Show me. Switch anytime. Your progress stays with you.
Switch anytime. Your progress stays with you.
Explore what AI can do.
Try a challenge. Change the outcome. See how it works.
I build AI workflows, interactive tools and creative experiments. Try them here, then explore the decisions behind them.
AI LabCatch AI mistakes
Backdrop: rendered in Blender for this site
Does the taskWaiting
Waiting for a request
Checks the workWaiting
Not checked yet
Spots problemsWaiting
Watching for stalls
Event replayNo run yetInject fault: available at Intermediate. Compare runs: available at Intermediate.
Try it. Then see how I built it.
Every experience has all three levels.
Catch AI mistakes
Spot it. Change it. Test it.
Keep orders moving
Move resources. See what changes.
Direct a scene
Move the camera. Shape the mood.
How the Forge works
Every experiment here runs the same loop I use to build with AI. One loop, three places: an email draft, a floor with three queues, and a fight scene.
1
Task
Does the task (builder)
The request is written down first, with what 'done' means. Nothing starts from a vague ask.
2
Attempt
Does the task (builder)
One builder makes one attempt and owns every file it touches. Its own 'it passed' is noted, not trusted.
3
Check
Checks the work (verifier)
A different checker compares the attempt with the ORIGINAL request, field by field. Only this can say Verified.
4
Retry or report
Bounded
A failed check gets a limited number of repairs. When they run out, the report says Failed and what is still wrong.
5
Decide
A human gate
Nothing is installed or published until a person approves. In this site, that person is you.
And one more seat: Spots problems (watcher)
The watcher reports progress, stalls and idle time. It never proves anything is correct: a process that looks busy for ten minutes but produces nothing is a stall, not progress. Try the "Attempt stalls" fault in AI Lab Advanced.