"What are some examples of AI agents?" is one of the fastest-growing questions we see in search data this autumn, together with "what's the best AI agent for a small business?" Most of the answers online are lists of products. We think the question underneath is a different one: what would an agent actually do in a business like mine? This post answers that with examples sorted by job. Then it does something most lists don't. It goes through the AI products we've built at Autor (they're all on our work page) and says plainly which ones are agents and which ones aren't.
First, a working definition
The most useful line we know comes from Anthropic's engineering guide, Building effective agents. It separates two things that both get called "agents." Workflows are systems where a language model and tools are orchestrated through predefined code paths. Agents are systems where the model directs its own process and tool use, deciding how to get the task done.
We use a two-part test that follows from that:
- Does it decide its own next step? A fixed script that calls a model at step three is a workflow. That's often the better build, and nothing to be ashamed of.
- Does it act through tools? If it only produces text for a person to copy somewhere, it's an assistant. An agent books the slot, updates the record or sends the message.
Most real business systems are a mix. That's fine. The label matters less than knowing which parts make decisions, because those parts need guardrails.
Is ChatGPT an AI agent? Is Claude?
The plain chat window is an assistant. You ask, it answers, and you do the work. Both companies now ship agent modes on top of it. OpenAI's ChatGPT agent can carry out multi-step tasks using tools. Anthropic publishes the same building blocks for developers in the Claude Agent SDK. So the honest answer is "the chat isn't, but the product can be." For a business, the more useful question is whether a general-purpose agent can reach your systems with the permissions and approval steps you need. That's usually where a general tool stops and a specific one starts.
AI agent examples by the job they do
These are the patterns we see most in small and mid-size businesses. Each one is narrow on purpose. In our experience, narrow agents are the ones that reach production. Our guide to building an AI agent explains why.
| Job | What the agent decides and does | Systems it acts in | Where a person stays in the loop |
|---|---|---|---|
| Answering customers | Works out what the caller or chatter wants, answers from your own information, takes details | Phone, website chat, messaging, calendar | Anything outside what it's been told it can answer |
| Lead follow-up | Decides who is waiting, what to say next and on which channel | CRM or lead list, email, text | Approves outbound messages before they send |
| Booking and rescheduling | Finds a slot that fits the rules, books it, confirms | Calendar, booking system | Exceptions and double-bookings |
| Re-engaging past customers | Picks who is due, drafts the reminder, books whoever replies | Customer list, messaging, calendar | Approves the list and the wording |
| Document intake | Reads an incoming form, invoice or application, pulls out the fields, files it, flags gaps | Email inbox, storage, accounting or CRM | Reviews anything it flags as uncertain |
| Internal questions | Finds the answer in your policies, past tickets or docs and cites where it came from | Knowledge base, drive, ticketing | Owners keep the source documents current |
| Scheduling staff | Builds a roster that satisfies rules and availability, then proposes changes when someone drops out | Scheduling system, notifications | Manager signs off the roster |
Notice what isn't on the list: "an AI that runs the business." Every example is one job with one measurable number attached, such as response time, the share of leads followed up within the hour, or overdue customers re-booked.
Which of our builds are actually agents
We've put every product below on our work page. Here's how each one scores on the two-part test, including the ones that fail it.
| Build | What it is | Agent? |
|---|---|---|
| Loquent (our own product) | An AI workforce for small businesses: owners hire AI employees by role, and the same employee works across phone, website chat, WhatsApp and text | Yes. Each employee decides what a conversation needs and acts through tools. It asks the owner before it acts |
| Avatraxx | A fitness platform with LLM-driven training plans that adapt to each user's progress and engagement | Partly. The plan adapts in a loop, but within a defined programme. Closer to a smart workflow |
| Bastet Rising (by Aranexx) | A women's health platform that turns wearable data into insights, with an NLP engine for health questions | Partly. It answers questions and personalises recommendations. It doesn't act in outside systems |
| Rosterbees | Automated shift scheduling for care homes that accounts for compliance rules, availability and skills | Mostly no. It's an optimisation engine. It decides a lot, but through rules and models, not an LLM directing tools. That was the right call for rostering |
| InteliGaze AI | Computer vision that analyses in-store visitors and updates digital signage | No. It's perception and analytics. Very useful, but not an agent |
| AI Behavior Monitoring | Thermal, radar and audio sensing that detects anomalies and triggers alerts | No. It detects events and alerts people; it doesn't plan its own steps |
| AI-Driven AR Platform | Augmented-reality repair guidance for automotive technicians, using computer vision to identify components | No. It guides a person step by step, and the person does the work |
Two of seven are clearly agents. We think that's a healthy ratio. Plenty of valuable AI is perception, prediction or optimisation, and calling all of it an "agent" makes it harder for buyers to compare what they're being sold. Our longer primer, what AI agents are and how they differ from chatbots, covers the categories.
Lessons from building Loquent as an agent
Loquent is the build where the agent question mattered most, because the AI employees talk to the business's customers directly. Three things shaped it, and they apply to any customer-facing agent:
- Approval before action. An employee shows the owner a card and waits before it acts. That one design choice does more for trust than any model upgrade.
- One memory across channels. A customer who called on Monday and texts on Thursday shouldn't start over. That needs a shared memory of every conversation, not a separate bot per channel.
- Hire, don't configure. Small-business owners don't want to write prompts. They pick a role and answer a few questions. The engineering cost of making that simple is real, and it's where most of the work went.
We wrote up the earlier version of the build in how we built Loquent. If you're wondering whether "AI employee" is just a new name for an agent, what is an AI employee? explains the difference.
So what's the best AI agent for a small business?
The best one does a single job you can measure and connects to the systems you already use. For common jobs like answering and following up, a ready-made product usually wins. That includes our own Loquent (getloquent.com), built by Autor Technologies in Toronto. For a job that's specific to how your business makes money, a custom build usually wins. What's the best AI agent platform? compares the platform options, and build vs. buy covers the decision.
FAQ
What are some simple examples of AI agents? An agent that answers your website chat and books appointments on your calendar. An agent that reads incoming invoices and files them with the right fields filled in. An agent that follows up on every quote request that hasn't replied. Each one makes its own decisions about the next step and acts in a system, rather than just producing text.
Is a chatbot an AI agent? Usually not. A chatbot answers within a conversation. It becomes an agent when it can decide on and take actions outside the conversation, like booking, updating a record or sending a follow-up, through tools it's been given.
What's the difference between agentic AI and AI agents? "Agentic" describes the behaviour: a model planning steps and using tools. An "AI agent" is a system built around that behaviour for a job. Many products are partly agentic: a fixed workflow with one or two steps where the model decides.
Do small businesses need a custom AI agent? Most don't, at first. For common jobs, a product is faster and cheaper to try. Custom makes sense when the work depends on your own data and systems in a way no product covers. Our AI development cost guide explains what drives the cost of a custom build.
How do I start? Pick one job, write down the number you'd measure it by, and list the systems it would need to read and write. If you get that far and want a second opinion, how we work explains how we scope a first agent, and how to evaluate an AI development studio covers what to ask whoever you hire, including us.