An AI agent is software that receives a goal, chooses the steps to reach it and acts inside your tools: it reads an email, checks the CRM, drafts a reply, updates a record. A chatbot answers; an agent executes. For a small business, it pays off on repetitive, well-bounded tasks whose results are easy to check.
What is an AI agent, exactly?
You have probably used ChatGPT or Copilot to draft an email. You ask, the tool answers, and you copy, paste and send. An agent does the rest: it opens the inbox, finds the right contact, drafts the message and, depending on the permissions you give it, sends it.
The clearest definition we know comes from Anthropic, the company behind Claude. It separates two families of systems (“Building effective agents”, December 19, 2024):
- Workflows: a language model and tools chained along a path written in advance. Step 1, step 2, step 3.
- Agents: the model decides its own path and which tools to use to reach the goal.
The distinction matters for your budget. Many products sold as “agents” are well-built workflows, and that is often good news: a predictable workflow is easier to test and fix than an agent free to make its own choices.
What changed in September 2026
If the word “agent” seems to be everywhere lately, it is not your imagination. Two announcements target small businesses directly.
- HubSpot, September 16, 2026. At its Fall 2026 Spotlight, HubSpot introduced a new version of Breeze Assistant that, in its words, no longer just answers questions: it takes action and hands tasks to specialized agents (prospecting, content, campaigns, customer service). It all lives in the CRM, where your contacts and deals already are.
- Meta, September 29, 2026. Meta opened its Muse agent to small businesses, with connections to Shopify, QuickBooks, Stripe, Slack, Canva and Notion, on top of Facebook Pages, Instagram accounts and Meta ad accounts (TechCrunch, September 29, 2026). A free tier exists, with usage limits.
What they have in common: the agent no longer lives in a separate chat window. It is plugged into the tools your business already runs on. That connection is what makes it useful, and it is also what makes it risky. Learn more: our HubSpot services.
Where Canadian businesses stand
Statistics Canada measures AI use every quarter. In the second quarter of 2026, 19.2% of Canadian businesses reported using AI, compared with 12.2% a year earlier and 6.1% in 2024 (Canadian Survey on Business Conditions, collected April 1 to May 6, 2026). Use has tripled in two years.
Three other figures from the same survey deserve your attention:
- The smallest businesses are not behind. 19.9% of businesses with 1 to 4 employees used AI, compared with 27.8% of those with 100 employees or more.
- Virtual agents and chatbots rank third among reported uses (28.2%), behind data analytics and text analytics.
- Two businesses in five (40.0%) say AI is not relevant to them.
That last figure is healthy. Not every business needs an agent, and the right question is not “do we need an agent?” but “which task, exactly, do you want to hand over?”.
What an agent really does for a small business
An agent pays off when three conditions are met: the task comes back often, its rules fit in a few lines, and its result is easy to check. These uses meet all three:
- Qualifying inbound requests: reading the form, enriching the CRM record, suggesting a priority and the right rep. A human approves.
- Answering frequent customer questions from your own documents (return policy, lead times, warranties), with a handoff to a person as soon as the question goes off script.
- Preparing sales follow-up: summarizing a call, updating the deal, drafting the follow-up email.
- Keeping the CRM clean: spotting duplicates, filling in missing fields, flagging deals with no activity.
- Preparing reports: pulling the numbers from your ad platforms and your analytics into a single table, every week. Learn more: our analytics approach.
What an agent does poorly today is decide alone on something ambiguous, costly or irreversible: a refund, a price quote, a legal answer. Gartner puts it bluntly: most agentic AI propositions lack significant value or return on investment, because current models lack the maturity to achieve complex business goals on their own (Anushree Verma, press release of June 25, 2025).
Three risks to check before you plug in an agent
- The “agent” that isn’t one. Gartner calls it “agent washing”: assistants, chatbots or classic automation rebranded as “agents”. By its estimate, only about 130 of the thousands of agentic AI vendors are real. The same release predicts that over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls (Gartner, June 25, 2025). Ask for a demo on your own data, not a video.
