Where IBM and Anthropic draw the line between an AI agent and a chatbot

IBM says assistants and chatbots react while agents plan and act toward a goal. Anthropic says agents choose their own steps where workflows follow fixed code. Gartner says many products are relabelled chatbots. Read on 2026-11-06.

By Mackenzie Bowes · mackenzie@mackenziebowes.com · @_mackenziebowes on X

Contents

The two companies whose definitions are read here draw the line in different places. IBM puts it between a system that waits for your request and one that works toward a goal on its own. Anthropic puts it between a system whose steps are fixed in code and one whose model chooses the steps. Both lines agree on one thing: a chatbot that only answers each message is not an agent. Neither source is neutral, and no standards body defines the terms in the sources read here.

IBM draws it as reactive or proactive

IBM's page on agents and assistants says: "AI assistants are reactive, performing tasks at your request. AI agents are primarily proactive, autonomously planning and taking actions to achieve a defined goal using the tools and permissions available to them."[1] It lists ChatGPT, Microsoft Copilot, Salesforce Agentforce and IBM's own watsonx Assistant as chatbot examples, and says an assistant "waits for permission to act on the recommendation."[1] An agent, in IBM's words, "refers to a system or program that can autonomously complete tasks on behalf of users or other systems by planning its own workflow and using available tools."[1] After an initial prompt, IBM says, agents "can continue working without further input."[1]

Anthropic draws it as fixed steps or chosen steps

Anthropic's article separates two kinds of what it calls agentic systems: "Workflows are systems where LLMs and tools are orchestrated through predefined code paths." And: "Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks."[2] The article is from 19 December 2024 and carries a note that much of the tooling it describes has changed since.[2] It does not define the word chatbot. It also advises "finding the simplest solution possible", noting that "Agentic systems often trade latency and cost for better task performance."[2]

Why ChatGPT is not one or the other

IBM lists ChatGPT as a chatbot.[1] A comparison page from the analyst site AIMultiple, updated 10 August 2026, says that "In August 2025, OpenAI retired Operator and folded its abilities into ChatGPT Agent."[3] Synthesis, this page's own reading: the product name ChatGPT now covers a chat window and, in the same product, an agent feature, so the question to ask is which mode a task runs in, not which product.

Relabelling

Gartner's June 2025 press release describes "agent washing", "the rebranding of existing products, such as AI assistants, robotic process automation (RPA) and chatbots, without substantial agentic capabilities", and estimates that "only about 130 of the thousands of agentic AI vendors are real."[4] It is an estimate, and the release shows no method.

What it costs to be an agent

IBM names the trade-off: "If an AI agent has trouble creating an effective plan or evaluating its progress, it may get stuck repeating the same actions or fail to complete its assigned task."[1] It says assistants "have fewer dependencies than AI agents and typically execute simpler workflows, making them easier to deploy and manage."[1] For measured results of agents on real tasks, see Do AI agents actually work?.

A working test

Synthesis, this page's own test and not a source's claim: ask who chooses the next step. If every step can be written down in advance, it is a workflow. If it only answers the message in front of it, it is a chatbot or assistant. If the system picks its own next step and its own tools toward a goal, it is an agent in both IBM's and Anthropic's senses.

Not yet proven

If you are choosing between a chatbot and an agent for a business task and can say what tipped it, I would like to compare notes: mackenzie@mackenziebowes.com.

Sources

  1. IBM, AI Agents vs. AI Assistants, IBM Think, read 2026-11-06, undated. www.ibm.com/think/topics/ai-agents-vs-ai-assistants
  2. Anthropic, Building effective agents, 2024-12-19. www.anthropic.com/research/building-effective-agents
  3. AIMultiple (Cem Dilmegani and Ezgi Arslan), Top 30+ Agentic AI Companies, updated 2026-08-10. aimultiple.com/agentic-ai-companies
  4. Gartner, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, press release, 2025-06-25. www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027

Terms on this page

  • Agentic AI: AI that works toward a goal by planning steps, making decisions and acting with more or less independence.
  • AI agent: A system where a language model decides for itself how to do a task and which tools to use.