Methodology / Foundation first

An agent depends on everything
you define beneath it.

The foundation does not change when you switch models. Evidence then determines whether the organisation is ready for the next step.

The original foundation pyramid from the paper: model, ten layers and an agent at the top

From foundation to application

Instructions

Define behaviour, boundaries and stop criteria. What must the model never do?

Foundation: your chosen model with suitable privacy settings. At the top: an agent that starts on a trigger, prepares an action, waits for approval and logs it. Instructions and context remain transferable.

Ownership

The builder is not the reviewer.

In small organisations, people can combine several roles. Keep building and independent review separate, and budget for review time.
Executive sponsor and agent owner with finance, risk, builder, reviewer and data steward

01

Chat: you ask, AI answers.

02

Artifact: you save, review and share.

03

Routine: the same output on a schedule.

04

Agent: trigger, proposal, approval and log.

Put AI to work

Ready to bring AI into your team’s daily work?

Talk to Wendy and Lavanya about team training, private executive training or a custom agent.

Book a training conversation