I’m Steven Smith, a product and project lead in Tokyo. I build and ship products with AI, open-source the tools that keep the output honest, and help teams build real AI capability of their own.
Workshops, learning paths and writing that teach people to direct AI and to judge what it gives back.
I help teams rework their day-to-day around AI, then put guardrails in place so the output stays consistent.
Research, UX, design systems and front end, start to finish. AI does the heavy lifting. I decide what’s good enough to ship.
It all sits on years of project delivery, product strategy, UX and research.
Background
Most of my career has been project management. Product and UX came later, and for the past two years AI has been the whole focus.
Sydney · 2010–2015
IT infrastructure and service delivery projects for corporate clients, then running operations and delivery for remote monitoring and security systems. That’s where I learned to keep engineers, clients and executives talking to each other.
Tokyo · 2015–2019
I ran three bilingual job websites for recruiters and job seekers, looking after the user experience, content and SEO.
Goodpatch · since 2019
Product and project manager at a UX/UI design consultancy. I lead research, design and development teams for clients in financial services, automotive and enterprise, often as the go-between for Japanese clients and overseas teams.
Now
I build production apps with AI. This year I shipped an iOS app for an enterprise client as the sole designer and front-end developer, working through AI agents. So the quality doesn’t come down to luck, I build my own skills and harnesses that check the work against a standard before I call it done.
Implementation and training
I help organisations put AI into everyday work, from workflow automation to agent systems, including a harness for a client’s business-development team. I also run AI training for people in every kind of role. The aim is that they can keep going after I’ve left the room.
Harness
The tools I build with, open for anyone to use.
Two harnesses so far, with a design skills kit and a design harness to follow. I built them for my own work, and I’m opening them up so you can try them on yours. A harness is the environment around the agent: one is for building software, and the kit builds one for any other kind of work.
Open sourceHarness
A ready-made harness for building software with AI. Install it into a project and the rules, skills, memory, security and automatic checks are in place from day one, on whatever stack you use, for engineers and non-engineers alike.
Harness parts it covers: Rules, Skills, Memory, Security, Governance, Checks.Every automatic check comes with a test that proves it can fail.
Open sourceHarness
Builds a harness for any kind of work, not just software. Point it at where your work already lives, such as a folder, a repo, Slack or Notion, and it turns that workspace into a robust AI workspace. It finds and reconciles your rules, puts the right checks in place, and turns repeated corrections into new rules over time.
Harness parts it covers: Rules, Security, Governance, Checks.Every check it creates is tested by deleting it and confirming a test fails.
Things I’m building for people to use.
In developmentProduct
A guided, project-based way to learn AI by building real things. Progress is measured by what you make.
What I’m learning about designing and building with AI, written up as I go.
Field reportEN / JAOctober 2026
AI can make a page in seconds. Getting one that looks designed takes the same things good design always took: context, references, a system, the fundamentals, an idea of its own, and review. AI tells are only one part of it.
Read the pieceField reportEN / JAOctober 2026
An open-source reverse-engineering tool, an agent, and about 30 minutes per app. The process, what came out, and what the copies couldn’t reach.
Read the pieceField reportEN / JAOctober 2026
The AI checks and improves its own work in loops, people keep the decisions, and a harness holds it all to a standard. Shown on a real app I led.
Read the piece
