Vog began on June 2, 2026 as a thought exercise. The opening question, “what’s the best language for a fleet of AI coding agents?”, inverted within the hour into “what would the best one be?” The spine fell out almost immediately: static typing, full immutability, no metaprogramming, actors, integrated red-green testing. By that afternoon a near-complete design existed, built on pipes and a unified result type born from the insight that “fail is wrong, sometimes the result is just not ok,” a boundary that admits absence but never unsafe nulls, and a compiler-enforced separation of the code that touches the world from the code that doesn’t.
It was agent-first from the first minute, but its instincts (locatability, loud-over-silent failure, a willingness to reject received wisdom like the for loop) came from decades of hitting the industry’s footguns. The design is really that experience run as a challenge and puzzle: how many footguns can one language make unrepresentable? The result is a language deliberately unfriendly to human fingers, readers-before-writers taken to the point where only a tireless machine writer can afford the verbosity, designed largely by voice on a phone during commutes as an evenings/weekends project, with an LLM as a foil and a coding agent committing the changes back home. It stands on the shoulders of giants, and the author is proud and excited for its future.