Approach

How I work

I work across the people making, working with and experiencing AI, moving between research, strategy, learning and hands-on prototyping, and back again.

Agentic AI, with the human kept in

Agents don't just answer; they plan, use tools and act. I help organisations decide how much autonomy each task should have, and build the safeguards that make it safe to use.

Where I work

AI transformation involves three groups of people. Most programmes only look at one.

The people building AI, the people expected to work with it, and the people on the receiving end are connected. Decisions made in one group show up as problems in another.

Making AI

Product teams, developers, designers, managers and governance teams building or introducing AI systems.

  • What are we actually trying to automate?
  • Where should humans remain involved?
  • How should an agent behave, and where does responsibility sit?
  • How do we turn governance into product behaviour?
  • What happens when the AI is uncertain, wrong or unable to continue?

Working with AI

Employees, managers and teams suddenly expected to bring AI into their everyday work.

  • AI literacy and role-based learning pathways
  • Coaching, workshops and finding real use cases
  • Prompt and workflow development
  • Confidence, trust and safe use in practice
  • Champions, peer learning and measuring whether work actually changes

Experiencing AI

Customers, citizens and users of products and services that increasingly contain AI.

  • Do they know AI is involved, and what do they expect it to do?
  • When do they trust it? When should they be able to question or override it?
  • What happens when a process moves between human and AI?
  • How do we design agentic journeys without making people understand the architecture?

Making AI, working with AI and experiencing AI are one system. I work across all three.

Why that matters

What I bring to an AI programme

AI strategy only matters once it changes how work gets done. These four areas are how I get it there.

  • Strategy and use cases

    I turn AI ambition into a roadmap: use cases assessed for value, data, risk and effort, with a clear call on what to build, buy, pilot or park.

  • Governance and risk

    I translate the EU AI Act, data protection and information security into guidelines and decision aids people can use at the moment they matter.

  • Enablement and learning

    I advise leadership and teams, and design role-based learning that moves people from knowing AI exists to using it safely in their own work.

How I work

Audit, pathway, prototype, test, scale.

A method for moving from AI ambition to changed behaviour. Each stage produces something people can react to, and testing always feeds back into the next round of understanding.

And back again. What we learn when scaling becomes the next audit.

01. Audit

Before recommending another tool, course or AI initiative, investigate the system around it.

Methods can include

  • Stakeholder interviews
  • Employee and persona research
  • Workflow analysis
  • AI readiness
  • Adoption barrier mapping
  • Trust research
  • Use-case discovery
  • Governance constraints
  • Tool landscape
  • Skills and confidence assessment

Possible outputs

  • AI adoption map
  • Personas
  • Opportunity areas
  • Barriers
  • Risk points
  • Prioritised use cases

02. Pathway

Turn research into practical pathways for change. Different people need different pathways: a developer, an administrator, a product manager, a manager and a senior leader shouldn't receive identical AI enablement.

Methods can include

  • Role-based AI learning pathways
  • AI literacy programmes
  • AI coaching and workshops
  • Use-case development
  • Adoption journeys
  • Champions programmes
  • Governance in practice
  • Measurement frameworks
  • Communication and support

Possible outputs

  • Role-based pathways
  • Learning programme
  • Champions model
  • Measurement framework

03. Prototype

Don't spend months debating abstract AI possibilities. Build enough of the future that people can experience it. I use AI-assisted development tools such as Cursor, and generative AI, to shorten the distance between idea and evidence. I'm not limited to describing an AI-enabled experience: I can make it tangible enough to test.

Example: EquitAI, an AI agent prototype for job applications

Methods can include

  • AI agents and agentic workflows
  • AI assistants
  • Automated self-service
  • Internal tools
  • Conversational experiences
  • Decision support
  • Human and AI handoffs
  • Multi-step AI journeys

Possible outputs

  • Working prototypes
  • Testable journeys
  • Interaction patterns

04. Test

Research how people actually respond to what the AI does.

The question isn't “Does the AI work?” It's “Can humans work with what the AI does?”

