The Limits of a Chatbot
Chatbots are everywhere. You can ask one to explain photosynthesis, write a poem about your dog, or translate a sentence into Japanese — and it will do it in seconds. That feels pretty powerful. But behind that impressive surface is a fundamental limitation: a basic chatbot only produces text. It cannot do anything else. Understanding that boundary is the starting point for understanding what an AI agent is and why agents had to be invented.
What a Language Model Actually Does
A chatbot is powered by a large language model, or LLM. An LLM is trained on enormous amounts of text — books, websites, code, articles — and learns the statistical patterns of how words and ideas follow each other. When you send a message, the model generates the most plausible next sequence of words given everything in the conversation so far. That is literally all it does: predict and produce text, one token at a time. This is genuinely remarkable. The model has absorbed so much knowledge that its text predictions often contain accurate facts, logical reasoning, and creative ideas. But notice what is missing: the model does not look anything up right now, does not open a file, does not send a message, and cannot remember last Tuesday's conversation. It sits in a bubble, taking in text and producing text.
A large language model is a neural network trained on vast amounts of text to predict the most likely next words given a context. It generates its responses entirely from learned patterns — it is not reading the internet live, and it has no hands to take action in the world.
The Three Core Limits
Chatbots without any extra machinery face three hard limits that matter enormously for real tasks. First: no live information. The model's training data has a cutoff date. It does not know what happened yesterday. If you ask it today's stock price or tomorrow's weather, it cannot actually check — it can only guess based on patterns or admit it does not know. Second: no persistent memory. Each conversation typically starts fresh. The chatbot does not remember that last week you told it your name, your project goals, or what you decided. When the context window closes, everything is gone. Third: no ability to act. A chatbot cannot send your email, save a file, book a flight, run a calculation in a spreadsheet, or search a database. It can tell you HOW to do those things — it can even write code that would do them — but it cannot execute them on your behalf.
A chatbot can write a perfectly correct email draft and pretend to send it — but it has not sent anything. It can write code that would book a flight — but no flight is booked. Producing text that describes an action is not the same as taking the action. This distinction is the heart of this module.
Match each chatbot limit to the real-world problem it causes.
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
Where Chatbots Still Excel
None of this means chatbots are weak or unimportant. For tasks that are genuinely about text — explaining concepts, drafting writing, brainstorming ideas, summarizing documents, tutoring, translation, coding assistance — chatbots are extraordinarily capable. The limits only bite when the task requires touching the world beyond the conversation window. If all you need is great text, a chatbot is the right tool. If you need the text to trigger real action, connect to live data, remember you across sessions, or complete a multi-step workflow, you need something more. You need an agent.
Chatbots are not broken because they cannot act — they were designed for conversation. The limits are not bugs; they are design choices that made LLMs safer and simpler to deploy. Agents extend chatbots with new capabilities rather than replacing them.
A student asks a chatbot for today's top news story. The chatbot gives a confident-sounding answer, but the story it describes happened eight months ago. What chatbot limit explains this?
Which of the following tasks is a basic chatbot WITHOUT any tools genuinely capable of completing on its own?
Find the Limit
- Step 1: Think of three things you have asked (or would want to ask) a chatbot to do.
- Step 2: For each one, decide: is it a pure text task, or does it require the chatbot to act in the world (access live data, remember you, or trigger an action)?
- Step 3: For any task that hits a chatbot limit, rewrite the request as if you had a more powerful AI system. What extra capability would it need?
- Step 4: Share your examples with a partner. Together, agree on which single chatbot limit would be most annoying to live with — and why.