[ An experiment ]
Pull the plug.
The same task. The same rules.
One keeps what it learns.
Let them explore. Then cut the power.
Keeps its memory
An unfamiliar space. A place to begin.
Starts again
The same space. The same starting point.
They will explore at the same speed. You decide when to interrupt them.
[ After the interruption ]
What should
carry on?
A lost connection. A restarted machine. Someone else taking over. The work already done still matters.
This small experiment makes one part visible: remembering where a system has been changes what it has to repeat. In real work, useful memory also needs to preserve decisions, unfinished tasks and what still needs checking.
What is actually happening?
Two copies of the same search algorithm explore an identical map. There is no AI model running here. The movement, saved progress and recovery are computed in your browser.
When you cut the power, both searches stop. On return, one reads its saved map and position. The other has to explore from the start. Faint lines on that side show the earlier run for comparison; the search itself cannot use them.
The example demonstrates saved state, not human judgment, reliable AI reasoning or a production deployment.
Your saved run
Starting the experiment saves its map, route and progress in this browser. The map and progress are not sent to us. Refresh the page to try a real return.
No run saved yet.
[ In your operation ]
Work that can continue.
Some work outlasts a session. Field records, unfinished tasks and workflows across tools need a way to continue when a device restarts or another person takes over.
We can build saved state, recovery checks and handover into the commissioned system, with people able to inspect what is being resumed.
Discuss your systemAnother question to explore
Cut the connection.