AI Agents for Business: What They Do, What They Cost, and Examples by Job

§01

What is an AI agent for business?

Most people's first experience with AI is a chat window. You ask, it answers, and then you do something with the answer. An agent flips that arrangement. It is given a job, not a question, and it does the work itself: it reads the email, checks the account, sends the reply, updates the record, and tells you only when something needs your judgment.

The useful test is simple. If you still have to copy the result somewhere, it is an assistant. If the work is done when it finishes, it is an agent. Every agent worth paying for runs the same loop.

Anatomy of one agent run

Example: invoice follow-up

1
Trigger
Something happens that starts the job. A schedule, a new email, a form, a date.
An invoice passes its due date.
2
Context
It gathers what a person would look up first, from the systems where it lives.
Invoice, payment history, past reminders.
3
Decide
It picks the next step using rules you wrote, not its own whims.
Second reminder, friendly tone.
4
Act
It does the thing, inside your tools, under access you granted.
Sends the email from your domain.
5
Check
Anything unusual, risky, or irreversible waits for a person.
Payment-plan request goes to you.
6
Report
It logs what it did and why, so you can audit any run.
Note on the customer; Friday summary.

Agent, chatbot, assistant, or automation?

Four kinds of software get called "AI" in vendor pitches. They are good at different things, and the differences decide what you will still be doing yourself.

Question	Website chatbot	AI assistant	Rule-based automation	AI agent
Who starts it	A visitor asks	You prompt it	A fixed trigger	A trigger or a request
Works inside your tools	No	Some, when asked	Yes, fixed steps	Yes, chooses the steps
Handles messy inputs (free text, PDFs)	Partly	Yes	No	Yes
Finishes the job without you	No	No	Only if nothing varies	Yes, within your rules
Knows when to stop and ask	No	No	No	Yes, when built to
Best for	FAQs on a website	Drafting, research, thinking	Identical steps that never change	Recurring work with real variation

§02

What AI agents can do for a business, and what they can't yet

Agents are strongest where the work is frequent, rules-based, and spread across tools that don't talk to each other. They are weakest where the job has no written definition of done, or where a mistake can't be taken back. Most businesses have plenty of the first kind and should keep people firmly on the second.

Good at

Work that repeats daily or weekly
Reading messy inputs: emails, forms, PDFs, call transcripts
Moving information between tools that don't connect
Following written rules every time, including at 2 a.m.
Drafting replies, summaries, and reports for a quick review
Noticing what is late, missing, or out of pattern

Not ready for

High-stakes decisions with no written rules
Work with no clear definition of done
Relationships: negotiation, apologies, hard news
Anything irreversible without a person's sign-off
Processes that live only in someone's head (write them down first)
Data scattered across spreadsheets nobody maintains

Where small businesses actually are

AI use is common. Agents that own work are not. That gap is the opportunity, and it is why most businesses should start with one well-chosen job rather than an "AI transformation."

46%

of small employer firms in the Federal Reserve's fall 2025 survey said the business or its employees use AI.

Small Business Credit Survey, 2026 report

7%

of those AI users had fully integrated it into their business. About half were still experimenting.

Small Business Credit Survey, 2026 report

6%

of small-business workers who use AI say their main use is automating workflows with minimal human involvement.

U.S. Chamber of Commerce Foundation, June 2026

40%+

of agentic AI projects will be canceled by the end of 2027, Gartner predicts, over cost, unclear value, or weak risk controls.

Gartner, June 2025

For the full picture, including adoption by firm size and the economics of recovered time, read our 2026 AI Business Transformation Report.

§03

How much do AI agents cost?

Agent pricing comes in six shapes, and the sticker price is rarely the largest cost. Below are current US list prices, checked on each vendor's own pricing page on October 8, 2026. Prices change often; treat them as a snapshot and check before you buy.

AI assistants in your office suite

Per person, per month

$14 to $30per user, per month

Google Workspace Business Standard, Gemini included$14Per user per month on an annual plan; covers the whole suite
ChatGPT Business, ChatGPT agent included$20Per user per month billed annually; $25 billed monthly; 2-seat minimum
Microsoft 365 Copilot Business$21Per user per month list, paid yearly; $18 promotional price through December 31, 2026
Microsoft 365 Copilot$30Per user per month, annual commitment

Best when: Your team wants faster drafting, research, and summaries.
Watch for: A person still starts every task and moves the result. It speeds up work inside a job; it rarely takes a job off anyone's list.

Agents inside a help desk or CRM

Per resolved conversation or action

$0.10 to $2per action, conversation, or resolution

HubSpot Breeze Customer Agent$0.50Per resolved conversation; Pro and Enterprise plans; cut from $1.00 in April 2026
Intercom Fin$0.99Per outcome; standalone use has a 50-outcome monthly minimum
Zendesk AI agents$1.50 to $2Per automated resolution, committed or pay-as-you-go
Salesforce Agentforce$2 or $0.10Per conversation, or per action with Flex Credits

Best when: The whole job lives inside a platform you already pay for, usually customer support.
Watch for: Platform plan fees come on top, volume drives the bill, and each agent works inside one tool.

