When someone tells you they’re “going all in on agentic AI,” check what they’ve already shipped.
I was in a conversation about a Head of AI role. The hiring manager was enthusiastic about building agentic systems, autonomous workflows, the whole vision. Then he mentioned the current state: nine use cases running on Dify — a low-code workflow platform.
That single fact told me more than anything else in the conversation.
Revealed preference is an economics concept: don’t listen to what people say they value, watch what they actually spend time and money on. In interviews, the equivalent is: don’t listen to what the team says they’re building next, look at what they’ve already built.
Nine Dify workflows means the team’s muscle memory is configuration, not engineering. “Going agentic” might mean adding a ReAct loop inside Dify, or it might mean a genuine platform shift — but the base rate says the former is far more likely.
This isn’t about Dify being bad. It’s about the gap between aspiration and revealed capability being the most diagnostic signal in any interview. A team that’s already built custom agents and wants to scale them is a different animal from a team that’s assembled low-code workflows and wants to graduate.
The practical test is simple. In any interview, ask: “What’s the most interesting thing the team shipped in the last year?” The answer — or the struggle to produce one — tells you what the role actually is, stripped of the job description’s ambition.
Stated intention is marketing. Revealed preference is the product.