AutoWebWorld: Synthesizing Infinite Verifiable Web Environments via Finite State Machines

The performance of autonomous Web GUI agents heavily relies on the quality and quantity of their training data. However, a fundamental bottleneck persists: collecting interaction trajectories from real-world websites is expensive and difficult to verify. The underlying state transitions are hidden, leading to reliance on inconsistent and costly external verifiers (e.g., human or LLM judges) to evaluate step-level correctness. To address this, we propose AutoWebWorld, a novel framework for synthesizing controllable and verifiable web environments by modeling them as Finite State Machines (FSMs) and use coding agents to translate FSMs into interactive websites. Unlike real websites, where state transitions are implicit, AutoWebWorld explicitly defines all states, actions, and transition rules. This enables programmatic verification: action correctness is checked against predefined rules, and task success is confirmed by reaching a goal state in the FSM graph. AutoWebWorld enables a fully automated search-and-verify pipeline, generating over 11,663 verified trajectories from 29 diverse web environments at only \$0.04 per trajectory. Training on this synthetic data significantly boosts real-world performance. Our 7B Web GUI agent achieves state-of-the-art on WebVoyager, outperforming all baselines within 15 steps. Furthermore, we observe a clear scaling law: as the synthetic data volume increases, performance on WebVoyager and Online-Mind2Web consistently improves.

AI assistants that can use websites for you—booking a flight, filling a form, navigating an online store—need to practice on huge numbers of example tasks to get good. But gathering those examples is expensive and unreliable: when an AI clicks a button on a real website, no one can see what is actually happening behind the screen, so a human or another AI has to guess whether each step was done correctly. These guesses are slow, costly, and often disagree with one another. We take a different approach: instead of collecting examples from real websites, we build our own. We first write down the complete rulebook of how a website behaves—every page, every button, and exactly what each action does—and then automatically generate a working website from that rulebook. Because we wrote the rules ourselves, we always know the correct outcome of every action, so the examples verify themselves with no human checking needed. Using this method we created 29 websites and over 11,000 verified practice tasks at a tiny fraction of the usual cost. AI agents trained on our synthetic examples performed better on real websites than agents trained on far larger real-world datasets, and kept improving as we generated more data—showing that carefully built, self-verifying practice can teach AI to act on the real web reliably and cheaply.