AI agent
An AI agent is software that is given a goal and works out its own steps, using tools it can reach. An automation that follows fixed steps is a different thing, because a person set the path in advance. Agents still need goals and rules from people. When someone sells you an agent, ask what it can reach, what it can change and when it stops to ask.
- RequestA person gives it a goal.
- Chosen stepsIt decides for itself what to do next.
- Connected toolsIt uses only the tools it has been connected to.
- ApprovalIt should stop and ask a person before anything that matters. Ask whether it does.
A person gives the agent a goal and the agent chooses its own steps using the tools it has been connected to. It should stop to ask a person before anything that matters, and whether a given tool does is a question to ask.
What the word means
Give it a goal and it works out what to do with the tools it has. That is the thread through both big-company definitions: Google Cloud describes AI agents as software systems that use AI to pursue goals and complete tasks on behalf of users. IBM describes an AI agent as a system that autonomously performs tasks by designing workflows with available tools.
Sources: Google Cloud, read 11 Oct 2026; IBM, read 11 Oct 2026
What it does not mean
IBM also says that although AI agents are autonomous in their decision-making, they still need goals and rules defined by humans. So the word does not mean a tool that decides for itself what your business should do.
Source: IBM, read 11 Oct 2026
Agent is a loose word, and on its own it tells you nothing about what a tool is allowed to touch. The useful questions are what it can reach, what it can change, and what happens when it is stuck.
Fixed steps against chosen steps
Anthropic separates two kinds of agentic system: workflows, where models and tools are orchestrated through predefined code paths, and agents, where the model dynamically directs its own process and tool use. In plain words: steps a person wrote down in advance, against steps the tool picks as it goes.
Source: Anthropic, read 11 Oct 2026
| The job | Fixed steps | Chosen steps |
|---|---|---|
| Text a customer the day before their appointment | A rule does it, with the same message every time. | Not needed. |
| Send a reminder when an invoice goes overdue | A rule does it, and your accounting software may already have the setting. | Not needed. |
| Read a messy customer email, work out what they want and decide which tools to check | Hard to write down as fixed steps in advance. | This is where an agent may earn its place. |
A made-up example
Example, made upThis is a made-up example. A cafe owner asks a tool to sort out this week's supplier orders. A fixed-steps version pulls the stock list and sends the same email to each supplier, every time. An agent version decides for itself which supplier to check first, which tool to open and when to stop. If it was built to pause, it asks the owner before it sends anything to a supplier.
Limits, and when a simpler option is enough
Anthropic says agentic systems often trade latency (waiting time) and cost for better task performance, and that the autonomous nature of agents means higher costs and the potential for compounding errors.
Source: Anthropic, read 11 Oct 2026
Anthropic recommends finding the simplest solution possible and only adding complexity when needed, which it says might mean not building agentic systems at all.
Source: Anthropic, read 11 Oct 2026
Anthropic says an agent can pause for human feedback at checkpoints or when it hits a blocker. Whether a particular tool does is a question to ask, not something to assume.
Source: Anthropic, read 11 Oct 2026
Questions to ask a seller who uses the word
- What can it reach, and what can it change?
- What happens when a request is outside what it knows?
- Where does it stop and ask a person before it acts?
- Could fixed steps do this job, and what does the agent add that they do not?
Sources and checking
- Google Cloud: What are AI agents?, read 11 Oct 2026
- IBM: What are AI agents?, read 11 Oct 2026
- Anthropic: Building effective agents, read 11 Oct 2026
Last checked against the sources: 11 Oct 2026. Where a source is a seller, what it says is the seller’s statement, not a measured result. Tax is shown exactly as the source gives it.
Published by Three Fixes. An AI drafted this page from the sources above, and a separate review pass checked it against them on 11 Oct 2026.
The next step
Read a sample plan
See what a Three Fixes starting plan looks like before you decide anything.
Read a sample plan