Superprocess automates business processes that need judgment. AI agents, business rules, system integrations and human approvers run inside one governed workflow, and every run leaves a history you can replay. I’m a co-founder.
The idea comes from a pattern we kept seeing. An agent demo goes well, then stalls when it reaches an operations team, because the demo never had to answer the questions an operation runs on. Who approves this, and what do they see? What happens when step six of nine fails at 2 a.m.? Six months from now, why did it decide that? What happens when the policy changes? There’s more in What breaks when an agent meets an operation.
We give each part one job. The process, described in BPMN, owns the work. Policy lives in rules rather than in prompts. An agent is a step inside the process: each one gets the model that suits it and only the tools that step allows, and model providers can be swapped. Execution is durable and backed by Temporal, so retries, failures and long waits resume from the last completed step. A process can wait days for a document or a person without losing its place. More on that in What a durable process does while nothing happens.
People retain authority. Routine work runs within policy limits, and material or uncertain exceptions pause for a named owner with the case attached. Autonomy grows in steps: assist, then recommend, then act within limits. Model calls, tool use, approvals, costs and system updates all land on one trace. More on that in People retain authority.
We’ve completed a pilot and production implementation is underway. The processes we work on include invoice factoring review, order-to-cash exceptions, and demand and supply planning.