Agent Concepts Overview

Skills give agents persistent capabilities — what they can DO. Concepts give them persistent pattern vocabulary — how they THINK. A catalog of recurring relational concepts — desire path, cargo cult, Chekhov’s gun, anchoring — that appear across many independent domains. They go by many names.

​

Why a catalog of concepts?

The goal is to aid recognition, transfer, and recombination.

• “Oh, there’s a name for that.” The shape you noticed but never quite named.
• “Oh, this same thing shows up over there.” A pattern from one domain you suddenly see in another.
• “That’s how I’d build this.” Combining named patterns to produce something new.

Once a pattern has a name, you can reason about it without rediscovering it. The desire path worn diagonally across a campus lawn is the same shape as the undocumented API call everyone uses instead of the sanctioned client, or the workaround a team actually follows instead of the official process. Once you’ve named desire path, the “where does actual use diverge from the designed route?” diagnostic carries across all three — and across the next domain you encounter it in. Once recognized, you can predict its consequences, infer its missing pieces from analogous instances, and combine it with other patterns to make new ones. The catalog is one attempt to make the patterns and their combinations available as a working concept vocabulary humans and AI agents can share.

​

How entries earn their place

• Cross-domain. The same structural shape recurs in genuinely distinct fields, not just within a single discipline.
• Compositional. Concepts combine — backpressure = flow + feedback-loop — so the catalog functions as primitives that build bigger structures.
• Empirically grounded. Each concept carries documented examples from across domains; a one-off observation is a candidate, not an entry.