AI agents

Funanga's AI agents take over document-heavy and repetitive knowledge work. An agent sits at a defined point in an existing process, picks up the work that arrives there, and hands its result back to the system that owns the record. The agent works alongside the systems a company already runs.

Every agent runs inside a control frame. A person owns the review step and can approve, correct or reject. Output is evaluated against test cases, guardrails limit what an agent may do, and a fallback hands the case back to a person. Each run leaves a full trace of the steps taken and the sources used. Quality is measured, not assumed.

  • Intake - reading incoming documents, mail and attachments
  • Classification - sorting each item by type and routing it to the right queue or owner
  • Extraction - pulling named fields out of invoices, contracts, forms and letters
  • Matching - checking extracted records against your system of record
  • Drafting - preparing replies, notes and summaries for a person to approve
  • Answering - responding to questions against your own content
  • Handover - passing unclear cases to a named person with the reasoning attached
The loop around an agentAn agent is defined against a process, connected to the data it needs, run as a step, reviewed by a person where a decision needs an owner, then evaluated and versioned. What the evaluation shows feeds back into the process definition.THE LOOP AROUND AN AGENTDefine the processConnect the dataAgent runs the stepHuman reviewsEvaluate and version

A first project starts small. We pick one process, agree what good output looks like, connect the systems the agent needs to read and write, and run it next to the current way of working until the results hold up. You keep your data and your systems of record.

Write to commercial@funanga.com with the process you have in mind. We will come back with a scope and the next step. For the software the agents run on, see Platform.