AI in the operating model
The pilot works. The operating model doesn’t know what to do with it.
Most organisations have AI pilots. Few have changed how work and decisions flow because of them. An agent that drafts, checks, plans or acts is a new actor in the system: it needs a place in the decision rights, guardrails it cannot cross, context it can trust, and someone accountable for what it does. That is operating-model work, the same work we do for teams.
Where the work is building the agents themselves, around your customer, I do it under a second name: Claint. Same person, same method, one question first: which job does this do for the customer?
What it looks like from the inside
- Twenty pilots, and not one decision is made differently.
- Agents are bought per department, and each one automates a hand-over that should not exist.
- The business case is a cost saving. Nobody asked what changes for the customer.
- Nobody can say who is accountable when an agent acts.
- Security and compliance arrive at the end, as a gate, and stop the pilot.
- Success is measured in licences and prompts, not in lead time or outcome.
Six things we design together
Value
Start from the customer, not the cost
We map the customer’s job and the value stream behind it, and put agents where work waits or decisions queue, not where a vendor demo looks good. Claint calls it a customer at the core.
Decision rights
For people and for agents
Which decisions an agent may prepare, which it may take, and which stay with people. Written down per level, with an escalation path.
Guardrails
A charter the agent works under
What it may read, change and send outside, what needs approval before or after, and when it has to be honest with the customer even if that doesn’t sell. Security, privacy and audit are part of the design, not a gate at the end.
Context
Data and knowledge it can trust
An agent is only as good as the context it gets: the customer’s intent and history, strategy, backlog, definitions, policies. We make that context explicit and keep it current.
Measurement
Customer outcomes, not usage
Lead time, quality, retention and trust. If an agent does not move them, it is a cost, not a capability.
Capability
A learning loop you own
Models are interchangeable. What lasts is the loop built from your workflows and judgement. Leaders and teams learn to run it on their real work, through coaching rather than a tool training, so it stays in the organisation.
How we start
The operating-model view and the build follow one motion. Each step delivers something you keep, so you can stop anywhere and still come out ahead.
AI State of Affairs
The audit lens (people, structures and content) applied to where AI already sits in your organisation and what it changes, or fails to change, in how decisions and work flow.
One customer job, one agent, one measurable outcome
A first agent as an experiment in the transformation backlog, with a hypothesis and a leading indicator, not a pilot running beside the organisation. Claint’s Align, Shape and Build take it from a one-page opportunity map to a working prototype on your own data in about three weeks.
Scale what moves the numbers
What improves flow or customer outcome gets a place in the operating model and its governance. What does not, we stop. Claint calls this Evolve: workflow after workflow, until your people run the loop without us.
Beyond Agile and Claint
Two names, one way of working
Beyond Agile redesigns the system around delivery: governance, decision rights, funding and flow. Claint builds agentic AI around your customer inside that system. Where you need both, you get one person accountable for both.
AI rollouts fail on adoption, incentives and workflow design, not on the model. That is operating-model work first, and technology second.
Questions clients ask
Do we need our own models?
Not to start. The first question is which customer job an agent serves and where decisions and work get stuck. Where data cannot leave the organisation, agents can run on local models; that is a design choice we make per use case.
Which platform do you build on?
Yours. Microsoft, Salesforce or another stack: we design and govern the solution, and the platform underneath stays interchangeable.
What about the EU AI Act?
Risk classification, transparency, human oversight and logging are designed in per use case, in line with the regulation and your own policies, rather than checked afterwards.
Will agents replace our teams?
That is the wrong frame. Agents take over preparation, checks and routine decisions; people keep judgement, accountability and customer contact. Treat agents as a way to cut headcount and you automate today’s hand-overs instead of removing them.
Pilots, but nothing changed?
Tell me where the pilots are and what has not moved for your customers. One conversation is usually enough to see where agents fit in your operating model.