If you’re an author, you already understand the difference between an assistant and an agent. You just might not realize it yet.
An assistant does what you ask. You say “format this manuscript” and they format the manuscript. You say “send this email” and they send it. They’re helpful, responsive, and always waiting for the next instruction.
A literary agent operates differently. You say “I want to sell this book” and they take it from there: reading the market, identifying editors who’d be a good fit, pitching on your behalf, negotiating terms, and circling back with a deal (or an honest conversation about why one didn’t happen). You define the goal. They figure out the steps.
That same distinction is now playing out in artificial intelligence. And it changes what you can ask these tools to do for you.
What Makes an Agent an Agent
A chatbot answers your question. An AI assistant follows your instructions. An AI agent takes a goal and works toward it, deciding what steps to take, what tools to use, and what to do when something doesn’t go according to plan.
The key difference is autonomy. When you ask ChatGPT to “rewrite this paragraph in a more suspenseful tone,” that’s an assistant interaction. One request, one response. But when you tell an AI agent to “research the top ten comp titles in cozy mystery from this year, summarize their themes, and draft a positioning statement for my novel,” the agent breaks that goal into subtasks, executes them in sequence, and delivers a finished result. It reasons about what to do next at each step, rather than waiting for you to spell out every move.
A Word That’s Meant “One Who Acts” for Six Centuries
“Agent” traces back to the Latin agere, meaning “to set in motion, to do, to perform.” It entered English in the late fifteenth century meaning simply “one who acts.” The word has always carried that sense of initiative, of a thing that does rather than a thing that waits.
In AI, the concept took formal shape in 1995. Stuart Russell and Peter Norvig published Artificial Intelligence: A Modern Approach, the textbook that would define the field for a generation of computer science students. They defined an agent as “anything that can perceive its environment through sensors and act upon that environment through actuators,” then went a step further, reframing the entire discipline of AI as “the study and design of rational agents.” That same year, researchers Michael Wooldridge and Nick Jennings proposed four properties that define an intelligent agent: autonomy (it controls its own actions), reactivity (it perceives and responds to changes), proactiveness (it takes initiative toward goals), and social ability (it can interact with other agents and humans).
But the intellectual roots go deeper. In 1973, MIT researcher Carl Hewitt developed the “actor model,” a framework where each independent computational unit receives messages, maintains its own internal state, and decides how to respond. Hewitt’s actors weren’t called agents, but they behaved like them. The idea that intelligence could emerge from autonomous entities making their own decisions, rather than from a single monolithic program following a script, is the direct ancestor of every AI agent running today.
For decades, all of this stayed safely inside research papers. Then 2023 happened. OpenAI introduced function calling, which let large language models autonomously use external tools. An open-source project called AutoGPT went viral by demonstrating a fully autonomous AI that could break a high-level goal into subtasks and execute them with minimal human oversight. Almost overnight, “AI agent” jumped from the academy into everyday conversation.
How Agents Actually Work
Four capabilities separate an agent from a standard chatbot.
Tool use. A chatbot can only generate text. An agent reaches beyond its own abilities by using external tools: searching the web, reading files, running code, accessing databases, even controlling software on your computer. Just as you wouldn’t try to format a manuscript without a word processor, agents use the right tool for each subtask rather than forcing everything through text generation alone.
Planning. When you give an agent a complex goal, it doesn’t attempt everything at once. It breaks the work into smaller steps, considers the best sequence, and adjusts when something unexpected happens. This is the reasoning layer, the part that makes an agent feel less like a search engine and more like a collaborator who can think a few moves ahead.
Memory. A standard chatbot forgets everything the moment your conversation ends. An agent can retain information across sessions, building up context about your preferences, your projects, and what has (and hasn’t) worked before.
Autonomy. Tie these three together and you get the defining trait: the ability to operate in a loop of perceiving, reasoning, acting, and observing the result, then adjusting course without needing a human to approve every step. You set the destination. The agent navigates.
Why This Matters for Your Writing Life
The tools you’re already using are moving in this direction. Claude can browse the web, write and execute code, and work through multi-step research tasks on your behalf. ChatGPT’s custom GPTs let you build specialized agents tuned to your workflow, whether that’s generating character profiles, analyzing pacing across chapters, or drafting query letters. Writing platforms like Sudowrite and NovelCrafter are building increasingly agentic features that go beyond “answer my question” into “manage this process for me.”
For authors, the practical shift is in what you can ask for. With a chatbot, you ask questions (“What’s a good synonym for ‘walked’?”). With an assistant, you delegate tasks (“Rewrite this scene from first person to third”). With an agent, you describe outcomes (“Research my genre’s bestselling cover designs from the past year, draft three concept briefs based on what’s trending, and outline back cover copy for each”). The scope of what you can hand off grows dramatically.
Understanding the distinction also helps you choose the right tool for the job. If you need a quick answer or a single rewrite, a chatbot works fine. If you need a complex, multi-step process handled from beginning to end, you want something with agentic capabilities. Knowing the vocabulary means knowing what to look for on the feature list.
The word “agent” has meant “one who acts” since the fifteenth century. In AI, it’s finally living up to the name.