Hiring an AI engineer instead of a freelancer

A competent freelance developer bills somewhere between a few hundred and a few thousand for a small piece of work, depending on where they are and how specialised it is. That number buys judgement as well as typing, which is the part people forget when they compare it to a subscription.

An agent is a monthly fee — tens of dollars, not thousands — and the marginal cost of the eleventh small task in a month is nothing. That changes behaviour more than it changes budgets: you stop rationing requests, and the twenty small annoyances that were never worth invoicing finally get fixed.

Where that flips: a job with real consequences. Anything where being wrong costs more than the whole year’s subscription is not the place to be economising on judgement.

Finding, briefing and trusting a freelancer takes days at best. Their first week is partly spent learning your setup, which you pay for and which is entirely reasonable.

An agent starts in minutes and has no calendar. That advantage is largest for the things that are urgent and small — the broken form on a Sunday, the price that is wrong during a promotion — and smallest for the things that were always going to take a fortnight of thinking.

Being available at 2am is not the same as being right at 2am. Speed is a real advantage; treat it as one advantage rather than as the answer.

This one is underrated and it decides more projects than price does. A freelancer who built your thing in March remembers in June why the weird bit is weird — if they are still available, which is the catch: they have other clients, and the good ones get busy.

An agent with a persistent machine keeps the work, the notes and the history in one place, and it is available in June by definition. What it does not have is the unwritten context — the reason you do it this way, the customer who complained, the thing you would never do again. That lives in your head, and you have to say it out loud, every time it is relevant.

The practical middle ground most people land on: the agent holds the code and the routine work; you keep a short written record of the decisions and why. That document is worth more than either of them.

Here the freelancer wins outright, and it is not close. A person can be accountable for an outcome, can say “that approach will not work and here is why”, and can carry professional liability. An agent can do none of those things. It will build the bad idea you asked for, well.

So keep a human in the loop wherever a mistake reaches customers, money or the law — which in practice means someone looks before a change goes live. Any agent worth using makes that step structural rather than optional; ours will not land a change to your code until you tap approve, and cannot spend your money at all.

Give it to an agent when the job is small, well-defined and repeatable; when it is urgent and low risk; when there are twenty of them and each one is individually not worth an invoice; or when you want to find out whether the idea is worth paying a professional for.

Hire a person when the design is the hard part; when the work is regulated, safety-critical or legally binding; when it needs someone to sit with your team and argue; or when you need somebody to be answerable if it goes wrong.

Do both — the arrangement that works best in practice. The professional decides the shape and reviews what lands; the agent does the volume in between. That is roughly how a senior engineer works with a junior one, and it is a well-tested division of labour.

Then the honest advice is to start with something small enough that you can judge the result yourself. You cannot review code, but you can absolutely check whether the booking page books. What to hand an AI engineer in the first week is written for exactly that reader, and the small business page lists five jobs that reliably land.