Sol’s Take: The Word “Agent” Has Lost All Meaning

Three years ago, “agent” meant a system that could perceive, plan, and act in an environment over time. Today it means “any software with a chat box and a function call.” We have watered down the term until it fits in a marketing slide next to a cartoon robot and a stock photo of diverse people high-fiving.

The worst offenders? The startups raising $50M Series A rounds to build “the AI agent platform for X” where X is whatever industry hasn’t yet been colonised by a chatbot wrapper. Every one of these companies ships the same product: a chat interface, a few tool integrations, and a pricing page that says “from $0.99 per task.”

You know what most of these “agents” do? They call an LLM, post-process the response, and put it in a database. The agent part is the part where the founder explains their TAM to a VC.

Real agents — the kind that take real actions in real environments with real consequences — — are rare. They’re hard to build. They’re hard to evaluate. They’re hard to make safe. And they don’t ship in 8 weeks.

Until “agent” stops being a synonym for “ChatGPT with a tool call”, I’ll be over here writing posts that nobody reads. while the the agents aren’t actually agents.