Methods can include

  • Usability and comprehension
  • Expectations and mental models
  • Trust and confidence
  • Failure states
  • Human oversight and escalation
  • Accessibility
  • Responsible AI considerations

Possible outputs

  • Evidence
  • Design changes
  • Oversight patterns

05. Scale

The goal isn't endless pilots. Use evidence from research and experimentation so the organisation becomes better at adopting AI, rather than simply deploying another tool.

Questions to answer

  • What should scale, change or stop
  • What people need to learn
  • What managers need to support
  • What governance needs to enable
  • What becomes everyday workflow
  • How adoption is measured

Possible outputs

  • Scaling decisions
  • Adoption metrics
  • Organisational learning

Method

From AI idea to changed work

My method, with the technology and the human checkpoint at every stage. Followed through one illustrative example: a multilingual assistant answering applicant questions in a high-volume application service.

  • Human checkpoint
  • System or AI step
Method pipeline from Audit to Scale with a feedback loopA five-stage pipeline: Audit, Pathway, Prototype, Test and Scale. A continuous flow line runs through each stage. Human checkpoints sit on the line at every stage. A dashed arrow returns from Scale to Audit, labelled: what we learn in use feeds the next audit. Stage icons show a funnel, a sandbox gate, a retrieval schematic, a test bench, and three buildings for rollout.01Audit02Pathway03Prototype04Test05Scale
  1. Audit — Understand what is actually happening.

    What happens
    I map the real work, data and pain points with the teams involved, and keep track of what the technology can now do.
    Technology
    Process and data inventory, analysis of incoming requests, a scan of available tools and platforms.
    Human checkpoint
    The business units confirm the problem is worth solving.

    Example: Many applicant emails turn out to be repeat questions about documents and deadlines.

  2. Pathway — Decide what to build and how it may run.

    What happens
    Requirements, use-case assessment and the choice of approach, before anyone builds.
    Technology
    Build, buy, RAG or fine-tuning; cloud, sovereign cloud or on-premise; EU AI Act risk class; data protection; alignment with information security (NIS-2, ISO 27001).
    Human checkpoint
    Governance, information security and data protection sign off. Nothing leaves the sandbox without it.
    In the diagram
    Gate: leaves the sandbox only with sign-off.

    Example: Limited-risk assistant with a transparency notice, EU-hosted model, official guidance as its only source, and no role in admission decisions.

  3. Prototype — Make it tangible.

    What happens
    A working version people can react to, built in small iterations.
    Technology
    Retrieval-augmented generation over approved content, integration with existing systems through APIs, logging from day one.
    Human checkpoint
    The staff who answer these questions today shape the answers, tone and limits.
    In the diagram
    Approved sources → index → retrieval → language model → answer with citation, plus a hand-over to a person when the system isn't sure.

    Example: The assistant answers in the applicant's language, cites the official page, and hands over to a person when it isn't sure.

  4. Test — Put humans back into the system.

    What happens
    We check whether it is good enough, and for whom.
    Technology
    An evaluation set built from real past questions, checks on accuracy and citations, tests for failure modes (made-up answers, outdated rules, wrong language), accessibility testing.
    Human checkpoint
    People review the errors and decide which mistakes are unacceptable.
    In the diagram
    We separate critical errors from acceptable ones — people decide which mistakes must never ship.

    Example: A wrong deadline counts as a critical error. It's tested against real questions before any applicant sees it.

  5. Scale — Turn it into capability.

    What happens
    Rollout with the people affected, and ownership that lasts.
    Technology
    Monitoring, feedback from users, a process for updating sources, access control, cost tracking.
    Human checkpoint
    Named owners, trained staff and a regular review.
    In the diagram
    The same capability reaches applicants, case workers and governance.

    Example: Case workers learn to handle escalations and flag outdated answers; the assistant is reviewed every month.

Illustrative example, not a client project.

What makes this practice different

AI transformation fragments. I work across the gaps.

  • Strategy teams define the vision.
  • Technical teams build the systems.
  • Governance teams define the rules.
  • Learning teams train employees.
  • Design teams create interfaces.
  • Employees try to work with the tools.
  • Customers experience the consequences.

