Steal This Move
Short answer: NASA’s Perseverance team let Claude plan two Mars drives. The route went through the same simulator as any human plan, then drivers checked it against ground images the AI never saw and refined one stretch. Organizations can copy both: no more lenient gate for AI work, and a named check for the context it was missing.
What did NASA do?
For two drives on Mars, the Jet Propulsion Laboratory let Claude plan Perseverance’s route from orbital imagery and slope data. Claude marked hazards, set the waypoints and revised its own path. Engineers then sent the commands through the rover’s digital twin, the same simulation every human-planned drive passes, which checks more than 500,000 variables. Rover drivers also reviewed the route against ground-level camera images Claude had not seen, and refined one stretch beside a band of sand ripples. JPL expects the approach to cut planning time roughly in half.
Why does this matter for organizations?
Two failure modes are common when AI drafts real work. One is a more lenient review because the output looks finished. The other is a review that checks the AI against the same information it already had, and so confirms its own blind spots. NASA avoided both: the plan faced the standard gate, and it faced the one source the model could not have used.
Every team has a ground-level view: the customer history in someone’s head, the policy change not yet written down, the constraint that came up in a hallway. That is where an AI plan is most likely to be wrong.
What should leaders put in place?
One gate for all plans. AI-drafted plans go through the review a person’s plan would get. No fast lane because it came back polished.
Name the missing view. The reviewer writes down two or three things they know that the AI did not, and checks the plan against those first.
Record what the check changed. When the review alters an AI plan, note what and why. Over time that list shows what context to hand the AI up front.
The goal is faster planning that is still safe to send. NASA got both by keeping the last look with the people who could see the ground.
Checks like this hold when a team practices them together. Start with a guided session →
Keep reading: The Formula 1 tire call nobody took away · Use AI to find the expert, not to replace one
This is the thinking behind a Trusted AI Culture.
Sources: NASA Jet Propulsion Laboratory, January 30, 2026. Anthropic, "Claude on Mars," January 2026. The Register, January 31, 2026.
Written with AI in the loop: my idea, AI drafted and sharpened, my judgment on the way out. Every word is mine to stand behind. — Darren