AI agents could matter more to small businesses than big ones

A large company can solve a capacity problem by adding people, teams and systems. Small businesses often can't. That's why AI that can do more than answer questions could have an outsized effect on smaller companies.

Imagine two companies discover the same problem. There's a repetitive piece of work taking 20 hours every week. The first company has 5,000 employees.

The second has five. Twenty hours matters to both. But it means something very different to the company with five people.

That's almost half of somebody's working week. And that is why I think some of the most interesting effects of AI agents may appear in surprisingly small businesses. Not because small businesses have better AI.

Not because agents are somehow designed for them. And not because a five-person company is about to replace a 500-person company with a collection of digital employees. It's because capacity means something different when you don't have much of it to begin with.

Small businesses have always had a capacity problem

Running a small company means doing a strange number of different jobs. Sales. Marketing.

Customer service. Finance administration. Operations.

Research. Website. Technology.

Reporting. Chasing things. Fixing things.

Remembering things. Often the same person is doing several of them. Large businesses can create specialist roles.

Small businesses tend to create Tuesday afternoons. I'll do the invoices Tuesday afternoon. I'll sort the CRM on Friday.

I'll update the website when things quieten down. I really need to go through those enquiries. Someone needs to organise all of this.

And then nobody does, because customers need something now.

AI assistants already help with some of this

Generative AI has already given small businesses access to useful capabilities. You can ask AI to: write,

research, summarise, analyse,

brainstorm, prepare documents, help with code,

and explain things. That can make one person considerably more capable. But there's still a limitation.

You have to drive. You decide when to use the AI. Find the information.

Give it the task. Review the result. Put the result somewhere.

Trigger the next step. The AI helps with the work. You still coordinate it.

Agents change the coordination part

This is why agents are interesting. An agent isn't simply an AI that gives a longer answer. The important shift is that the system can potentially:

understand a goal, work out what needs doing, use tools,

retrieve information, take an action, look at what happened,

decide what comes next, and continue. That moves AI from:

help me do this step towards: help me move this piece of work forward.

For a small business, that difference could be substantial.

Think about an enquiry

A customer sends an enquiry. Today, perhaps the owner: reads it,

checks the CRM, looks at the website, finds the relevant service information,

writes a reply, updates the CRM, creates a reminder,

and follows up three days later. AI can already help write the reply. But an agentic workflow could potentially help with much more of the process.

It might: understand the enquiry, retrieve relevant customer information,

find the appropriate service information, prepare a response, create the CRM update,

prepare the follow-up, and bring the unusual part to the owner. That's not necessarily replacing sales.

It's removing a collection of administrative steps surrounding sales. For a small company, that can matter enormously.

Small businesses don't have a department for everything

This is where the relative effect becomes interesting. A large company may already have: sales operations,

marketing operations, IT, data teams,

analysts, customer-service systems, administrators,

developers, and internal support. A small company may have Sarah.

Or Dave. Or whoever happens to know how the spreadsheet works. There are countless useful things small businesses simply don't do consistently because nobody has the spare capacity.

Not because they're unimportant. Because something more urgent always wins.

AI may fill gaps rather than replace people

I think this is a much more useful way to think about it. Imagine a five-person business doesn't currently: analyse every customer enquiry,

prepare a weekly competitor briefing, keep the CRM perfectly updated, follow up every dormant lead,

document every internal process, review every customer conversation for recurring issues, or produce detailed management reporting.

Nobody necessarily loses a job if AI starts helping with those things. There wasn't a person doing them. The capability was missing.

That's an important distinction.

This is where the economics get interesting

Suppose a business would benefit from a particular function. But it only needs five hours of that function each week. Hiring someone full-time makes no sense.

Outsourcing may work, but perhaps the task is tightly connected to internal systems or happens continuously throughout the week. Historically, the answer may simply have been: We don't do it.

AI changes that. Not because an AI agent is equivalent to hiring a specialist. It isn't.

But because parts of that function may become economically viable for a much smaller company.

Think capability, not employee

This is why I don't particularly like the phrase: digital employee. It encourages businesses to imagine an AI agent as a person-shaped replacement for a job.

I think a better question is: What capability are we missing? Perhaps:

monitoring enquiries, organising information, researching prospects,

preparing follow-ups, checking systems, producing reports,

maintaining records, or coordinating a workflow. Now we can decide which parts:

need AI, need ordinary automation, need software,

and need a person. That's much more practical than hiring an imaginary AI employee.

Agents may be particularly useful between systems

Small businesses often accumulate software organically. A website. Email.

CRM. Accounting. Payments.

Calendar. Cloud storage. Project management.

Perhaps an industry-specific system. Individually, they work. Together, not always.

So people become the connectors. Copy this. Paste that.

Download this. Upload that. Check whether something happened.

Send a reminder. Update the other system. Agents become interesting when they can help coordinate work across those boundaries.

But don't start by building a super-agent

This is where I think businesses could go badly wrong. The demo looks impressive. So the company decides:

Let's build an AI agent that runs our operations. That's far too broad for where I'd start. Pick one bounded workflow.

