Is this for you?

• You have an agent or automation running, but no one is sure who supervises it or what happens when it gets something wrong.
• The process got slower after adding AI, because people now check everything twice.
• No one can say exactly where the AI's responsibility ends and a person's begins.
• People spend more time managing the tool than working with it (BCG 2026: true for 47% of people).
• You're building a new process or service and want AI's role designed in from the start.

What I do

• New process — work designed from zero, with AI built in from the start.
• Redesigning an existing process — from a map of current work to the new version.
• Service blueprint — what the customer sees and what happens behind the scenes, on one page.
• Internal processes — approvals, case handling, reporting, and other work your people do.

The AI layer

This is the core of the service — AI governance at the level of the process, not the document. For every step, I define:

• Autonomy level: AI does it alone · AI proposes, a person approves · a person does it, AI assists · human only.
• Human checkpoints: where and on what basis someone reviews the result before it moves on.
• Permissions: what the AI can access, what it can change, what it can never touch.
• Handoffs: how work moves between AI and a person, with what context and in what state.
• Error paths: what happens when the AI is wrong, doesn't know, or isn't confident — and who notices.
• The starting principle: begin with the simplest thing that works, and increase autonomy once there's evidence for it (in line with Anthropic's own guidance to start simple).

What you get

• A workflow design — steps, roles, tools, handoffs.
• An autonomy map — the AI autonomy level for every step, with the reasoning behind it.
• A roles and permissions matrix — who (or what) does, approves, and is informed.
• A service blueprint — front stage and back stage on one page.
• Error scenarios — what happens when the AI is wrong, and how the process catches it.
• A plan for increasing autonomy — the signals that tell you the AI is ready to do more.

How it runs

1. Starting point — a map of the current work (from Workflow mapping, or a fast version done at the start).
2. Initial design — the new workflow and a first autonomy map.
3. Team workshop — we walk the process through real, concrete cases.
4. Error scenarios — we check what happens when something goes wrong.
5. Build-ready version — the design, the roles matrix, and the autonomy plan.
6. Optional — a process prototype (Prototype) and a test (Testing).

Roughly 3–6 weeks [to be confirmed].

How we'll know it worked

• Everyone in the process can say what they're responsible for, and when the AI acts alone.
• What happens when the AI gets something wrong is known before it happens.
• The build team has a design they don't have to guess the rules from.
• Metrics are agreed: case time, number of corrections, share of approvals made without changes.

What this doesn't cover

I don't build agents or integrations, and I don't choose the AI model. I design the work the agent is meant to operate inside; your team or a vendor builds it.

Visualizations

The autonomy ladder

Autonomy is set step by step, not for the whole system.

A process with checkpoints (sample artifact)

Sample artifact: who does what, and where a person decides.

Roles and permissions matrix (sample artifact)

Sample artifact: responsibilities, including the agent's.

Related services

• Workflow mapping — a map of current work as the starting point.
• Interaction & interface — how people see and correct the AI's work.
• Scale readiness — preparing people, and the criteria for scaling.