AI Velocity Pods

An AI Velocity Pod is how Ailoitte delivers agentic AI: a small senior team, a fixed price, and an outcome that is written down before the price is. It is the delivery model behind Agentic in Production, and it exists because the usual model, a large team billed by the hour, does not survive contact with an agent.

Why time-and-materials fails for agents

Hourly billing assumes that effort and progress move together. With agents they do not. A week of prompt tuning can produce nothing measurable, and a single afternoon spent writing the refusal list can be the most valuable work on the project. When the client pays for hours, they pay for the first and cannot see the second.

Hourly billing also removes the pressure to define “done”. An agent can always be a little better. Without a fixed outcome, the project drifts into an endless pilot, which is the most expensive way to not ship.

Outcome-based engagement

An engagement whose scope is a measurable result for the client, such as an agent passing a named evaluation set inside a stated budget, rather than a quantity of effort.

What a pod actually commits to

A pod commits to three artefacts and one number.

The evaluation set: a fixed collection of real cases, with expected outcomes, that the agent must pass at an agreed rate. The client provides the cases; the pod turns them into a test that runs on every change.

The refusal list: the actions the agent must never take and the requests it must decline, enforced in code. This is written with the client’s risk and compliance people in the first weeks, not the last.

The runbook: who owns the agent, how to switch it off, what the traces look like, what the monthly budget is and what happens when it is hit.

The number: the price. It is fixed because the three artefacts define the work precisely enough to price it. If the client changes the eval set or the refusal list, that is a scope change, and it is priced as one.

Why small and senior

Agent work rewards judgement more than volume. The hard decisions, what to let the model decide and what to hard-code, where to put the human in the loop, how to make a refusal explainable, are made better by a few people who have made them before than by many people who have not. A pod is small enough that everyone in it has read every prompt and every trace of every failure. That is not possible in a team of twenty.

How this connects to fixed price

Fixed price is not a pricing trick. It is a forcing function. To price an agent you have to define what it must do, what it must refuse and what it may cost per run. Those are exactly the decisions that most enterprise agent projects postpone until the pilot review, which is where they die. The pod model makes them the first two weeks of work.

If you want the longer argument, read Fixed price for agentic AI: how pods make it possible. If you want to work this way, the work with me page has the details, and Ailoitte’s own site is at ailoitte.com.