CanItWork turns a plain-English description of a system — a circuit, a web service, a hospital, a drone fleet — into an interactive model you can run, break, and repair. AI agents propose the design and the fixes, but a deterministic simulator decides: a design that fails its checks is refused, and a fix is kept only if a replay shows it works. Built at SharkByte 2026 with a team of four, where it won MLH Best Use of Vultr and the Best Use of Prune API challenge.
Thirteen kinds of system, each with its own deterministic simulator and 3D view — from breadboard circuits to databases, power grids, and hospitals.
Recovery council of agents proposes fixes after a failure; the simulator replays each one and ranks them by measured result and added monthly cost.
Shared runs replay from a ~2 KB link and verify a SHA-256 of the result, anchored on Solana so anyone can check it.