← All workAI engineering

AI agent orchestration

In short

  • WhatTeams ask questions of their own data in plain language.
  • HowAgents work within the same permissions as the person asking.
  • ResultAnswers you can check; decisions stay with people.

How a question is answered

How a question is answeredA question in plain language goes to the agent, which works within the asker's permissions and reads the data without changing it. It returns an answer with its evidence, and a person makes the decision.A questionin plain languageThe agentworks within yourpermissionsYour dataread, never changedAn answerwith its evidenceA person decidesHow a question is answeredA question in plain language goes to the agent, which works within the asker's permissions and reads the data without changing it. It returns an answer with its evidence, and a person makes the decision.A questionin plain languageThe agentworks within your permissionsYour dataread, never changedAn answerwith its evidenceA person decides

Agents propose. People decide.

The agent does the searching and the drafting; a person makes the call.

What it delivers

  • Evidence with every answerEach figure traces back to the record it came from.
  • Within your permissionsThe agent sees only what its user is allowed to see.
  • People decideAgents suggest; anything consequential waits for a person.
  • On the recordWhat was asked and what was done is logged.

Principles

  • Code calculates the figures; the model explains them.
  • If the data cannot answer a question, the agent says so.
  • Nothing consequential happens without a person’s approval.

The engineering detail behind this work stays private: it belongs to the organisations I build for.

Want to talk about this work? Email me or download my CV.