Practical AI

DIY or hire someone: where the line with AI really falls

I will start with the part that works against my own trade: a good deal of what AI can do today, you should do yourself, and pay no one for it. Writing, summarising, brainstorming, drafting a letter, roughing out an idea, all of that can be done with a £25 a month subscription and a bit of practice. If a provider offers to "manage your AI" at this level, they are selling you nothing. So the real question is not "do it yourself or get it done", it is: where exactly does the line fall?

The "do it yourself" side

The rule is simple. As long as the task touches a single tool, stays in front of your own eyes, and a mistake would be spotted immediately, do it yourself. You open an assistant, ask it, read it back, decide. If the result is poor, you see it and start again. The risk is nil, the learning is fast, and no one will do it better than you because no one knows your trade and your customers as well as you do. Paying someone for this level is paying someone to type in your place. The only investment worth making here is not a provider, it is an hour spent learning to use it well, and it pays for itself within the first week.

As long as a mistake is spotted straight away and costs nothing, do it yourself. It is when it becomes invisible that you need someone.

The "get it done" side

The line is crossed at one specific point: when several tools need to be wired to each other, when it runs without you, and above all when a mistake becomes invisible. Take a system that sends messages on its own to hundreds of customers, or that copies data from one piece of software to another overnight. If it gets something wrong, you do not see it happen, and by the time you notice, the damage has repeated itself dozens of times. That is where a professional earns their fee: not for wiring the tools together, plenty of people can do that, but for putting in the safety nets, the checks, the backups that catch the mistake before you do. You are not paying for the connection, you are paying for the safety net.

The day I charged to say "you are on your own"

One case stuck with me because it cuts against the idea that a provider always tries to sell you the biggest package. Someone came to me with a long list of things they wanted to "put into AI". Together, we did a proper analysis, a paid piece of work, to lay it all out. Conclusion: behind his long list, his real need was quite simple, and he could sort it out alone, with the right off-the-shelf tools and my recommendations. He left without buying a project, and pleased about it. He had paid for clarity, not for execution. To my mind, that analysis was worth more than a project sold by force: it saved him from spending thousands of pounds on something he already had in hand.

That story sums up my conviction: the job of a good provider is not to do for you everything you could do yourself, it is to tell you where the line falls, then to hold the part that genuinely cannot be bodged. Many people leave my office with two or three recommendations and a subscription to set up on their own. They come back later, when a real project appears, the kind where a mistake repeats silently and costs dearly. On that day, they know exactly what they are paying for, and it is not for typing.

The test to run on your own. For each thing you want to hand to AI, ask three questions: does it touch a single tool? Does it stay in front of my eyes? Would a mistake be spotted straight away? Three "yes": do it yourself, keep your money. A single "no", especially on the last one, and it is time to talk to someone, not for the technical side, but for the safety net.

Do watch out for the opposite excess too, the "I will do everything myself" on principle. Spending twenty hours painfully learning what a professional would set up cleanly in two is not a saving, it is your time going somewhere other than where it is worth the most. The right line is never "can I technically do this", it is "is my time better spent here than serving my customers". For everyday writing and thinking, train yourself and do it alone: one hour is enough to get started. For anything that runs without you and can go wrong silently, that is the territory of automation set up and watched over. Between the two lies your judgement, and that is exactly why it cannot be outsourced.