- You answer for what your agent says. In February 2024, the British Columbia Civil Resolution Tribunal ruled against Air Canada over wrong information its website chatbot gave about a bereavement fare. The airline argued it was not responsible for what its chatbot said. The tribunal answered that it remained responsible for all the information on its website, and ordered it to pay $812.02 (Moffatt v. Air Canada, 2024 BCCRT 149). The amount is small; the principle applies to everyone.
- Quebec’s Law 25 also covers automated decisions. Section 12.1 of Quebec’s Act respecting the protection of personal information in the private sector requires a business that makes a decision based exclusively on automated processing of personal information to inform the person concerned, no later than when it informs them of the decision. On request, it must also tell them which information was used and the main factors behind the decision, and let them have it reviewed by a person. An agent that alone turns down a credit application or a job candidate is directly concerned. For advice on your situation, consult a lawyer.
Where to start
Before buying anything, answer these five questions. Sorry, this is not the most spectacular part, but it is the one that decides the outcome.
- Which task? One only, described in a sentence, with its weekly volume.
- With which data? An agent plugged into a messy CRM makes bad decisions faster. Cleaning the data comes before the agent.
- With which permissions? Read-only first, drafts next, automatic sending last, and only for what is easy to correct.
- Who approves? Name the person who reviews, and for how many weeks.
- How do you measure? Time saved, error rate and customer satisfaction, measured before and after. Without a baseline, no result can be proven.
And if you already use HubSpot, check what your subscription includes before adding yet another tool: Breeze agents work on the data you already have.
Our take
An AI agent is a very fast colleague with no judgment, who only knows about your business what you give it. You do not hand it the keys on day one. You give it one task, you measure, and you widen the scope when the numbers justify it. That is how we see things, and we will not change.
One last point, often overlooked: agents do not only work for you. Your customers’ agents also look for suppliers, compare offers and cite sources. Being understood and cited by these engines is what GEO (also called AEO) is about. Learn more: our approach to AI search optimization.
One task. Clean data. A measurement.
Frequently asked questions
What is an AI agent, in one sentence?
An AI agent is software that receives a goal, decides the steps to reach it and acts on its own inside your tools, such as your email or your CRM. It differs from a chatbot, which answers your questions but leaves the work to you. An agent executes, within the limits of the permissions you give it.
What is the difference between an AI agent and a chatbot?
A chatbot answers a question in a chat window. An AI agent chains several actions to reach a goal: looking up information, updating a record, drafting and sending a message. Many tools sold as “agents” are in fact rebranded chatbots or automations, which Gartner calls “agent washing”.
Is an AI agent useful for a small business?
Yes, for repetitive, well-bounded tasks that are easy to check: qualifying requests, answering frequent questions, preparing follow-ups or reports. In the second quarter of 2026, 19.9% of Canadian businesses with 1 to 4 employees already used AI, according to Statistics Canada. Start with a single task, measured before and after.
Who is liable if an AI agent gives a customer wrong information?
The business. In 2024, a British Columbia tribunal held Air Canada liable for wrong information given by its website chatbot, noting that the airline is responsible for all the information on its site. Keep a human approval on anything that commits your business: prices, refunds, terms.
Does Quebec's Law 25 apply to AI agents?
It applies as soon as an agent processes personal information, and section 12.1 specifically covers decisions based exclusively on automated processing: you must inform the person and let them have the decision reviewed by a human. An agent that prepares a decision approved by a person raises fewer issues. For your specific case, consult a lawyer.
Ready to move from reading to doing? See how we handle AI agent development for small businesses: one well-scoped task, measured tests before launch, then a monthly follow-up.
Further reading
- HubSpot partner agency
- HubSpot AEO: how to read mentions, citations, sentiment and share of voice
- Analytics
- AI search engine optimization
- AEO or GEO: two names for the same discipline
Further reading: how agents are paid for in HubSpot credits, and what drives the rest of the bill, in HubSpot pricing: what drives the cost of your subscription.
Further reading: before handing lead qualification to an agent, write down what a qualified prospect means for you, in MQL, SQL, SAL: definitions, differences and how to align them.