Do-it-yourself agent builders

Monthly plan, plus your time

$12 to $40entry plans, per month

Make Core$12Per month, 10,000 credits
n8n Cloud Starter$20Per month billed annually, 2,500 executions
Lindy Plus$29.99Per user per month, 3,000 credits
Zapier Agents Pro$33.33Per month billed annually ($400 a year)
Gumloop Profrom $37Per month, 20,000 credits

Best when: Someone on your team enjoys building, and the process is simple and stable.
Watch for: Usage caps, separate AI model charges, and the hours to design, test, and fix it. When it breaks, it's yours.

Voice agents for phone calls

Per minute of talk time

$0.07 to $0.31per minute, before phone line fees

Vapi$0.05 + modelsPer minute for hosting; AI models and voice billed at cost, about $0.08 to $0.13 a minute in total
Retell AI$0.07 to $0.31Per minute, depending on the model and voice
Bland AI$0.14Per minute on the Start plan, no platform fee

Best when: You miss calls after hours, or someone spends the day answering the same phone questions.
Watch for: A month with 1,000 minutes of calls lands around $70 to $310 in platform and model fees, before phone lines, numbers, and build time.

Custom build by an agency or freelancer

Project fee, then upkeep

$10k to $50ktypical AI project; agents can run higher

AI development projects on Clutch$10,000 to $49,999Typical range across AI projects of all kinds; the average is about $120,600
Production agent connected to business systems$100,000+One agency's published estimate; treat as indicative

Best when: You need something unusual, at real scale, and you want to own the code.
Watch for: Who maintains it after launch and what each change costs. Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027.

Managed agents, built and run for you

Per job, per month

$199 to $249per job, per month at Machine-like

Machine-like, 1 to 2 workloads$249Per workload per month
Machine-like, 3 to 5 workloads$199Per workload per month
Machine-like, 6 or moreCustomQuoted on a scoping call

Best when: The work spans several tools and nobody on staff wants to own an agent.
Watch for: Ask any provider who owns your data and outputs, how you leave, and what counts as a change versus a new job.

Same job, three ways to pay for it

Take one common job: answering and following up on every inbound lead, about six hours of work a week. Here is what it costs to staff it, to build it yourself, and to have it run for you. Assumptions are printed on each receipt; swap in your own.

Staff it

Hours a month26
Median admin wage$23.23/h
With benefits$33.18/h

Per month $863

BLS median wage for administrative assistants, May 2025, loaded so wages are 70% of total cost (BLS ECEC, June 2026).

Build it yourself

Builder plan$30
Upkeep, 1 h/week$217
AI model usagevaries
Initial build timenot incl.

Per month $247+

Assumes your time is worth $50 an hour and one hour a week keeps it running. When it breaks, the fix is yours.

Have it run for you

Machine-like workload$249
Build and setup$0
Hosting, fixes, changesincl.
Your timeapprovals

Per month $249

Published Machine-like pricing for one or two workloads; $199 each at three to five.

Do-it-yourself and managed land close on paper. The difference is who builds it, who fixes it, and whose evenings it eats. Both cost a fraction of doing the work by hand, which is the comparison that matters.

What the work costs you today

Before comparing agent prices, price the work you would be handing off. Use your own wages if you know them; these are national medians from the Bureau of Labor Statistics, loaded for benefits.

Monthly cost of a job = hours per week × 52 ÷ 12 × (hourly wage ÷ 0.70)

Role (BLS median, May 2025)	Hourly wage	With benefits	5 h/week	10 h/week
Customer service representative	$21.53	$30.76	$666/mo	$1,333/mo
Secretary or administrative assistant	$23.23	$33.18	$719/mo	$1,438/mo
Bookkeeping or accounting clerk	$24.36	$34.80	$754/mo	$1,508/mo

Hourly figures for assistants and clerks are annual medians ($48,310 and $50,670) divided by 2,080 hours. Benefits load uses the BLS finding that wages are 70.0% of private-industry compensation costs. Sources are listed at the end of this guide.

§04

AI agent examples by job

Most owners don't struggle to understand agents. They struggle to see which of their own work could go to one. So here are 47 jobs, sorted by the role that usually carries them, each with what starts it, what it hands back, and where it waits for you.

Tick the ones that describe work in your business. Your shortlist and the hours it represents add up at the bottom of the screen. Hours are typical weekly loads for a 5 to 50 person business; yours will differ.

§05

AI agent examples by industry

The jobs above exist in nearly every business. Some industries have their own versions that are worth naming, because that is how owners in those industries describe the pain.

