Load Path lets several AI agents plan and fund bridge repairs under a fixed budget, and makes every action auditable. Each tool call, by an agent or a person, passes one policy gate before it runs: role, spend limit, budget envelope, jurisdiction, approved plan, and a reason on rejections. The decision, actor and policy version are written to a SHA-256 hash-chained ledger in Atlas. A MILP picks treatments for 198 Ontario bridges within a 250M envelope. When an owner rejects a condition reading, only the work that depended on it is reverted, newest first, an inspection is requested and the plan is re-solved. Stop rolls back the whole session. Data: Ontario bridge conditions, Open Government Licence – Ontario.
Team
