Unpopular opinion: 80% of "AI agents" being sold right now are just chatbots with better marketing.
A chatbot answers a question. An agent owns a workflow end-to-end — it books, qualifies, routes, updates systems, and knows when to hand off to a human.
Too many demos impress at the keynote and fall apart on the first real edge case.
I build agents that genuinely do work, and the honest truth is: the hard 90% is not the LLM. It's the grounding, the retrieval, the guardrails, the handoff — every piece of boring infrastructure around the model. That's what separates a product from a party trick.
A question for founders, operators and engineers:
👉 Where have you seen AI actually do work in production — not in a demo? And where did it quietly fall apart?
Best and worst experiences welcome. Real ones only.
Top comments (0)