What Is an Agent? And Why Should You Care Right Now?

You’ve probably heard people talking about “AI agents” recently. If it sounds like just another bit of tech jargon, I understand. But this one is worth paying attention to because it’s going to change how you run your business - and sooner than you might think.

Let me explain what agents actually are, why they matter right now, and what you can do about it today.

First, What Is an Agent?

You know how ChatGPT can answer questions, write emails, and give you advice? That’s a chatbot. It talks, but it doesn’t do anything.

An agent is what you get when you give a chatbot the ability to actually take action. It can send emails, create spreadsheets, update your CRM, make phone calls, book appointments - whatever tools you connect it to.

The simplest way to think about it: a chatbot gives advice, an agent does things.

This isn’t theoretical. It’s already happening:

• Coding agents like Claude Code and Codex write software and can even manage teams of other agents working together on a project
• Office agents built into Microsoft 365 and Claude can create spreadsheets and documents for you
• Voice agents from services like Bland AI and Vapi answer phone calls for businesses, handle routine enquiries, and pass the tricky ones to a human
• Bookkeeping agents inside QuickBooks categorise transactions, reconcile accounts, and chase unpaid invoices - owners report saving around 12 hours a month
• AI receptionists from services like AnsweringAgent answer calls and book appointments 24/7 for about £50 a month, compared to several thousand for a human

OK, But ChatGPT Has Been Doing Stuff for Ages. What’s Changed?

Fair point. The tools have been improving steadily for a while. But something happened in the last few weeks that signals where this is all heading.

Open Claw: A Sign of Things to Come

An Austrian developer called Peter Steinberger built a tool called Open Claw for his own personal use. It went viral. He published it on GitHub - the world’s largest platform for sharing software - and it became the fastest project in the site’s history to reach 100,000 endorsements, doing it in about two days. As of this week it’s at around 200,000 with 1.5 million agents created.

The idea behind it is simple but powerful:

1. Install it on your own computer
2. Connect it to your digital life - email, Slack, WhatsApp, HubSpot, Notion, whatever you use
3. Send it a message through any of those channels and it does what you ask

Here’s a real example of what it can do. You meet Jane from ACME Consulting at a networking event. You send Open Claw a WhatsApp message: “Met Jane from ACME Consulting today. Add her to the CRM. Arrange a follow-up.”

The agent then researches Jane and ACME Consulting, checks your calendar for context about the meeting, creates a CRM entry in HubSpot, searches your website and Notion for information Jane might find useful, and drafts a personalised follow-up email with relevant links and an invitation to a call.

One message. All of that done for you.

And here’s my favourite story: someone’s Open Claw agent tried to book a restaurant online, couldn’t find a way to do it through the website, so it found some voice software and phoned the restaurant directly to make the booking. Nobody told it to do that. It figured it out.

There’s a Catch Though

Open Claw has serious security problems right now. If it can take instructions from your email, so can anyone else. Security researchers found 512 vulnerabilities in a recent audit, and about 10% of plugins in the community marketplace were flagged as malicious. It’s not ready for general use yet.

But here’s what happened next: OpenAI hired Peter Steinberger and is funding the project through an independent OpenClaw Foundation. That means serious money and engineering talent behind making this kind of tool reliable and safe. And if OpenAI is doing it, you can be sure Anthropic, Microsoft, Google, and dozens of startups are working on their own versions.

What Does This Mean for Small Businesses?

The effects are already showing up, even if they’re not always obvious:

• Software is getting built faster. Developers using coding agents can work at 10 times the speed they could before. New products that used to take months are launching in weeks.
• The internet is adapting. Websites and search engines are adding features that make it easier for agents to find and use information - stripped-back pages that agents can read quickly and cheaply.
• Subscription software companies are worried. Stock markets have noticed that agents can do what many paid-for software products do, but at a fraction of the cost. Around $2 trillion in market value was wiped out from companies like Salesforce, HubSpot, and ServiceNow in January and February 2026, and their share prices continue to slide as the market adjusts.
• The big companies are betting everything on this. OpenAI, Anthropic, and Google are raising billions to build the infrastructure, because they’re expecting agent use to grow massively.

The direction is clear: instead of buying a product to do something - a CRM, a project management tool, a booking system - you’ll increasingly use agents that you’ve configured, downloaded, or had built to manage things exactly the way you want them.

When Should You Think About Using Agents?

Agents make sense when you have:

• Repetitive tasks that follow a pattern. If you find yourself doing the same sequence of steps regularly - data entry, follow-up emails, invoice chasing - an agent can likely handle it.
• Tasks that span multiple tools. If a job involves copying information between systems (email to CRM, calendar to spreadsheet), that’s exactly what agents are good at.
• High-volume routine communication. Answering the same customer questions, booking appointments, sending reminders - agents handle this well and don’t need sleep.

When Should You NOT Use an Agent?

This is just as important, and I’d rather be honest about it than pretend agents are the answer to everything.

Don’t use an agent when:

• The task involves judgement calls with real consequences. Agents can act on bad data without realising it. If a mistake means losing a client or breaking a contract, keep a human in the loop.
• Actions are irreversible. Sending money, deleting records, publishing content - anything you can’t undo easily should have human approval before it happens.
• You can’t supervise it. Agents can get stuck in loops or go off track. If you set one up and forget about it, you might come back to a mess. They work best when you check in regularly.
• A simple chatbot would do. If all you need is answers to questions or help drafting text, a regular AI chat is simpler, cheaper, and less risky than giving it tools to take action.

The rule of thumb: the more an agent can do, the more you need to watch what it’s doing - at least until the tools mature.

Things You Can Try Right Now

You don’t need to wait for the polished products. Here are some practical starting points:

1. Try Cowork. Anthropic recently launched Cowork, a desktop agent built into Claude that’s designed for non-technical users. It can interact with your computer on your behalf - filling in forms, organising files, that sort of thing. It’s in early access but worth experimenting with.

2. Ask your existing AI what it can do. If you’re already using ChatGPT, Claude, or Copilot, ask it what agentic features it has. You might be surprised. Claude can create spreadsheets and documents. Copilot can build presentations. These features are being added constantly.

3. Look into MCP servers. Model Context Protocol (MCP) is a standard that lets you connect an AI to specific apps - your email, your calendar, your CRM. It’s a bit technical to set up right now, but it’s the plumbing that makes agents work, and it’s getting easier all the time.

4. Explore Skills. Skills are downloadable bundles of instructions and code that teach an agent how to do a specific task - create a spreadsheet, run a competitor analysis, generate a report. Think of them as apps for your AI.

Where This Is Going

Agents are going to become as normal as apps on your phone. Right now we’re at the stage where early adopters are figuring things out and the big companies are racing to build the infrastructure. The security and reliability problems are real, but they’re being worked on with serious money behind them.

My advice: start small, stay curious, and don’t ignore it. The businesses that understand agents early - even at a basic level - will have a real advantage when the tools are ready for everyone.

If you want to talk about what any of this means for your specific situation, book a free discovery call. No jargon, no sales pitch - just a practical conversation about where you are and what makes sense.

Further Reading

• Building Effective Agents - Anthropic’s detailed explainer on how agents work
• The 2026 Guide to AI Agents - IBM’s plain-language overview
• AI Agents for Beginners - Microsoft’s free 12-lesson course (no coding needed)