Case 04 · AI product

Teams of AI agents, delivering client work.

Role
Product lead
Organisation
A global services group
Platform
In-house multi-agent system
Status
Delivering client work
Controls
Governance · Provenance · Security & audit logs

How it works, in three shots

A brief comes in, inside set budgets, permissions and approved models.

Agent teams do the work. Every output is checked for IP and lookalikes.

The client gets the output. Every step is logged and auditable.

Illustrative diagram of how the platform works.

The problem

The question wasn’t whether AI could make assets. It was whether a services business could turn that into repeatable client delivery: quality it could stand behind, spend it could control, and work that’s safe to hand to a client.

What I did

  1. a Owned the product: what the platform does, for whom, and in what order.
  2. b Co-led with engineering and built hands-on, prototyping new agent teams myself.
  3. c Specialist agent teams for art, 3D, game dev and presentation work, plus a builder for creating new teams.
  4. d Made control part of the product: governance, provenance checks, and security and audit logs, built into the platform rather than bolted on.
  5. e Pilots and go-to-market: it now delivers client work, from pitches and vertical slices to full products.

The control layer

AI work only sells if the client can trust it. So every brief runs inside four controls.

Governance

Who can run what, with which tools and models, within what budget.

  • Spend controls
  • User permissions
  • Tool & LLM access

Provenance

Every output is checked before it reaches a client: for IP infringement, lookalikes of existing work, and more.

  • IP infringement
  • Lookalikes
  • + more checks

Security logs

Every access, tool call and model call is recorded, including what was blocked.

  • Access
  • Tool calls
  • Model calls

Audit logs

A traceable record for every output: who briefed it, which agents and models worked on it, and what was approved.

  • Who ran it
  • Which models
  • What was approved

Where it stands

Live

delivering client work today

Outputs

pitches, vertical slices and full products; clients get the work, not the tooling

Next

regional teams running it with clients, on revenue-generating projects

Outside games

Any firm that sells deliverables, from proposals and designs to reports and campaigns, can work this way: agent teams do the heavy lifting, people own the result, and a control layer keeps spend, IP and access in check.

The judgment call

“Sell the output, not the tool. Clients pay for results they can use.”

Next case · 05 An internal knowledge assistant