Person/David Crawshaw/Blog/26/02/Eight more months of agents

• Eight more months of agents

  • Author: Person/David Crawshaw
  • Source: crawshaw.io
  • Journal link: 2026/02/17

  • High-level takeaways

    • I 100% agree with most of this framing on the state of AI-assisted engineering: frontier model capability is moving quickly, practical workflows matter most, and teams that adapt their tooling around agents gain a real speed advantage.

  • Notes on the argument (summary + excerpts)

    • Model progress is the main driver right now; harnesses matter, but capability jumps are coming mostly from frontier models.

      • "Right now, it is all about the model."
        

      • "In February last year, Claude Code could write a quarter of my code. In February this year, the latest Opus model can write nine tenths of my code."
        

    • The coding workflow is shifting away from IDE-centered interaction toward agent-centered interaction.

      • "In 2021, the IDE had won."
        

      • "In 2026, I don't use an IDE any more."
        

      • "The only IDE-like feature I use today is go-to-def... here I am, 2026, and I am back on Vi."
        

    • Using weaker/cheaper models can teach the wrong operational instincts when the frontier keeps moving.

      • "if you try some penny-saving cheap model... you do worse than waste your time, you learn the wrong lessons."
        

      • "you will not know what models will be capable of unless you use the best."
        

    • Built-in safety sandboxes create too much friction for serious use; in practice, people bring their own isolated runtime.

      • "Built-in agent sandboxes do not work"
        

      • "You have to turn off the sandbox, which means you have to provide your own sandbox... use a fresh VM."
        

    • He is building [[exe.dev]] to make unconstrained, disposable VM-based agent execution practical for everyday one-liner ideas.

      • "This is why I am building exe.dev."
        

      • "I need a VM, with an unconstrained agent, that I can trivially start up and type the one liner I would have otherwise put into an Apple Note named TODO and forgotten about."
        

      • "A good portion of the time Shelley turns a one-liner into a useful program."
        

    • His broader product thesis: customers now arrive with agents, so products should be easy for programmers (and thus agents) to compose against.

      • "I implemented that entire Stripe product (as it relates to me) by typing three sentences."
        

      • "the best software for an agent is whatever is best for a programmer."
        

  • Personal annotations integrated

    • The points above line up with my own experience, including doing similar CLI-centric workflows (e.g., with langstar) instead of relying on heavy UI-first flows.