Partner at the desk.
AI is now part of the work, not just something we talk about. Daily, in design exploration, research, prototyping, writing, automation. Cloud, or a self-hosted stack for private repeatable work. AI is in the loop.
The stance
From idea to tested build.
Rapid mid to high fidelity design, fast stakeholder sign off, then a clean handover so development and the proper Figma build can run in parallel.
Rapid iteration
Mid to high fidelity screens, fast. Enough resolution for stakeholders to react to something real, not a wireframe.
Stakeholder sign off
Tight feedback loops on the iterations. Decisions land quickly because there is something concrete to point at.
Handover to development
Dev begins mapping solutions and architecture. Patterns and general refinement happen here, while the build catches up.
Hand built in Figma
Built from the ground up, correctly, so it can be connected and properly tested through AI and automation later in the lifecycle.
Architecture and refinement on the dev side overlap the manual Figma build, so the correct foundation gets laid without holding the timeline.
On Figma AI composers
I have tested a lot of them. Most get you halfway there, then the other half, the refactoring, takes too long. In today's market it is still a matter of building it in Figma from the ground up correctly, so it can be connected and tested through AI and automation later. The build that lasts is still the hand built one.
What I use, every day.
Six areas. Each one earns its keep on real client work.
Agentic coding
AI in the IDE for multi-file edits, terminal work and test loops.
Tools
- Claude Code
- Cursor
- Repo prompts
- Sub agents
The big models
Frontier models for prose, planning, code and image work.
Models
- Claude Sonnet/Opus
- GPT-4 class
- Gemini Pro
- Model routing
Image & vision
Generative tools for moodboards and pipelines. Vision models for UI critique.
Pipelines
- Midjourney
- ComfyUI
- Vision API
- Screenshot critique
Local LLMs
Self-hosted models for work that can't leave my machine.
Stack
- Ollama
- Llama 3.x
- Qwen
- LM Studio
Automation
Workflow automation for the boring. File watching, reports, email triage.
Flows
- n8n
- Local agents
- Webhook pipes
- Cron jobs
Research & writing
Long-form research into structured notes. Drafts, edits, voice matching.
Uses
- Doc summarisation
- Transcript coding
- Voice matching
- Spec writing
Shipping product with AI inside.
AI in the workflow is one thing. AI in the product is another. Below is what I think about when the model is part of the surface, not a chat window glued on.
LLM in the loop UX
The model is one collaborator in a workflow with humans. Design the handoff first, then the model behaviour.
Confidence & control
Two screens for one model output. One for when it's confident and the user trusts it. One for when it's not and the user has to choose.
Streaming & tool use
Streaming changes the interaction grammar. So does tool use. The new wait state is "the model is thinking, here is what it's looking at".
Evals & observability
Prompts are software. They regress. Production AI without evals is a roulette wheel. I design observability into the UI from day one.
Not a fad, a stack.
I have been a senior product designer through every model wave since 2019. Gravicus used ML and NLP back then. Chainalysis used graph models on chain data. The current LLM wave is the loudest, but the discipline goes back further than that.
Designed UIs for compliance models that surfaced risk from messy enterprise data.
On chain investigation tooling. Address clustering, transaction graph analysis.
When Claude 2 and GPT-4 shipped tool use, the daily workflow shifted.
Built the local pipeline. Self hosted Llama and Qwen. The whole thing runs on my desk.
Designing AI assisted surfaces into ArchCo's operational layer.
Let's design the thing, not the demo.
Building product with AI inside is more interesting than building AI products. If you have a workflow where the model should sit beside the user not in front of them, I'm the right call.