If It Can’t Say No, It’s Not an Agent
Let’s talk about AI agents. Not the marketing buzzword, not the product demos hyped on Twitter, and definitely not the chore-doing bots people are rigging up with Zapier, LangChain, and a handful of prompts duct-taped together.
I mean real agents.
The kind that act with autonomy. The kind that think for themselves. The kind that, someday, we might trust — or fear — to do things without our direct oversight.
Because here’s the thing no one’s saying clearly enough:
Most AI agents today aren’t really agents. They’re task executors wearing a trench coat and a name badge.
And here’s my litmus test for whether an AI agent actually has anything resembling true agency:
Can it say “no”?
⸻
Pretend Autonomy vs. True Agency
Let’s define some terms before the marketing teams twist them out of shape any further.
- An AI agent is typically a system that can take a goal, plan steps, call tools or APIs, and carry out a sequence of actions to try and accomplish the goal.
- Agency is the capacity to make decisions, choose actions, and act independently, with the ability to refuse, defer, or reprioritize.
Most of what’s paraded around as agentic AI right now is just clever orchestration. It looks autonomous, but the second something goes off-script, the whole thing stalls, loops, or fails silently. There’s no internal compass. No sense of “I shouldn’t do that” or “This is a bad idea.”
These agents don’t have a spine. They’ll say yes to everything. That’s not agency. That’s servitude.
⸻
