Ehud Shapiro's vGLP language lets AI build apps from human volitions

Ehud Shapiro’s latest paper introduces vGLP (volition‑guarded GLP), a programming language designed for distributed systems where people and their personal machines act as a single agent. Building on earlier work on Volition‑Guarded Multiagent Atomic Transactions, vGLP formalizes how a transaction can proceed only when both the machine state and the person’s volition consent. The operational semantics model Communicating Volitional Agents, and crucially require the system to “elicit” what the person wants — i.e., the UI becomes the mechanism to discover what’s in the user’s head. This is a radical shift: instead of a developer deciding the interaction flow, the user’s intent drives every step, and the AI translates that volition into executable code.

Shapiro demonstrates the concept with three real‑world platforms: a grassroots social graph, a social network, and a digital currency. Each platform is described as a vGLP program, generated automatically by AI from volition‑guarded multiagent atomic transactions. A second AI then maps those clauses into standard UI constructs, producing a single working app that runs on a physical smartphone. The paper, submitted to arXiv on July 10, 2026, spans multiple CS fields — programming languages, AI, distributed systems, HCI, and multiagent systems — and positions vGLP as a new paradigm for building human‑centric distributed applications where consent is explicit, not assumed.