One outcome. One set of information. One defined level of authority.

Learn there. A useful first agent might be incredibly boring. That's fine.

The boring agent may be the valuable one

Imagine an agent that checks every morning for: new enquiries without a response, proposals awaiting follow-up,

customer requests that haven't been completed, and records missing a required update. It doesn't run the company.

It doesn't devise corporate strategy. It doesn't negotiate contracts. It notices things.

It prepares things. It moves routine work. It tells a person when something needs attention.

For a small team, that could be extremely useful.

The first big win may be remembering

This sounds almost too simple. But small businesses lose enormous amounts of capacity to keeping track of things. Did they reply?

Did we send that? Has the supplier come back? Did anyone chase the quote?

Was the customer updated? Did that payment arrive? Did we ever finish that?

Software has helped with this for years. Agents potentially add the ability to understand the context around those outstanding items rather than simply triggering a reminder because a date passed. That's a more useful kind of coordination.

Another big win could be preparation

There's a lot of work that doesn't need to be fully automated to save time. Before the owner starts work on something, the agent could potentially prepare: the relevant customer history,

the documents, the previous correspondence, the outstanding questions,

the latest information, and a suggested next action. Now the person begins with the work organised.

For a time-poor small business owner, that's significant.

Agents could give small teams more operational depth

A small company often has very little redundancy. If one person is busy, things wait. If one person is away, a process may slow down.

If the person who understands something leaves, knowledge disappears. Agentic systems won't eliminate those problems. But properly designed systems could help make processes less dependent on somebody remembering every step.

That creates operational depth without necessarily creating another management layer.

There is another advantage: small businesses can change quickly

Large organisations have advantages small businesses don't. Resources. Specialists.

Data. Infrastructure. Capital.

But small businesses have one important advantage when technology changes: They can sometimes redesign how they work much faster. Fewer systems.

Fewer approval layers. Fewer stakeholders. Less legacy process.

Not always, of course. Some small businesses have astonishingly complicated spreadsheets. But a small company may be able to redesign one workflow this month without launching a two-year transformation programme.

That makes experimentation interesting.

Small doesn't mean simple

This is important too. A ten-person company can still handle: sensitive customer data,

large payments, regulated work, complex contracts,

valuable intellectual property, and consequential decisions. So "small business" should never mean:

risk doesn't matter. The same principles apply. Access.

Permissions. Security. Review.

Logging. Failure handling. Human oversight.

The implementation may be smaller. The responsibility isn't.

The owner shouldn't become the agent's full-time supervisor

There's another trap. Imagine an agent saves ten hours of administrative work but creates six hours of: reviewing,

approving, correcting, monitoring,

and wondering what it's doing. That's not much of a win. A useful agent should reduce coordination burden.

Not simply create a new thing the owner has to manage constantly. This is why narrow, measurable workflows are so important.

Start with work that's easy to check

A good first agentic workflow has some familiar characteristics. It happens regularly. The goal is clear.

The information is accessible. The steps are reasonably understood. The output is easy to inspect.

Mistakes are recoverable. Exceptions can go to a person. You can measure whether it helped.

That's much more attractive than: Give AI access to everything and see what happens.

Autonomy should be earned

You also don't have to start with full autonomy. A sensible progression might be: WATCH

Observe the workflow and identify what needs attention. RECOMMEND Suggest the next action.

PREPARE Create the email, update or task. ACT WITH APPROVAL

A person confirms before anything happens. ACT WITHIN LIMITS Routine, low-consequence actions happen automatically.

ESCALATE Anything unusual goes to a person. As confidence increases, the workflow can change.

The technology doesn't need maximum authority on day one.

Small businesses should be especially careful with account access

A five-person business may have surprisingly concentrated access. The owner account can do everything. The main email account sees everything.

The CRM administrator has broad permissions. One cloud-storage account contains most of the company's files. Giving an agent access through one of those accounts can accidentally give it far more authority than the task requires.

Create access around the job. Not around whatever login was easiest to connect.

The economics could change hiring decisions

This doesn't mean: agents replace hiring. Sometimes the answer is absolutely to hire someone.

People bring: judgement, relationships,

creativity, experience, accountability,

leadership, empathy, and knowledge that an AI workflow doesn't replicate.

But AI may change when a growing company needs to add headcount. Perhaps a team can handle more customers before adding another administrator. Perhaps one technical person can support more internal systems.

Perhaps salespeople spend more time selling because less time disappears into CRM administration. Perhaps an operations hire happens later because routine coordination has been reduced. That's a meaningful business effect.

It could also change the first hires

This is an interesting possibility. Small businesses traditionally hire partly around the work that consumes the founder's time. If AI removes some of the administrative load, perhaps the next hire can be chosen for a different reason.

A stronger salesperson. A specialist. A customer-facing person.

Someone who creates new value rather than primarily absorbing coordination. AI doesn't necessarily mean fewer people. It could change which people become valuable first.

