I’m exploring how AI can extend what people are capable of, and building the tools and practices that make that possible.
For twenty years, my work has centered on helping teams accomplish more. I led product design at Fog Creek, helped Khan Academy grow from three people to a few hundred, and spent eight years building Radical Candor.
The organizations were different, but the underlying work was consistent: helping people do things they didn’t yet know they could do.
AI changes the scale of that possibility. One person can now take on work that previously required a team, but the experience is still far from accessible. Most people don’t have a useful mental model for dividing work between themselves and a machine. And the tools often demand too much setup, correction, and cleanup.
So this is the question I’m working on:
How do you build AI tools that make people measurably better at their actual work?
I’m approaching that question by building and teaching.
Compass is where I test the ideas. It started as a conversation engine, but the hard problems turned out to be everything around the conversation: remembering what a user has worked on before, guiding them into the right exercise instead of a blank prompt, surfacing the right resource at the right moment. Each of those required the AI to enhance the person’s judgment, not substitute for it.
That same principle drives the rest of my technical work: the Radical Candor website, internal data pipelines, and the smaller tools that keep the company running. Across those projects, I’ve shipped roughly 3,000 commits in the past year, nearly all of it built with AI.
The biggest lever isn’t speed. It’s having a clear model for the work and tools that stay out of the way. I want to make both easier for other people to access.
In development
A five-session, live-taught course for experienced professionals who want to work effectively with AI.
The course uses one durable mental model: the assistant you build (Brain, Hands, and Memory) and the work you give it (Brief and Mandate).
The model is deliberately tool-agnostic, so it remains useful when the tools change. About three-quarters of each session is hands-on.
Get in touch if you’re interested in bringing it to your team
Open source
One install that gives an AI agent governed access to every system it needs. Instead of connecting five vendor tools and flooding the agent's context with a hundred schemas, Charter exposes four tools, period. The agent discovers what it can do on demand, and the business rules ride inside each operation.
Every action is scoped to a named permission, tied to a human actor, and recorded in an append-only audit log. Keys are revocable in seconds, no redeploy. I built Charter because the alternative was either too dangerous (raw API tokens) or too rigid (configuration that can't express business logic).
Live
An AI coaching tool grounded in Radical Candor that helps people navigate difficult workplace conversations.
The conversation engine is the core, but the features that make Compass useful go well beyond chat: persistent memory so the coach builds on what you've already discussed, guided playbooks that walk you through specific skills step by step, a resource library you can search in natural language, AI-driven follow-ups that check in after your real conversation happens, and native apps on iOS, Android, and Mac.
I led a product design team at Fog Creek Software, the company behind Trello and closely involved in the creation of Stack Overflow.
In 2010, I moved to California to help launch Khan Academy. I served as both chief product officer and chief people officer as the organization grew from three people to a few hundred and reached learners around the world.
For the past eight years, I’ve been CEO and co-founder of Radical Candor, working with organizations ranging from early-stage startups to the Fortune 100.