An AI plan found two watch listings

Grex is a system for assigning work to AI teams and checking their results. Today, its watch-search team planned and executed a search without a person directing those steps.

The team was looking for watches that might be worth buying for resale. It found two listings. An independent check confirmed one as a watch; the other remained unverified because the marketplace blocked access.

The team chose its own search strategy

Grex runs this work as a mission: an assignment with a target, limits, and evidence requirements. Its planner turns that assignment into instructions for the agents doing the search.

At 3:07 p.m. Central, the planner wrote fourteen instructions. One tactic was to search for misspellings of a brand name. A seller’s typo can make a listing harder for other buyers to find.

The new plan replaced the human-written starting plan before it expired. The next search ran at 3:30 p.m. and produced the two listings.

One checked result, one unanswered question

The checker confirmed one listing’s product category from the marketplace’s navigation trail. It could not inspect the second listing, so Grex recorded that limitation.

This was a complete planning-and-search sequence without human intervention. It was not proof of two profitable purchases. One listing passed an identity check, and the other could not be checked.

The next planned test was a different capability on a second computer. That would test whether the same mission process worked beyond watch searches.