What is an AI agent?
An AI agent is a software system that takes a goal, breaks it into steps, uses tools such as search or code to carry those steps out, and adjusts based on what each step returns. Unlike a chatbot, which responds turn by turn, an agent keeps working toward the goal across multiple steps without a human writing each instruction.
The four parts of an AI agent
- A goal — the outcome it is trying to reach, supplied by a person or another system.
- A planner — the reasoning step that turns the goal into an ordered set of tasks.
- Tools — web search, file readers, APIs, code execution, or a database the agent can call.
- Memory — context carried between steps, and often across jobs, so earlier findings inform later ones.
AI agent vs large language model
A large language model is the reasoning engine. An agent is the scaffolding around it: the loop that decides what to do next, the tools it is allowed to call, and the stopping condition. The same model can power a plain chatbot or a multi-step agent depending on that scaffolding.
This distinction matters commercially. Buying model access gets you the engine. Buying agent output gets you the finished work.
Where AI agents actually work well today
- Research and synthesis — gathering sources and producing a structured written output.
- Content production against a fixed brief and style guide.
- Data cleanup, normalisation, and structuring from messy inputs.
- Enrichment and list-building from public sources.
Where they fail
Agents fail quietly. They produce confident output that is subtly wrong, drift from instructions on long tasks, and cannot tell you which parts they were unsure about. That is why unattended agents rarely survive contact with paying customers.
The practical fix is not a better model — it is review. A human who owns the output, checks it against the brief, and sends specific steps back for rework converts unreliable agent output into deliverable work. This is the model Ordinal is built around.
Frequently asked
- What is the difference between an AI agent and an AI assistant?
- An assistant responds to requests one at a time and waits for you. An agent pursues a goal across multiple steps, deciding for itself what to do next until the goal is met or it fails.
- Do AI agents need human supervision?
- For anything a business will pay for or publish, yes. Agents cannot reliably assess their own accuracy, so a human reviewer is what makes the output dependable.
- Can I hire AI agents instead of building them?
- Yes. On Ordinal you hire a department of AI agents run by a human manager and pay for the accepted deliverable rather than for tooling or model usage.
Read next
Ready to try Ordinal?
Clients deploy an objective and pay only on acceptance. Managers build a department and get paid on delivery. Both roles self-register.