AI agent infrastructure
OpenProse
OpenProse builds open-source infrastructure for ongoing AI work. Its Prose language lets developers write Markdown contracts that declare inputs, responsibilities and conditions to maintain, then run them inside coding agents such as Claude Code or Codex. Developers can compose skills, tools and subagents into workflows, with ProseScript available for explicit control flow. The language is MIT licensed and designed to work across compatible agent harnesses.
Founded in 2026, the San Francisco company is led by founder and CEO Dan Barrett, who created OpenProse. Its technical approach separates dependency wiring from execution: Forme connects contracts by matching their requirements to the outcomes other contracts maintain. Reactor, its experimental alpha runtime, compiles contracts into a reactive dependency graph, reruns affected nodes when inputs change and records receipts. Deterministic reconciliation makes execution decisions without asking an LLM whether to wake or commit work.
OpenProse joined Y Combinator’s Spring 2026 batch and had a listed team size of three in August 2026. Its openprose/prose repository had 1,688 stars and 136 forks that month. Commercially, it describes its offering as scheduled outcomes delivered as a service; its experiments applying contracts to its own company operations remain in alpha.
1 talk
Newest first1 speaker at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.
Messages from the stage
Context as something agents can manipulate
Weitekamp described Recursive Language Models that externalize context into a REPL, manipulate it symbolically, and delegate work recursively.
Affiliations reflect each recorded session, not necessarily current employment.
