Requests, Promises, Assertions, and Declarations: Rewriting Our AI Agent Workflow as a 1986 'Committed Speech' Machine

This article explores a novel approach to improving autonomous AI coding agents by applying Fernando Flores’s 1986 'Committed Speech' theory. The author argues that common failures in AI workflows—such as hallucinated completions, mid-flight bailouts, and silent plan deviations—stem from a lack of linguistic rigor. By replacing defensive prompt engineering with a formal 'Commitment Machine' based on four speech acts (Requests, Promises, Assertions, and Declarations), developers can enforce accountability. The system integrates Nuel Belnap’s four-valued logic to manage epistemic states and Tony Hoare’s Communicating Sequential Processes (CSP) to handle agent handoffs. By treating every task as a two-party contract between a 'Customer' and a 'Performer,' the workflow ensures that agents only proceed based on verified physical evidence rather than vague instructions. This architectural shift significantly reduces the need for complex, defensive prompt lawyering, resulting in more reliable, self-healing autonomous coding systems.