When to hire an AI agent development company, and when not to

A team left an established construction software company and went off to build a rival with AI built into it properly. Around fifty or sixty pounds a month.

When somebody looked into it, the domain had been registered a year earlier.

The original company then started hiring people to build their AI side.

I think about that one a lot, because it is the version of AI risk that actually turns up in small businesses. Not a machine taking anyone's job. A competitor getting quicker and cheaper at the same work while you were waiting to see how it all shook out.

The question you are actually asking

Most people who go looking for an AI agent development company are asking one of two very different questions, and it is worth knowing which.

The first is: I have a job in this business that no software does, and I want something built that does it.

The second is: everyone says we should be doing something with AI and I am not sure what.

Only the first one should be buying a build. The second should be buying a fortnight of somebody's attention and a shortlist.

What the evidence says about building versus buying

The British Chambers of Commerce, with the University of Essex, surveyed 668 UK firms, most of them small. Among businesses using generic AI tools as bought, 3 per cent reported a headcount decrease. Among those using bespoke AI, it was 21 per cent.

Their own authors say the data is cross sectional and should not be read as causal, and the bespoke group was small at around 65 firms. So treat it as a signal, not a proof.

But the direction is worth sitting with. Building something bespoke is a different kind of decision from buying a licence. It changes more, it costs more, and it is much harder to reverse.

There is a companion figure worth knowing. Deloitte, in its State of AI 2026 report, found that only 25 per cent of organisations have moved 40 per cent or more of their AI pilots into actual production.

Deloitte sells the kind of consultancy that finding argues for, and its respondents are large organisations. Read it with that in mind. Most builds do not survive contact with the business they were built for.

What a good build looks like

Small. Specific. Attached to a job somebody already does every week and can describe in a sentence.

A supplier sends a quote as a PDF with thirty eight plant lines on it, and somebody has to type all thirty eight into a spreadsheet before anyone can compare it against the other quote. Build the thing that does that. One clear job, a defined input, an output somebody can check in a minute, and a person who actually wanted it.

The bad shape is a platform. Something that will eventually handle quotes and scheduling and reporting and customer comms, delivered in phases, where phase one is a login screen.

If the first thing you are shown is a login screen, ask what it does before the login screen exists.

What you could do on Monday

Write one sentence describing the job you want done. Not the system. The job.

"Take the supplier PDF and give me the lines in a spreadsheet." That is a sentence. Someone can build that, and you will know within a fortnight whether it works.

"Streamline our quoting" is not a sentence. It is a heading, and it is how projects end up costing forty thousand pounds and getting quietly shelved.

If you cannot get it into one sentence yet, that is genuinely useful information. It means the next thing to buy is somebody's time to work out what the sentence is, not a build.

The honest limit

A bespoke thing is a thing you now own.

Somebody has to maintain it. When the tool it sits on top of changes, and it will, something breaks and somebody has to fix it. That somebody is either you, or a retainer, and both of those are real costs that rarely appear in the excitement of the first conversation.

For most small businesses, off the shelf tools plus somebody who has been shown how to use them properly will get you further than a build, for less money, with nothing to maintain.

I am aware that building things is the more profitable half of what I do, and I am telling you to buy less of it. That is not modesty. It is that I would rather do the small thing that works than the big one that gets abandoned in March.