AI amplifies the engineering culture you already have

With strong delivery practices, agents accelerate good work.

Without them, agents accelerate entropy: more code, produced faster, understood less.

The tools are the least interesting part.

"Agentic AI" is three different problems

• Developer tooling agents speed up code production. They need tests, CI/CD, and clear boundaries.
• Business process agents speed up operations. They need defined processes, data ownership, and explicit rules.
• Customer-facing autonomous systems act for the organisation. They carry the most risk.

This framework focuses on the first.

Adoption happens at three levels that do not move together

• Individuals progress through eleven stages, from copying out of a chat window to building software factories.
• Teams choose between four modes for each piece of work. They do not climb them.
• Organisations provide the infrastructure and governance that limit what teams can do.

A capable developer in an unready organisation still stalls.

Most people stall between supervised and autonomous

Moving from approving every step to reviewing output needs three things:

1. Tests that catch agent errors before production
2. CI/CD that gates deployment on those tests
3. Observability that detects what the tests miss

This is where to focus enablement effort.

Rollout and readiness are separate

An organisation can give everyone an agent subscription while nobody has test coverage.

• Rollout: try, scale, optimise
• Readiness: clarify, harden, ease, accelerate

Most adoption problems live in the gap between the two.

Every role changes, and nobody has the answer yet

• Product: specification quality becomes the bottleneck
• Senior engineers: value shifts from implementation to judgement
• Junior engineers: the learning pathway needs redesigning
• Quality: moves from a separate check to an enabling function
• Design: shifts from making things look right to defining what right looks like

Agentic AI exposes your real management philosophy

The transparency that makes agents auditable also makes human work measurable.

Organisations that trust their people use it to learn. Organisations that only claim to trust them use it to watch.

Governance without engineering foundations is performative.

Six principles for adoption

1. Solve the sharpest constraint
2. Own your interfaces, not your implementations
3. Structure for flow, not for now
4. Sense and classify, do not predict
5. Distribute decisions to where the knowledge is
6. Adopt as an experiment, not a programme