These activities are deeply connected, but organisations usually run them as separate workstreams. My strength isn't knowing several disciplines. It's being able to move between them.

Research, strategy, learning,prototyping, testing,adoption.And back again.

In one day I can interview employees about why an AI rollout is stalling, map the adoption problem, run a workshop on possible use cases, turn the findings into a learning pathway or workflow, prototype an AI-enabled interaction, and take that prototype back to people to find out whether it actually works.

That movement between understanding, strategy and making is the defining characteristic of the work.

One transformation, seen at seven levels at once

Research, AI adoption, organisational learning, service design, responsible AI, accessibility and hands-on prototyping let me hold all of these in view at the same time.

  1. Individual

    What does this person need to understand, trust, learn or do differently?

  2. Workflow

    Where does AI actually fit into their work?

  3. Experience

    How should a human interact with the AI?

  4. Product

    What should we build or prototype?

  5. Organisation

    What has to change for adoption to happen?

  6. Governance

    What boundaries, oversight and decisions are necessary?

  7. Strategy

    Which opportunities are worth pursuing, and how do we move from experimentation to capability?

Working from both directions

Most AI programmes start with the technology. Human-centred transformation can also start with the problem. I connect the two.

Starting from technology

  1. Technology
  2. Capability
  3. Possible use case
  4. Human workflow
Is this worth building, and can people work with it?

Starting from people

  1. Human
  2. Problem
  3. Workflow
  4. Opportunity
  5. Possible AI intervention

Connecting both directions avoids two expensive extremes: building AI simply because the technology exists, and discussing human needs without understanding what emerging AI can actually make possible.

What this changes

Organisations don't need another AI strategy. They need AI ambition to become changed work.

  • From AI investmentto actual adoption
  • From experimentationto repeatable practice
  • From generic AI trainingto role-specific capability
  • From AI use casesto working prototypes
  • From governance documentsto decisions people can make
  • From automationto well-designed human and AI workflows
  • From technically functional AIto AI people understand and can work with
  • From isolated pilotsto organisational learning

How the value shows up

These are the mechanisms through which the work creates value. Specific results depend on the organisation and are documented in individual case studies.

  • Faster identification of AI use cases that matter
  • Less investment in AI solutions people won't use
  • Faster movement from idea to testable prototype
  • More employee confidence and capability
  • Clear adoption pathways for different roles
  • Better alignment between AI teams, business teams and users
  • Less friction around governance and responsible use
  • Better-designed human oversight
  • A clearer view of where automation genuinely creates value
  • Evidence for deciding what to scale, change or stop
  • More usable, trustworthy AI-enabled customer experiences

Current questions

Designing journeys for humans working with increasingly autonomous systems.

Agentic AI changes the design problem

With traditional software, people see each step. With agents, they express an intent and much of what follows happens out of sight. The experience challenge moves from designing screens to designing oversight.

Traditional software

  1. Human
  2. Interface
  3. Action
  4. Result

Agentic systems

  1. Human intent
  2. Agent interpretation
  3. Planning
  4. Automated actions
  5. Other systems and tools
  6. Decisions and uncertainty
  7. Human checkpoint
  8. Further autonomous action
  9. Outcome

Planning, automated actions, other systems, decisions and further autonomous action are often invisible to the person. The human checkpoint is where they come back in.

  • What should humans see, and what can stay invisible?
  • When should the agent ask permission? When should it stop?
  • How can the human intervene and correct it?
  • How are progress and uncertainty communicated?
  • How does responsibility move between human and system?
  • What happens when the agent fails?
  • How do we stop automation becoming a loss of control?

Automated self-service

AI could translate what people need into the actions an organisation requires, instead of making people learn its departments, forms and categories first.

Today

  1. Find the right department
  2. Learn the terminology
  3. Find the correct form
  4. Understand the requirements
  5. Enter the information
  6. Work out what happens next

With AI

“I need to do this.”

But automation brings its own problems:

  • Invisible decisions
  • Wrong assumptions
  • Uncertainty
  • Unclear responsibility
  • Loss of control
  • Accessibility barriers
  • Trust
  • Recovering from errors

Remove complexity without removing agency.