This connects to the one-person technical team

I've already seen how much more one experienced technical person can build with AI. The same principle could extend beyond development. One person supported by well-designed AI systems may be able to operate across a much broader range of work.

But there's an important condition: the person still needs to know what good looks like. AI multiplies capability.

It doesn't remove the need for judgement.

This could favour experienced generalists

Small businesses have always valued people who can cross boundaries. Someone who understands: customers,

technology, operations, commercial reality,

and how the pieces connect. AI potentially gives that person far more execution capacity. They don't need to personally perform every mechanical step.

They can spend more time deciding: what should happen, what matters,

what needs fixing, and where the business should go next. That's a very different kind of leverage.

But there is a ceiling to founder leverage

This matters because "one-person billion-dollar company" type conversations can get silly very quickly. A person still has finite: attention,

judgement, energy, relationships,

and decision-making capacity. AI can increase execution enormously. It doesn't make human attention infinite.

Eventually, the bottleneck moves. Perhaps from producing work to: deciding,

reviewing, prioritising, selling,

leading, or maintaining relationships. The objective shouldn't be to see how few humans a company can survive with.

It should be to use people's time where it creates the most value.

Small businesses can also get this wrong faster

Speed cuts both ways. A large organisation may take months to approve an AI integration. Frustrating.

A small business owner may connect an agent to the company email, CRM and cloud drive on Tuesday afternoon because the demo looked brilliant. Also potentially a problem. Being able to move quickly doesn't remove the need to think about:

data, permissions, authority,

security, and failure. Move quickly on the experiment.

Be deliberate about the access.

The opportunity isn't "run your business with AI"

I don't think that's the useful promise. The opportunity is much more grounded. What would your business do better if routine coordination consumed less human attention?

Could every genuine enquiry receive attention quickly? Could every lead get followed up? Could important information be easier to find?

Could repetitive administration happen without somebody remembering? Could reports prepare themselves? Could customer context be assembled before somebody needs it?

Could systems stay updated without constant copying? Could a five-person team operate with some of the process discipline previously associated with a much larger organisation? Those are interesting questions.

Small businesses shouldn't copy enterprise AI strategy

A large organisation may need: central AI platforms, complex procurement,

large governance programmes, multiple specialist teams, and organisation-wide transformation initiatives.

A small company probably doesn't. Start with one workflow. Perhaps two.

Find something painful. Map it. Work out where AI adds something.

Decide what information it needs. Limit what it can do. Measure the result.

Then expand if it works. Small businesses have the advantage of being able to keep this practical. Use it.

The agent isn't the strategy

This is perhaps the most important point. Don't decide: We need AI agents.

Decide: We need to stop losing sales enquiries. Or:

We need to reduce the admin around every new customer. Or: We need this process to continue without somebody manually moving it through five systems.

Now you have a business problem. An agent may be an excellent component of the solution. Or ordinary automation may solve most of it.

Or a better integration. Or a small custom application. Start with the work.

Always.

Why I think small businesses are worth watching

Big companies will undoubtedly use agentic AI. They have enormous incentives to do so. But the effect in small businesses could look different.

When a company already has thousands of people and sophisticated systems, an agent may improve an existing capability. When a company has five people, an agent may help create a capability the business simply didn't have before. That's why the relative impact could be so interesting.

Not: five people suddenly become 500. Something much more believable:

five people stop spending so much of their limited capacity keeping the machinery moving. That matters.

The leverage is in what the team gets to stop doing

This is where I think the conversation about agents becomes useful. Don't begin by asking which people an agent could replace. Ask which parts of people's days are currently being consumed by work that doesn't need their particular judgement.

The copying. Checking. Searching.

Chasing. Preparing. Updating.

Moving. Remembering. Coordinating.

If AI and automation can absorb more of that, a small team can put more of its limited human attention into: customers, decisions,

relationships, ideas, quality,

sales, and growth. That's a much more interesting version of the AI agent story.

AI agents could matter enormously to small businesses

We don't yet know exactly how far agentic systems will go or how quickly businesses will trust them with increasingly important work. And different companies will adopt them very differently. But I think the economics are worth paying attention to.

Large companies have historically been able to buy capacity with: people, departments,

systems, consultants, and infrastructure.

Small businesses have had to make do with less. AI doesn't erase that difference. But it may reduce parts of it.

A small company can increasingly access: intelligence, software-building capacity,

research capability, automation, and workflow coordination

that would previously have required substantially more resource. That's why I think agents could be particularly consequential for small businesses. Not because they replace the business.

Because they may allow a small number of people to spend much more of their time on the parts of the business that actually need people.

Where to go next

• AI agents aren't digital employees. Here's what they actually are. Start with what an agent actually does before deciding whether your business needs one.
• What should businesses actually automate with AI in 2026? Find the repeated work where AI and automation can create genuine capacity.
• AI is changing how much one technical person can build Why AI is increasing the execution capacity of experienced individuals and small teams.
• The important question isn't what AI can do. It's what you should let it do. As agents become more capable, permissions and authority become part of the design.

Related: AI agents aren't digital employees. Here's what they actually are..