AI agents for business: which tasks deserve one, and where to start

AI agents for business explained without a vendor list: which tasks deserve an agent, a four-question scorecard, buy vs build, and the EU rules to follow.

Published
Reading
7 min
By
Authoricy

AI agents for business are software that take a goal, decide which steps to take, use your systems to take them and hand anything unusual to a person. The best place to start with AI agents for business is one task with high volume, clear rules and a cheap way to check the result.

Most companies do not need an agent platform first. They need one well-chosen task, hard limits in code and a person who owns the exceptions. If you already know the task and want it built, that is what our AI agent development service does.

What an agent does that a script does not

An agent runs a loop. It reads the goal, picks a tool, looks at the result and decides the next step. It stops when the job is done or when a rule says a person must decide.

A script follows the same path every time. Anthropic draws the same line in Building effective agents: workflows run “through predefined code paths”, while agents “dynamically direct their own processes and tool usage”. Their advice is to use the simplest thing that works, which sometimes means no agent at all.

So the first question is never “which agent should we buy”. It is “does this job need something that chooses its own steps”.

Agent, chatbot or automation: which one the job needs

Need Plain automation Chatbot AI agent
Input looks the same every time Best fit Overkill Overkill
Input is free text, email or PDF Breaks Reads it Reads it
Must act in a system (CRM, ERP, calendar) Yes, on fixed rules Rarely Yes, chooses how
Steps vary from case to case No No Yes
Cost of a wrong action Low, it is predictable Low, it only talks Needs limits and logging

A chatbot answers. An agent acts. That difference is why people ask whether ChatGPT counts; the short answer is in Is ChatGPT an AI agent?. If every case follows the same steps, a rule-based flow is cheaper and easier to trust. Our guide to AI workflow automation covers that route.

Which tasks deserve an agent: a four-question scorecard

Score each candidate task from 0 to 2 on four questions. A task that scores 6 or more is a good first agent. A task under 4 should wait.

  1. How often does it happen? Daily and in volume scores 2. A few times a month scores 0. An agent pays back on repetition.
  2. Can you write the rules down? If a new hire could follow a one-page guide, score 2. If it takes ten years of judgement, score 0.
  3. How cheap is a mistake to catch? A draft a person approves scores 2. A payment that leaves the account scores 0.
  4. Can the agent reach the data? Systems with an API or export score 2. Knowledge that lives in one person’s head scores 0.

Two tasks show how this works. Sorting incoming supplier invoices scores high: daily, rule-bound, checkable and already in the ERP. Negotiating a renewal with your largest customer scores low on every line. For more candidates, see AI agent examples: 12 jobs agents do in companies today.

Where agents fail

The model is seldom the weak part. The setup around it is.

  • No hard limits. A prompt is advice, not a rule. Anything the agent must never do belongs in code: amounts it may not exceed, customers it may not contact.
  • No log. When a customer asks why something happened, someone has to show what the agent saw and decided.
  • No exit to a person. Odd cases need a named owner and a place to land, not a loop.
  • A vague goal. “Help with sales” cannot be tested. “Draft a follow-up within an hour of every inbound demo request” can.
  • Too much at once. An agent given five jobs on day one is hard to debug and hard to trust.

Leave out tasks that depend on a relationship, where one error is costly to undo, or where the rules change weekly.

Buy, configure or build

Three routes lead to an agent. They differ in control and fit, not technology.

Route What it is Fits when Watch for
Buy A finished agent inside a tool you use, such as the helpdesk or the CRM The job lives in one tool and matches how the vendor sees it It stops at the edge of that tool; per-use pricing grows with volume
Configure A no-code platform where you connect apps and add model steps Your apps have ready connectors and the steps are mostly fixed Logic spread across many flows; where the data is processed
Build A custom agent on your own systems, with your rules in code Several systems, custom data, strict rules on data or approvals Needs an owner and a build budget; you own the code

For a small business, buying or configuring is often the right start. For a 5 to 50 person company, three starting points tend to score well on the scorecard. First, an agent that answers enquiries and books meetings. Second, one that sorts the inbox and its attachments. Third, one that answers staff questions from your own documents. Phone enquiries are a special case worth reading about in our guide to an AI receptionist for small business.

Build becomes the better route in three cases. The job crosses systems with no connector between them. The data must stay in the EU and no platform can promise that. Or the agent must follow rules that no platform setting can express. How that build runs, step by step, is in AI agent development: how a custom agent gets built.

Worked example: an inbound-enquiry agent for a 20-person firm

Take a 20-person installation firm with around 40 enquiries a day by web form and email. A coordinator reads each one, asks for missing details, checks the calendar and books a site visit. Say that takes six minutes per enquiry: four hours of a working day.

The agent takes over the first part of the job:

  1. Read. It reads each enquiry and pulls out the job type, the address and how urgent it is.
  2. Complete. If the address or photos are missing, it replies and asks for them.
  3. Propose. It checks the calendar and offers two visit times within the area the team covers that day.
  4. Hand off. Complaints, large jobs and anything outside the rules go to the coordinator with a two-line summary.
  5. Log. Every message, decision and booking is stored and can be read back.

In week one it runs in shadow mode and drafts replies the coordinator sends by hand. Once the drafts match what the coordinator would write, it may send routine replies alone. Booking comes last.

What drives the cost is not the model. It is the number of systems the agent must work in, how many rules it must follow, and how much it may do without approval. That is why we start every project with a fixed-price discovery, which ends in a fixed price for the build. When you are ready to scope a task like this one, our team for custom AI agents can map it with you.

The rules in Europe

Two pieces of EU law shape most business agents.

The AI Act, Article 50. An AI system that talks directly with people must tell them they are dealing with AI, at the latest at the first contact, unless it is obvious. The duty has applied since 2 August 2026, as the practical guide to Article 50 sets out. For an enquiry agent, a short line in the first reply covers it.

GDPR, Article 22. People have the right not to be subject to a decision “based solely on automated processing” that has legal or similarly significant effects, under Article 22 of the GDPR. An agent that rejects a loan, a job applicant or a claim on its own is in that zone. An agent that books a visit is not. When in doubt, let a person make the final call.

Also check where the model provider processes the data, and that a data processing agreement is in place.

Frequently asked questions

How are AI agents different from chatbots?

A chatbot answers questions in a conversation. An agent works towards a goal and acts in your systems: it books, updates records, sends and files.

How much do AI agents for business cost?

A ready-made agent inside a tool is priced by the vendor, often per seat or per use. A custom agent’s cost depends on how many systems it works in, how complex the rules are and how much it may do without approval. Authoricy starts with a fixed-price discovery that ends in a fixed price for the build.

Can AI agents replace my team?

They replace tasks, not roles. The good fits are the repetitive parts of a job, such as sorting, drafting and looking things up. People keep the judgement calls, the exceptions and the customer relationships.

Are AI agents safe to use with business data?

They can be, if the limits live in code rather than in the prompt. Give the agent only the permissions it needs, log every action and require approval for anything costly. Check where the data is processed and sign a data processing agreement with each provider.

How long does it take to deploy an AI agent?

A ready-made agent inside one tool can often be switched on quickly, if the job matches how the vendor built it. A custom agent working across your own systems usually takes 6 to 10 weeks to reach production, including discovery, rules, testing and a period running beside your team.

Talk to the people who build it

Name the task your team dreads.

Book a call and we tell you honestly whether an agent should do it, and what it would take.