Writing
Notes on building AI and SaaS products: how I work, working with founders, and lessons from my own studio products.
Where should an AI agent ask for human approval?
Useful automation needs boundaries: which actions an agent can take on its own, which need approval and what the person approving needs to see.
Read
How I would test a product idea before building the full application
Before investing in the complete product, design a smaller experiment that could genuinely change your mind.
Read
Your AI feature is cheap to demo. What does it cost to operate?
Why the cost of an AI feature is the cost of a successful result, not a single model call, and what to work out before you add one to a subscription product.
Read
AI-generated UI still needs an accessibility review
Why an interface that looks right in the browser is not finished, what recent research says about AI-generated code and accessibility, and what I check by hand.
Read
Do you need an AI agent, or just a better workflow?
The difference between a workflow and an agent, where fixed rules beat model decisions, and how I decide how much freedom an AI system actually needs.
Read
The new model looks better. Should you actually switch?
A model release is a candidate for testing, not an automatic upgrade, so here is how I check whether a new model is better for your product and how to switch safely.
Read
Your AI agent's tools are part of the user experience
The tools an AI agent uses shape what your customers see, so they deserve the same design care as any screen, from clear names to useful responses and errors that help recovery.
Read
What a domain expert and a technical cofounder should validate together
How I think a domain expert and a technical partner should share the work of deciding what deserves to be built, before either of them commits to a long build.
ReadPage 2 of 4 · 32 articles