What Is an Agent? The Business and Legal Definition
August 12, 2026
An agent is a person or entity authorised to act on behalf of another — the principal — in dealings with third parties. This is the law of agency, and it long predates the software sense of the word.
The defining feature isn't doing work for someone. It's that the agent's actions legally bind the principal. A contractor who builds your website performs a service. An agent who signs a contract in your name has committed you to it.
The essentials:
- Authority is the whole concept. Without it, an agent is just a person making promises they can't keep.
- Authority can be actual or apparent, and the second one catches people out — a principal can be bound by authority they never actually granted.
- Agency creates duties, chiefly loyalty and care, running from agent to principal.
- "AI agent" borrows the word by analogy, and means something meaningfully different. Disambiguation at the end.
The relationship
Agency has three parties, and the third is why it matters.
The principal grants authority and is bound by what the agent does within it.
The agent acts on the principal's behalf, and generally isn't personally liable on contracts made within their authority.
The third party deals with the agent but ends up in a relationship with the principal.
That last point is the practical consequence. When a purchasing manager signs a supply contract within their authority, the company owes the money — not the manager. When a freight forwarder books cargo space on a shipper's behalf, the shipper is liable for the freight charges.
Property is the exception that proves how much the formalities matter. In England and Wales an accepted offer on a house binds nobody: a contract for the sale of land must be in writing, incorporate all agreed terms, and be signed by both parties under section 2 of the Law of Property (Miscellaneous Provisions) Act 1989. Until contracts are exchanged, either side can walk away — so an estate agent communicating acceptance has not bound the seller to anything.
Types of authority
This is the part that generates most disputes.
| Type | How it arises | Example |
|---|---|---|
| Express actual | Stated explicitly, written or spoken | "You may sign contracts up to £50,000" |
| Implied actual | Reasonably necessary to carry out express authority | A shop manager ordering routine stock |
| Apparent (ostensible) | The principal's conduct leads a third party to reasonably believe authority exists | A former employee still holding company cards and an office |
| Ratified | No authority at the time, but the principal approves afterwards | A deal signed without permission, then accepted |
Apparent authority is the one that surprises people. A principal can be bound by authority they never granted, if their own conduct led a third party to reasonably believe it existed. Leaving someone in a role, letting them keep credentials, or continuing to honour their past deals can all create it.
The practical implication: when authority ends, tell the people who relied on it. Internal revocation isn't enough — apparent authority persists until the third party has reason to know otherwise.
Duties an agent owes
Agency is a fiduciary relationship, which sets a higher bar than ordinary contract:
Loyalty. Act in the principal's interest, not their own. No secret profits, no undisclosed conflicts, no acting for both sides without informed consent.
Care and skill. Exercise the competence reasonably expected for the role.
Obedience. Follow lawful instructions, even ones the agent thinks are unwise.
Disclosure. Pass on material information. Knowledge the agent acquires is generally attributed to the principal — which cuts both ways, since a principal can be treated as knowing something only their agent knew.
Accounting. Keep the principal's property and money separate and account for it.
In return the principal owes compensation, reimbursement of proper expenses, and indemnity for liabilities properly incurred.
Common business agents
- Estate agents — authority to market a property and communicate offers, not to conclude a sale; in England and Wales that needs an exchange of signed contracts
- Sales representatives — often authority to solicit orders but not to conclude contracts
- Freight forwarders and customs brokers — contracting with carriers on the shipper's behalf
- Insurance agents — authority varies sharply; a broker generally represents the insured, an agent the insurer
- Officers and employees — authority scaling with role, which is precisely where apparent authority arguments arise
- Attorneys-in-fact — authority under a power of attorney, express and usually narrow
The recurring theme: the title tells you little; the grant of authority tells you everything. "Sales agent" describes a job, not a scope.
"Agent" in AI means something different
The word has been borrowed, and the two meanings now collide in search results often enough to be worth separating.
A legal or business agent is a person or entity with authority to bind a principal. The concept is about authorisation and consequence.
An AI agent is software that pursues a goal by deciding its own steps and calling tools — the distinction Anthropic draws between workflows and agents, and the framing IBM uses for AI agents. The concept is about autonomy in execution.
Two things are worth separating here, because they get conflated constantly.
Software has no legal personality. It cannot hold authority of its own, cannot be a party to a contract, and owes no fiduciary duties. In that sense an AI agent is not an agent at all — there is no principal-agent relationship between you and your software.
That does not mean automated actions have no legal effect. A system acting on your instructions can still form contracts that bind you. The Law Commission's work on smart contracts concluded that contracts defined and performed by computer programs can be legally binding under existing English law, and the same logic applies to an automated purchasing system placing an order. The obligation runs to the operator, not the software.
So the practical position is narrower than "AI can't bind anyone": it can commit you, and you carry the consequences. That's an argument for scoping what a system is permitted to do and keeping a person on anything consequential — which is why the AI employees framing oversells, and why accountability can't be delegated to software.
If you arrived here looking for the AI meaning, agentic workflows covers what those systems actually do, and AI agent builders covers how they're built.
FAQ
What is an agent in simple terms?
Someone authorised to act on another person's behalf in dealings with third parties, in a way that legally binds the person they represent.
What's the difference between an agent and an employee?
An employee works under the employer's control; an agent has authority to bind the principal in dealings with others. Many employees are also agents, but the categories are separate — an agent can be an independent contractor, and an employee may hold no authority to bind anyone.
What's the difference between an agent and a broker?
An agent represents one party and owes that party fiduciary duties. A broker typically facilitates a transaction between parties. The distinction matters most in insurance and real estate, where the labels carry specific meanings that vary by jurisdiction.
Can a company be an agent?
Yes. Agents can be individuals or entities — freight forwarders, advertising agencies and collection agencies are all companies acting as agents.
Is an AI agent an agent in the legal sense?
No — software has no legal personality, so it holds no authority of its own and owes no fiduciary duties. But that is not the same as saying an automated action has no legal effect: a system acting on your instructions can form contracts that bind you, and the Law Commission has concluded that smart contracts can be legally binding under existing English law. The obligation and the liability sit with the operator, not the software.