Experiments

A desktop app that runs a Linux sandbox inside Apple's Containerization framework to power a Claude-driven agent harness with a fleet of local MCP tools.

An experiment in personalized, adaptive web experiences — generating entire web pages at request time from site-owner facts, design systems, and a knowledge base.

An in-app assistant that surfaces widgets to get users to what they need quickly, rather than making them navigate for it.

A focused editing surface for building and tweaking the individual components that the page builder assembles into pages.

Pages authored on demand from a user persona, site description, and requested URL — with GPT-4-generated article slugs that author and cache the page when selected.

A short look at generating site navigation on the fly, as part of the broader just-in-time rendering experiments.

When websites can be generated on demand, content management becomes the thing that matters most. This explores digital asset management from inside Vision Pro.

Using mobile designs as context when generating desktop layouts and navigation — combining sitemap design with user flows and page building.

A take on no-code page building, working from the premise that no-code tools have to be simpler than the generative-AI tools they compete with.

A little generative social-story builder I made in December 2022, before most people had clocked what generative AI could do.

A short glimpse at hyper-personalized, localized page rendering — the direction I think page generation is ultimately heading.

An event-sourced workflow engine that streams agent task updates over sockets in real time, turning a design system and knowledge base into authored markdown.

Moving away from text prompts toward an AI peer that watches as you work and offers an assist when it can — with voice suggestions via Whisper.