Let a team of agents handle a business process from start to finish. The agents read, match and draft, and a person approves. A run that lasts longer than one session continues where it paused. You see what each stage costs while the run is going.
In use
Agent teams with a shared ledger run on Microsoft Foundry, and every stage of a run writes down its cost.
The Agentic Part
In this team every agent has a role: one reads, one matches, one drafts. A manager agent hands out the tasks to specialised agents as the situation requires, and other patterns let agents work one after another, in parallel, or by passing control on depending on context. All agents write their progress into a shared ledger, a common record in which every member sees the current state and which keeps the progress when a run lasts longer than one session. The agents run on Microsoft Foundry, Microsoft’s platform for agents in production. Writes to your data go through an approval step: the agents draft, a person confirms.
Main Features
- Specialised agents that each take one part of the process
- Nothing is saved until someone has reviewed the proposal
- Runs continue exactly where they paused
- The cost of every stage is visible live while the run is going
- Every run leaves an entry that explains the result later
- The interface you use today becomes, unchanged, the tool an agent calls
- Article 50 disclosure and an audit trail are built in from the start
Use Cases
- The monthly Excel worklist is in the office anyway. The agents match the buildings and propose a full month of technician routes grouped by location, with real driving times from Google Maps. Saving happens once someone has reviewed the proposal.
- A document arrives by mail. An agent reads it and creates a record with a review flag, and the original stays archived until a person confirms the record.
- A run takes longer than a working day. The ledger keeps the state, and the run continues after the pause where it stopped.
What we build for you
- The process the team of agents takes over, and the people who approve its results
- The programs and interfaces the agents access
- The roles of the agents in the team
- The model approved for each use case
- The people who may view the cost per stage
