AI and agentic transformation

AI is the easiest part of an AI programme to buy and the hardest part to land. The constraint is almost never model quality. It is whether the organisation underneath can absorb what the model produces.

Agentic systems raise the stakes rather than change the problem. An agent that takes actions rather than making suggestions needs clearer ownership, tighter guardrails and better data than a model that only advises, and most organisations are reaching for agents before they have any of the three.

What usually needs fixing

A portfolio nobody prioritised. Twelve ideas, all sponsored, none sequenced, competing for the same three people. Prioritisation is a business decision, not a feature list, and it has usually been avoided rather than made.

Build decisions taken by people who enjoy building. Sometimes correct, rarely challenged, and increasingly expensive as the capability you are rebuilding ships in products you already own.

Pilots with no route to production. The integration work was never scoped, no one owns the workflow, and the pilot succeeds on its own terms while changing nothing.

Agentic ambition ahead of governance. Agents that take real actions need an audit trail, a rollback and a human accountable for outcomes. Retrofitting those is far harder than designing them in.

Adoption treated as a training problem at the end rather than a design constraint at the start.

How I approach it

Every candidate has to name the decision it changes and show a baseline before it gets funded. That rule removes most of a typical portfolio in the first fortnight, and it removes the right parts.

What survives gets a real value case, an owner in the business rather than in technology, and a route into an existing workflow that exists on paper before anyone writes code. For agentic work, it also gets an explicit answer to what happens when the agent is wrong, who finds out, and how fast it can be stopped.

The organisations getting value from AI are not the ones with the best models. They are the ones that changed how a decision gets made.

Where I am careful

I will tell you when the answer is a licence you already hold, a process that should be retired, or nothing at all. I have lost work saying all three.

I am also cautious about agentic deployment into processes with weak data foundations. An agent acting confidently on untrustworthy data does damage faster than a dashboard reporting it.

What you get

• A prioritised portfolio, with the things to stop marked as clearly as the things to start.
• Build or buy assessments that price configuring and buying as seriously as building.
• Value cases with measured baselines and named budget holders who have accepted the benefit.
• An agentic readiness view: where agents genuinely fit, what guardrails they need, and where they are a fashion.
• Delivery oversight that holds vendors and internal teams to the outcome rather than the milestone.
• An adoption plan written at the start, not retrofitted at go live.

How this connects

AI sits at the top of the chain, which means it inherits every weakness below it. If the estate is untrusted or the analytics layer is disputed, AI amplifies the problem rather than solving it.

The free gate check will tell you whether your AI work is actually blocked by AI, which in my experience it usually is not.