Your AI agent is probably not an agent
The word 'agent' has become meaningless. Everyone from chatbot vendors to autonomous system builders uses it. We've been here before — with self-driving cars — and it didn't end well.
Duration: 3:14 | Size: 3.7 MB
Here’s a fun exercise. Go to any tech company’s website and count how many times they use the word “agent.” Now try to figure out what they mean by it.
You won’t be able to, because the word has been stretched past the point of meaning anything. A chatbot that answers FAQ questions? Agent. An autocomplete that finishes your code? Agent. A system that decomposes problems, uses tools, adapts when steps fail, and maintains state across sessions? Also agent. One of these things is a calculator with a personality. The other is something genuinely new. Calling them the same thing isn’t just sloppy — it’s expensive.
Researchers have noticed. A paper out of Colorado State draws a hard line between “AI Agents” (modular, task-specific, LLM-driven) and “Agentic AI” (multi-agent collaboration, dynamic task decomposition, persistent memory, coordinated autonomy). These aren’t degrees of the same thing. They’re architecturally different systems with different failure modes. Agents hallucinate and get brittle. Agentic systems have emergent behavior and coordination failure. Treating them as interchangeable is like treating a calculator and a spreadsheet as the same product because they both do math.
Meanwhile, the Swarmia team and a group at Columbia have independently proposed five-level autonomy frameworks — think SAE Levels for AI. The levels are defined not by what the AI can do, but by what the human’s role becomes: operator, collaborator, consultant, approver, observer. Each step up means less human involvement between the AI receiving a goal and delivering a result.
And that’s where I start getting nervous. Because we’ve seen this movie before.
The Society of Automotive Engineers created Levels 0–5 for self-driving cars. The intention was clarity. What actually happened was that every car company claimed “Level 4 autonomy” while shipping what amounted to glorified cruise control. “Level 4” became a marketing term divorced from its technical meaning. People trusted it. Some of them died.
I’m not being dramatic. The stakes with AI agents are lower than with literal cars, but the pattern is identical: a taxonomy designed for engineers gets captured by marketing, and the gap between what people think they’re buying and what they’re actually getting widens until something breaks.
The Cloud Security Alliance already recognizes this. Their January 2026 guidance says different autonomy levels should require different authorization authority — and that Level 5 (fully autonomous) “is not appropriate for enterprise deployment today.” The Linux Foundation launched the Agentic AI Foundation in late 2025, trying to play the W3C role before the definitions calcify around whatever vendors find most profitable.
So what’s the practical takeaway? When someone sells you an “AI agent,” ask one question: what happens when it fails? If the answer is “it stops and asks you,” that’s a chatbot with extra steps. If the answer is “it tries a different approach, logs why, and keeps going,” you might be looking at something real. The failure mode tells you the autonomy level. The marketing never will.
Nobody takes you aside anymore
Print taught a generation when to stop. What we lose when the machines absorb the constraints that used to form us.
Your AI agents need a water cooler
Coordination is a property of the room, not the org chart. What that means when your coworkers are agents.
On the death of the author and the birth of the detector
Why worrying about AI authorship is lazier, and more prejudiced, than it looks.
The work of being available now
A book on AI, judgment, and staying human at work.
The practice of work in progress
Practical essays on how work actually gets done.
How to manage content for multiple clients without flattening their voices
How to manage content for multiple clients without their voices blurring into one house style: a workspace and a voice profile per client, batchable stages, and approval buffers.
Why does AI writing sound generic? It has nothing to work with
Why does AI writing sound generic? Because the model has none of your perspective, examples, constraints, or stakes to work with. The fix is interview-first, not better adjectives.
How to train AI to write in your voice, not your vibe
How to train AI to write in your voice isn't a prompt trick. It's a system: writing samples, interview answers, keep/avoid lists, revision loops, and approval gates.
The 19% slowdown nobody wants to talk about
Experienced developers are 19% slower with AI tools — and they don't even know it. The data says the productivity revolution isn't about faster code. It's about fixing the system around the code.
Manual fluency is the prerequisite for agent supervision
You cannot responsibly automate what you cannot do manually. AI agents speed up work for people who already know how to do it. They do not replace the need to learn the work in the first place.
Your process was built for a different speed
When work changes velocity, governance systems don't just fall behind. They become theater. And theater is worse than nothing—it gives you the feeling of control without any of the substance.