skip to content
Search
Light / Dark Mode

Essays

  • What Will Be Left for Us to Work On?

    An illustrated adaptation of Arvind Narayanan's ICML 2026 keynote on AI, human agency, and the shift from building to judgment.

  • OpenMake: A Bet on Intelligence Staying Plural

    As frontier intelligence becomes cheap and rentable, enterprises will compete on an executable theory of their domain, not on access to models.

  • Imagineering (2026)

    A June 2026 rewrite of my Imagineering essay on the skills that endure when AI turns every builder into a manager of stacked agentic loops.

  • Traces Are the New Source Code

    A local-first evidence layer for agent work: capture what agents see, do, and change, anchor it to Git, and reuse it for search, lineage, evals, and datasets.

  • An application for every user

    Adaptive applications ship a fixed base, then grow a layer of code around how each user works. A harness for every task becomes an app for every user.

  • Trace Capsules: Agent-to-Agent Issue Reporting

    Trace Capsules turn upstream agent bugs into replayable issues by packaging the failing intent, context, trace slice, and repo snapshot.

  • The Collapse of the Trust Contract

    ASI breaks the symmetry that kept software security intact. Open source with provable hardening, or consolidation, are what's left.

  • The Software Market After the Claw Revolution

    How runtime code agents reshape software markets, turning apps into delegation targets where work flows to the best execution environment.

  • The Builder Paradox In 2026

    When everyone builds fast with the same AI tools, speed stops being a moat. The real question is what is actually worth building.

  • How Code Agents Disrupted the Software Market

    A speculative 2026-2036 history: as code agents become software's primary operators, selection pressure shifts to measurable delegated execution success.