Capability / Enterprise R&D
AI workflows that make R&D faster and more controllable
I design the operating systems around AI: the sources of truth, retrieval context, specialist agents, approval points and implementation path that turn AI capability into reliable delivery.
before generation
AI works from the organisation’s real standards and knowledge.
at decision points
AI assists the work; accountable owners choose what progresses.
that can evolve
Reusable foundations reduce repeated work as products change.
AI is not the workflow
A model, a prompt and a collection of tools are not yet an operating model. In an enterprise, the difficult part is deciding what the AI is allowed to use, how it receives current context, where expert judgment intervenes and how a good result becomes a maintainable product.
That is the work I lead: designing an AI workflow around the organisation’s real constraints, not around a generic demo. The goal is useful throughput without creating a parallel system that is impossible to govern or maintain.
Start with the knowledge people already trust
Strong AI workflows begin with sources of truth that people can maintain. Depending on the problem, these can include a UX pattern library, a production component library, product documentation, research, policies and delivery history. Retrieval-augmented generation and vector memory make the relevant context available when the work starts, rather than leaving the model to guess.
This is especially valuable when several teams are changing work in parallel. The same approved pattern or component can inform design, implementation and review—reducing avoidable reinvention and making changes easier to trace.
Connect the workflow, not just the tools
I use task-specific skills and purpose-built agents to make AI assistance repeatable. MCP-connected tools and APIs can provide controlled access to the systems the task actually needs. The workflow then defines the sequence: frame the need, retrieve the relevant context, generate a proposal, review it, approve it and implement it.
Each connection needs a purpose and an owner. A connector is useful because it brings authoritative information into the task—not because a long list of integrations sounds advanced.
Keep human judgment where it matters
AI can compress the time from problem to first credible output. It cannot own the organisation’s priorities, customer trade-offs or release decisions. I design explicit review and approval points so designers, engineers, business owners and stakeholders stay accountable for their part of the work.
That makes the workflow more credible to the people who must use it. It also helps the organisation scale AI adoption without treating quality, governance and speed as competing goals.
What I bring to the work
With 16 years in UX and CX, I combine product judgment with the ability to shape the operating model around it. I work across design, development, marketing and R&D: translating a business constraint into a system teams can run, measure and improve.
For a concrete example, see the AI-enabled design-to-code case study. It shows how I connected a Figma UX library, a frontend library, retrieval context, specialist skills and business approval into one governed delivery path.