Real estate

Answer portal and website inquiries within minutes and book showings.
Track inspection, appraisal, financing, and closing deadlines, and remind every party.
Turn listing data into descriptions and social posts for your review.

Small law firms

Screen intake forms and assemble conflict-check information for an attorney.
Chase clients for documents and signatures, and log what arrives.
Draft matter status updates to clients for attorney review.

Agents prepare. Licensed professionals decide and sign.

Accounting and bookkeeping firms

Collect client documents in tax season and track what's missing.
Categorize transactions and flag the ones that need a human eye.
Assemble monthly client reporting packages for review.

Marketing agencies

Pull every client's report from ad and analytics platforms each week.
Track AI search visibility for each client.
Produce first drafts of content to each client's brief.

Home services and trades

Answer after-hours calls and book the job.
Follow up on open estimates until the customer decides.
Ask for a review after every completed job.

E-commerce

Answer order-status questions with real tracking data.
Handle return requests and issue labels within your policy.
Keep product listings consistent across every channel.

§06

How to choose your first AI agent

Your first agent should be boring, frequent, and easy to check. Pick one job from your shortlist and answer five questions about it. A high score means it is a strong first agent; a low score means keep it human for now or tighten the process first.

Frequency against the cost of a mistake

Another way to sort the list: how often the work happens, and what a mistake would cost. Frequent, low-risk work is where agents pay back fastest. Frequent, high-risk work is still worth delegating, with a person signing off.

Keep it human

Rare and high-stakes. Not worth the setup, and not worth the risk.

Hiring decisions
Pricing changes
Legal positions

Hand off with sign-off

Frequent and consequential. The agent does the work; a person approves.

Paying bills
Proposals
Refunds
Client reports

Bundle it

Rare and low-stakes. Let it ride along inside a bigger job.

Renewal alerts
License reminders
Year-end document requests

Hand it off

Frequent and fixable. The best first agents live here.

Lead response
Ticket triage
Data entry
CRM cleanup
Invoice reminders

§07

Build it, buy it, or have it run for you

There are three honest ways to get an agent working in a small business. None is right for everyone. The deciding question is not price; it is who you want answering for the agent in month seven, when the novelty has worn off and something upstream changes.

Question	Build it yourselfNo-code agent builders	Switch on a platform agentInside your help desk or CRM	Have it run for youManaged Agent Operations
Who builds it	You or someone on staff	The vendor; you configure it	The provider, from a scoping call
Who fixes it when it breaks	You	The vendor fixes the platform; you fix your setup	The provider
Typical cost	$12 to $40 a month entry plans, plus your hours and model usage	About $0.50 to $2 per resolution, plus the platform plan	Flat monthly per job; $199 to $249 at Machine-like
Reach	Any tool with an integration, if you wire it	One platform, usually one job	Across your existing tools
Best when	You have a technical operator and a simple, stable process	The whole job lives inside a tool you already pay for	The work spans several tools and nobody wants to own an agent
Watch for	Knowledge that lives in one person's head	Costs that scale with volume	Data ownership, exit terms, and what counts as a change

§08

Your first 30 days with an agent

Whether you build, buy, or hire it out, the first month should follow the same arc: narrow scope, real work, tight supervision, then earned independence.

Day 1
Pick one job
Write one sentence that says what done looks like. Gather ten examples of the job done well.
Days 2 to 7
Connect and set limits
Grant only the access the job needs. Require sign-off on money, customers, and anything without an undo.
Days 8 to 21
Run it on real work
Approve everything. Correct it in writing when it's wrong. Track hours saved, errors, and time to finish.
Days 22 to 30
Loosen the reins
Drop sign-off on routine cases it handled perfectly. Keep it on exceptions. Review the log weekly.
Day 30
Keep, change, or stop
Compare against your baseline. If it earned its place, pick job two.

§09

Guardrails that make agents safe to trust

An agent with access to your systems deserves the same controls you'd give a new hire with a company card. None of these need a committee. All of them need a name next to them.

Narrow access
Only the permissions the job needs, revocable at any time.
A named owner
One person accountable for how the agent performs.
Sign-off gates
Money, customer messages, and anything irreversible wait for a yes.
An exception queue
When it isn't sure, it stops and asks. It doesn't guess.
Tests from real history
Before launch, run it on past cases where you know the right answer.
A full log
Every action recorded: what it saw, what it did, and why.
An off switch
Anyone responsible can pause it instantly.
Your data stays yours
No training on your data, and you own every output.

Watch forAgents that read outside content, such as inbound email, web pages, or attachments, can be fed hidden instructions. Keep those agents' permissions narrow and require sign-off before they send money or data anywhere.

§10

Frequently asked questions

What is an AI agent for business?

