The Promise None of Them Kept

The promise theory control loop is composed of three axioms: an agent is autonomous, can only promise its own behavior, and has local knowledge of the world. This leads to a control loop where the agent observes its environment directly, decides based on local reasoning, and commits to its own behavior. However, none of the popular infrastructure tools, including Chef, Puppet, Terraform, and Ansible, have fully implemented this promise theory control loop.

Each tool fails to observe, reason, and commit independently, instead relying on file-based inputs, central control nodes, and imposed changes. The missing piece that finally arrived is the reasoning layer provided by large language models (LLMs), which allows agents to observe live systems, weigh context, and decide whether to act.

While CFEngine came closest to implementing promise theory, it stopped short for reasons worth being precise about. The article argues that the promise theory control loop is still enough to differentiate tools, and Terraform's and Ansible's actual models are closer to batch scripts with diffing preambles than to the autonomous agents envisioned by promise theory.