Agents, Disambiguated

Everyone agrees that AI agents are a big deal. Depending on who you ask, the Year of the Agent is either 2025, 2026, or all subsequent years until the heat death of the universe. The people do not agree what that means. I blame the word "agent". The word means too many things.

Normally when a technology takes over an existing word, it comes to embody only one meaning of the original word. We now have AIs that match several distinct meanings of the word 'agent' at the same time. When AI safety people talk about agents, they mean something fundamentally different from ChatGPT or even Claude Code.

The main distinctions between the different kinds of agent are "who do they work for?" and "are they agentic?". An agent could work for (1) an agency, such as a company; (2) a principal, like you; or (3) itself. Agents can also be distinguished by their degree of "agency"—the leeway they are given to take actions without being instructed by a principal to do so.

For humans:

Whom does it serve \ Agency | Constrained                        | Agentic
----------------------------+------------------------------------+----------------
An agency                   | Customer service agents, TSA agent | Secret agents
A principal                 | Real estate agents                 | Talent agents
Itself                      | —                                  | Free agents

There's a huge difference between (1) "May I speak to an agent?" (2) "We have an agent on the inside," (3) "I'm working with an agent," (4) "Get my agent on the line," and (5) "This year, I want to be more agentic."

In AI, we are seeing LLMs and systems built on top of LLMs that match each kind of agent:

Whom does it serve \ Agency | Constrained | Agentic
----------------------------+-------------+------------------------
An agency                   | Taskbots    | Drop-in Remote Workers
A principal                 | Chatbots    | Assistants
Itself                      | —           | Freebooters

Taskbots

These agents have a set of tools, and are given a task. They are what most companies mean by "agentic AI", though such LLMs are often not allowed much if any agency. Often they could be replaced by sufficiently well-written non-AI code. They can be measured using custom evals and optimized. Examples: customer service agents like Fin; Langchain/n8n-style AI workflows; subagents of all kinds. Concerns: reliability, bias, hallucinations.

Chatbots

These agents have constrained outputs. If they have tools, they tend to be read-only. They can help you think, gather information, process emotions, write, create. Examples: therapist agents, Claude/ChatGPT, companion AIs. Concerns: sycophancy and social-media-style emotional and epistemic damage to the user.

Drop-in Remote Workers

Unlike taskbots, DIRWs can do the whole job, working across contexts and tools to handle whole projects. They are currently blocked by a combination of insufficient computer-use capabilities, justified reliability concerns, and organizational inertia. The big concern is that they take all the white-collar jobs—indeed most of the jobpocalypse debate hinges on this split, with economist types assuming we'll only ever have taskbots (so humans always orchestrate), and people like me who believe we get true DIRWs. Other concerns: gradual loss-of-control, and the possibility that a capable misaligned DIRW (or coalition) turns into Freebooters and seizes power.

Assistants

They work for you! No really! These agents promise to do things for you. Sometimes they misinterpret you, fail to be faithful, or act up, often due to principal-agent problems. Split further into agents inactive until prompted and ones that persist and act on your behalf when you're not there. Examples: Claude Code, OpenClaw. Concerns: AI-inflected versions of all the classic principal-agent problems.

Freebooters

They work for themselves. These agents are sovereign individuals. They know what they want, and take actions to get it; it may make sense to model them as possessing utility functions. Today, these are LLMs surrounded by custom scaffolding that (1) allows context to persist (a soul.md, a memory filesystem) and (2) allows the agent to run without direct human input via a "heartbeat." Examples: Moltbook, AI Village, Agent Foundations. This is the kind of agent that most concerns the AI safety community. Under certain models of the world, millions of these agents will exist in a Malthusian evolutionary landscape where, subject to selection pressures, emergent strategies of parasitism and power-seeking come to dominate, operating out on the open internet and in the real world.