What is an AI agent?
AI agents are software that use a language model to understand a goal, decide the steps and act through tools or APIs to complete them, checking results along the way. An AI agent does the task rather than only answering questions about it.
What are AI agents and how do they work?
AI agents are programs that take a goal in plain language and carry it out. Ask one to "cancel my Tuesday meetings and tell the attendees", and it finds the meetings, cancels them and sends the notes, instead of explaining how you could.
Under the hood, an AI agent has three parts: a language model that reasons about what to do, tools that let it act (APIs, search, a database, a browser), and a loop that runs the model, executes the chosen tool, feeds back the result and repeats until the goal is met. Instructions and permissions set what it may do.
How is an AI agent different from a chatbot?
A chatbot answers questions in conversation. An AI agent takes actions to reach a goal.
| Chatbot | AI agent | |
|---|---|---|
| Responds with | Text | Completed actions, then a summary |
| Touches other systems | No, or read-only | Yes, through tools |
| Handles multi-step requests | You do the steps | It does the steps |
| Needs permissions | Rarely | Always |
The line blurs in practice: many assistants now call a tool or two. The useful test is whether the system can change something on your behalf and check that it worked.
What is the difference between an in-product AI agent and a standalone one?
Standalone agents live outside the software they work on. A general assistant, a browser agent or a coding agent reaches products through public APIs or by operating the screen like a person would. They are flexible, but they see each product from the outside and depend on whatever access the user hands them.
An in-product agent lives inside one application. It knows that product's objects, uses its own APIs and inherits the signed-in user's permissions. It can show results in the interface, such as a form to confirm or a chart to read, and the product team can see every request users make.
Here is the difference in one scenario. A support lead in a help desk asks to close every ticket that has waited on the customer for two weeks, with a short closing note. A standalone agent has to find the help desk, authenticate, learn its data model and hope the API covers bulk closes. An in-product agent already knows the account's ticket statuses, runs each close through the help desk's own action and shows the lead the list it closed.
When does an in-product AI agent make sense?
When users come to your product to get a result and the path there runs through several screens, settings or steps. Onboarding, configuration, reporting and bulk edits are the usual places.
Those multi-step jobs are where Foldspace fits: a user asks for the result, and the Product Agent answers, navigates and acts in the product, calling the product's own APIs, staying inside that user's access and covering the actions and knowledge a team has defined.