A finance team asks "process this invoice." A chatbot writes a nice email about invoice processing. An agent reads the invoice, matches it against the purchase order, routes any discrepancy to the right approver, and returns a receipt. That is the difference.
What are the four capabilities that define an AI agent?
The four capabilities that separate an agent from a language model with a chat interface:
- Tool use. The agent can call real APIs, read a database, browse the web, or execute code. Without tool use, it is a text generator.
- Memory. The agent remembers prior context across turns or sessions. Short-term memory is table stakes. Long-term memory, grounded in a vector store or structured store, is what makes it feel competent.
- Autonomous planning. The agent decomposes a goal into steps and picks the right tool for each one. It is not following a script you wrote.
- Verifiable output. The agent returns evidence: a transaction ID, a changed record, a URL, a receipt. If you cannot verify what it did, you cannot trust it in production.
When is an agent the wrong tool?
Three signals that an agent is overkill:
- The workflow is already rule-based. A deterministic automation is cheaper, faster, and does not hallucinate.
- The latency budget is tight. Agents loop. Loops add latency. If the user is waiting under 500ms, use a classical API.
- The failure cost is high and oversight is low. Agents are statistical. Statistical systems need evaluation harnesses, cost caps, and escalation paths. If you do not have budget for those, do not ship an agent.
FAQ
Is ChatGPT an AI agent?
ChatGPT in its default chat mode is a language model with a UI. When it uses tools, browses the web, or executes code on your behalf, it operates as an agent for that session.
Do I need an agent or an automation?
If the task has rules and runs on a schedule, an automation is cheaper and more reliable. If the task needs reasoning across unstructured inputs, an agent is the right fit.