A rover 20 light-minutes from Earth can't wait for an engineer when its agent fails, so our harness repairs itself. An Explorer agent (Claude Haiku 4.5) drives a rover across a simulated planet with sand traps, dust storms and crater edges. When it has an incident, an Engineer agent (Claude Sonnet 5) reads the telemetry and the Explorer's own reasoning, diagnoses the root cause, and proposes a typed patch to the harness: rules, code-enforced guardrails, tool access and memory policy. The patch is kept only if it passes an immutable constitution and beats the current version on held-out planets it never trained on. Unsafe or unlucky patches are rejected in public: in our runs the constitution rejected a patch that raised the average score but killed a rover the baseline had kept alive. MongoDB Atlas holds the whole loop. Change streams wake the Engineer on each live incident, Atlas Vector Search with Voyage AI embeddings recalls past lessons, $geoNear finds nearby hazards, a time-series collection stores telemetry, and versioned documents hold the harness history, including every rejected patch.
More from The Harness Engineering & Model Wrangling Hackathon
Ripple designs circuit boards from plain English, grades them against hidden checks, and evolves its own harness from the failures, going from 0/4 to 3/4 on…
+1Offload learns the work you repeat, so next time you can hand it over. Its engine, REM (Replay · Evolve · Merge), is an agent harness with a day/night cycle…
Recursive Computer Use is a local-first desktop automation harness that enables AI agents to execute tasks on your computer, verify outcomes locally, and…
Da Vinci is a recursive improvement CAD harness designed to support physical engineering tasks through specialized AI agents, programmatic CAD and automated…
CyclopsDiary gives factory workers wearing Meta AI glasses a shared memory of the floor. Every worker's glasses feed one MongoDB memory that keeps what was…
A harness that teaches itself what context to show Jev, TypeSafe's fast decision model, proves the result on a storm it never saw, and can't cheat because…