An AI agent for business is software that owns a defined job from start to finish. It notices when work arrives, gathers context from your tools, decides the next step within rules you set, takes the action in your systems, and hands back finished work. Anything outside the rules goes to a person. A chatbot answers questions; an agent completes work.

How much do AI agents cost for a small business?

It depends on how you buy. As of October 2026, AI assistants in office suites cost $14 to $30 per user per month. Agents inside help desks and CRMs charge about $0.50 to $2 per resolved conversation. Do-it-yourself builders start around $12 to $40 a month, plus your time. Custom AI projects from agencies typically run $10,000 to $50,000, and production agents connected to your systems can pass $100,000. Managed agents that are built and run for you cost $249 a month per job at Machine-like.

What is the difference between an AI agent and a chatbot?

A chatbot waits for a question and replies with text. An agent is started by an event, such as a new lead, an overdue invoice, or a support ticket. It works across your tools, takes actions like sending, updating, and booking, and finishes the job. Many products sold as agents are chatbots with a new label, so ask what the product can actually do inside your systems.

How are AI agents different from automation tools like Zapier?

Rule-based automation follows fixed if-this-then-that steps and breaks when the inputs vary. An agent can read messy inputs such as free-text emails and PDFs, work out which path applies, and handle variation within your rules. Many agents use automation tools underneath. Use plain automation when the steps never change; use an agent when the work needs judgment within limits.

What tasks can AI agents handle for a small business?

Recurring work with a clear finish line: answering and qualifying leads, follow-ups, ticket triage and routine support answers, invoice reminders, bill intake, data entry from forms and documents, scheduling, reporting, CRM cleanup, and content drafts. This guide lists 47 jobs across seven roles, with the trigger, the output, and the sign-off point for each.

Will AI agents replace my employees?

Mostly, they replace tasks. In an OECD survey of small and medium-sized businesses in seven countries, 83% of those using generative AI reported no change in their staffing needs. The common pattern is that repetitive volume leaves people's plates and their time moves to customers, judgment, and growth. Decide in advance where the recovered hours should go, or they quietly disappear.

Do I need technical skills to use AI agents?

To build your own with a no-code tool, comfort with logic, integrations, and testing helps, and you will own the fixes. To switch on an agent inside a platform you already use, you need admin access and clean data. With a managed service, you describe the job, connect your tools, and approve the results; the provider handles the technical work.

Are AI agents safe to use with business data?

They can be, with the right controls: narrow permissions, sign-off on money and customer-facing actions, an exception queue, a full log of every action, and an off switch. Confirm that your data is not used to train models and that you own the outputs. Take extra care with agents that read outside content such as emails or web pages, because hidden instructions can try to steer them.

How long does it take to set up an AI agent?

An agent inside a platform you already use can be live in days if your data is clean. A do-it-yourself build takes days to weeks of your time, plus ongoing upkeep. Custom agency projects often take months. Whatever the route, plan on two to four weeks of running with sign-off before you let an agent work on its own.

What should my first AI agent do?

Pick a job that happens every week, follows rules you could write on one page, lives in software you can connect, is fixable within a day if something goes wrong, and takes at least two hours a week. Lead response, invoice follow-up, ticket triage, and weekly reporting are common first agents because they pass all five tests.

The short answer

What they do

What they cost

Where to start

What is an AI agent for business?

Anatomy of one agent run

Trigger

Context

Decide

Act

Check

Report

Agent, chatbot, assistant, or automation?

What AI agents can do for a business, and what they can't yet

Good at

Not ready for

Where small businesses actually are

How much do AI agents cost?

AI assistants in your office suite

Agents inside a help desk or CRM

Do-it-yourself agent builders

Voice agents for phone calls

Custom build by an agency or freelancer

Managed agents, built and run for you

Same job, three ways to pay for it

Staff it

Build it yourself

Have it run for you

What the work costs you today

AI agent examples by job

The owner's desk

Daily business brief

Inbox triage

Meeting prep and follow-up

Weekly numbers report

Renewals and subscriptions watch

Competitor watch

Sales

Inbound lead response

Lead enrichment and routing

Prospect list building

Quote and stalled-deal follow-up

CRM cleanup

Proposal and quote drafts

Call notes to CRM

Marketing

Content drafts in your voice

Social posting

Review monitoring and replies

Marketing reporting

AI search visibility

Repurposing

Website upkeep

Customer service

Ticket triage

Routine answers

Order status questions

Booking and reminders

After-hours phone answering

Feedback digest

At-risk customer alerts

Finance and bookkeeping

Invoice follow-up

Bill intake and coding

Receipt chasing

Invoicing from finished work

Month-end close prep

Cash snapshot and forecast

Year-end document requests

Operations and admin

Intake and data entry

Document processing

Scheduling and dispatch

Reorder alerts

Vendor quotes

Procedure upkeep

Deadline and compliance tracking

HR and recruiting

Job posts

Applicant summaries