You Were Never Declaring State. You Were Observing by Hand

What changes when your agent can observe reality instead of reading your notes about it

Every Terraform file you ever wrote was a note to a blind tool. You looked at your infrastructure, decided what it should be, wrote that decision into HCL, and handed it to a program that could not see for itself. The declaration was your observation, pre-computed and frozen into a file.

The same was true for Chef recipes, Puppet manifests, and CFEngine promises. You observed the system. You wrote down what you saw and what you wanted. You gave that note to an agent that could parse files but could not look around.

That was never “declaring state.” That was observing by hand and writing your observations into a format a blind agent could consume.

What the declaration actually was#

Mark Burgess described promise theory in In Search of Certainty as a model of autonomous agents reasoning about their environment and making local promises about their own behavior. The agent observes. The agent decides. The agent acts.

The tools we built to approximate this vision could not do the first step. A CFEngine agent could parse a promise body and converge toward it, but it could not observe the broader system and form a judgment about what promises to make. A Chef agent could execute resources in order and check whether each one needed convergence, but it could not look across the node’s full state and decide whether the recipe was even relevant.

So we wrote the observation for them. We called it “desired state” and stored it in files. The file was scaffolding for an agent that lacked the capacity to observe.

What changed#

Agents can observe now.

... continue reading