Designing and building software, from the interface to the systems behind it.
Most of the work is with founders: settling the technical decisions that are expensive to undo, then building the product, from the screens people use to the services and AI behind them.
Recent notes.
All notes →Be Careful With the AI Layer Your Product Depends On
When AI is central to a product, model access, data boundaries, routing, cost and reliability become infrastructure decisions.
Reliable Agents Need an Operating Model
Prompts help, but a reliable AI product also needs operating rules, review loops, permissions and failure handling.
Managing AI Agent Work Requires Clear Judgment
As agents take on more of the work, the job shifts toward clear goals, careful review and accountable decisions.
Agent Work Depends on Reliable Inputs
Useful AI products depend on grounded, current and structured information. A larger context window does not make up for it.
Agents Need Workflows Before They Can Help Reliably
Agent demos become useful products once the work has clear owners, scoped permissions, isolation and review, with a person accountable for what ships.
AI-Assisted Work Needs Reviewable Evidence
AI products should prove their work with screenshots, artifacts and checks that a reviewer can inspect.