AI development
I put AI in the product, not on a slide.
I design the feature, the limits, and the handoff when the model is unsure. You get something a user can click this month — not a chatbot stuck on a brochure.
Product features
Draft, search, recommend, and summarise with the user’s context still in view.
Agents at work
Read a queue, call an API, and stop when the answer has to be right.
Honest limits
The model does not invent your prices, your policy, or your customers.
How I build it
Useful AI has a job, a limit, and a person behind it.
I start from the work: support that should resolve, admin that should stop being a spreadsheet, a tool that should draft and then get out of the way. The model is part of the stack. It is not the product.
Inside the app
Features live next to the data they need. The user does not leave to “try the AI.” They finish the task they already opened.
Agents and workflows
A queue, an API, a clear stop. Support, ops, and internal admin are where this earns its place.
On the website
Guided intake, sharper search, and copy that stays on-brief. The site should feel like the right studio, not a demo.
What I will take
- A new AI feature in a product that already has users
- An agent that does a real job and hands off when it cannot
- Website intake or search that has to stay factual
What I will not take
- A chatbot bolted on a brochure and sold as transformation
- A model with no limits on prices, legal text, or medical advice
- A six-month discovery with no working preview
If the job is real, send the brief.
Tell us who it is for and what should be true when we are done. You get a next step, or an honest no.