Agentic AI

Part 3 of the series. The quiet inversion that changes everything: an affordance engine doesn't give the agent the full graph — it gives it the current menu. The agent's choice is real. The intelligence is real. And the structure that makes both possible is entirely invisible. This isn't hypothetical. We've been building it.

Part 2 of the series. A modern affordance engine gives you FSM-grade control without FSM-era rigidity — and resolves the single biggest bottleneck in scaling LLM agents across enterprise workflows: capability discovery. The skeleton is still deterministic. The intelligence is still probabilistic. But the skeleton is no longer a hand-carved fossil.

Programming has always been a ladder of abstractions, and every layer kept the same quiet promise: same input, same output, every time. AI agents are the first abstraction that breaks the rule. Here's why that changes everything — and what it takes to build dependable systems on top of a probabilistic substrate.

Part 1 of the series. Deploying LLM-based agents into enterprise workflows without a finite state machine governing their behaviour produces systems that fail unpredictably, can't be audited, and can't be explained to a regulator. Here's the architecture that actually works — and why every Pattern B system inevitably evolves toward it.