Over 26 rounds the tower view ended up darker than in earlier rounds, and the Judge Agent rated a major problem in it in every review. Let AI find the problems. You decide what good looks like.

Get more free Lab builds. A new one most weekends, straight to your inbox.

Let us build Critic Agents into your team's AI work

An AI Builder with Critic Agents works for reports, proposals, spreadsheets, code and designs, not just 3D worlds. We set up the workflow in the AI tools your team already uses, then teach your team to run it themselves.

Critic Agentsfor the work you already hand to AI
Hands-on trainingwith your team's real work
In the tools you useClaude, ChatGPT, Copilot, Gemini or Grok
Human sign-off built inAI checks, people decide
Book a free call

About this build

Built November 14 and 15, 2026 with Claude Fable 5.2 in Claude Code on our own PC. Both builds started from the same build prompt. For the AI Builder with Critic Agents, every round a brand new Critic Agent saw only that round's renders (six fixed shots, close-up crops and frames from the tour) with the same fixed brief, and returned problems ranked Critical, Major, Moderate or Minor. The AI Builder fixed them and rendered again, 26 times over almost five hours. The AI Builder Only had no Critic Agent: it rendered and looked at its own shots 11 times and decided it was done after about an hour. The pictures on this page are the final round of each build.

To judge the finished builds the same way, we gave the Judge Agent (a new Critic Agent with the build's own brief) the six final shots of one build at a time, unlabeled: Claude Fable 5.2, Claude Opus 5.5 and Claude Sonnet 5.5, three reviews of each build per model, nine per build and 18 in all. The percentages are the share of a build's nine reviews that rated a shot's worst problem Major or Critical. Every quote is the Judge Agent's own words: under a bar above 0%, from a review that rated that shot Major or Critical; under a 0% bar, a Moderate note. The replay test re-ran rounds 2 to 24 of the build with a blind Critic Agent, a Critic Agent that also read the AI Builder's notes, and the AI Builder itself, 207 reviews. A planted-flaw test of 252 runs checked that every reviewer catches planted glitches either way, and they did: about 95% caught in every setup, and 92 to 100% of the obvious ones. Severity is the Critic Agent's own label, and the six shots are listed from the biggest win for the AI Builder with Critic Agents to the two it lost; all six are on the page. The tour videos show the same two final builds.

After the experiment we added furniture, on-screen controls and faster first loading to the live AI Builder with Critic Agents world. Those changes are not part of the comparison. More builds in the Lab

Sunlight shining down through deep blue water

Find light in the deep waters of AI

Let's talk! A short call will help you learn more about our training and consulting solutions. We can also discuss whether our existing licensed tools can help, or whether you need a custom build.

Book a free call