Newshot

Show HN: WorldFoundry – AI agents need experience, not just training data

Original text(via HN AI)

AI agents can say “done.”Can they prove it?

A production outage. An unexpected state change. An independent evaluator. See what a superficially successful workflow misses—then take control of the real WorldFoundry simulation yourself.

The ticket must be critical, assigned to on-call, acknowledged with both “outage” and “incident,” and resolved. In this scenario, a separate system resets its priority during the workflow.

You're in control of the environment

Choose actions in any order. WorldFoundry updates the simulated ticket after every action and automatically checks all five requirements. No Python or API key required.

A concurrent overwrite occurs after your third successful action in Scenario 02. If it changes something important, act again before submitting.

Evaluation

WorldFoundry action trace

A green final answer isn't proof.

The environment stores what actually changed. Real actions, hidden state mutations, and independent checks expose gaps that a confident explanation can conceal.

Now test your own agent.

The WorldFoundry Python SDK accepts custom environments and agent adapters. Use the documented local quickstart to connect your own agent.

This browser experiment runs WorldFoundry v0.8.0 simulation and evaluation code. It does not currently execute an autonomous LLM. The reference run is intentionally scripted, and the interactive run uses visitor-selected tool actions. Model training/fine-tuning is not performed.

The text above was extracted automatically from the source page and may miss images, tables or code blocks. The original is authoritative.

Published by HN AI
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