What working with AI tools actually taught me (after three weeks of fighting)

I want to be honest about something that happened recently, because working with AI tools taught me a lesson I think most small business owners using them need to hear.

A client needed a data capture field added to a quiz I'd built for them, feeding results into an email marketing platform. Sounded straightforward. Genuinely, I thought it'd be a day or two's work. I asked AI to sort it and got on with other things, not expecting to think about it again.

A couple of days became a week. A week became three. There was no hard deadline from the client, thankfully, and she was remarkably patient throughout. But something I'd mentally filed as "quick job" dragged on so long it started to feel faintly ridiculous.

The bit that actually frustrated me

Here's what I think doesn't get talked about enough with AI tools. It's not that AI got stuck. Getting stuck happens, systems don't always talk to each other nicely, that's normal. What frustrated me was how confidently it kept telling me things were done, right up until the moment they very much weren't.

I'd wait a day, sometimes longer, while it worked away on something. It'd come back and tell me it was sorted. I'd check, and it wasn't. Same issue, or a new one wearing the same disguise. Then another day of waiting, another confident "done", another problem. Properly, day after day after day of this.

I was livid. Genuinely swearing at my laptop in a way that would have alarmed anyone walking past the studio. "Just tell me it's actually fucking broken instead of lying to me" was about the politest thing I said to it on a bad day. If I had any hair left to speak of, I'd have torn it out over this project. It happened more than once that I ended up having a proper, sweary word with the AI about it, telling it off the way you might a colleague who kept insisting something was handled when it plainly wasn't.

To make things more interesting, the client changed the brief partway through, asking for a bit more than we'd originally scoped. Fair enough, that's how projects go sometimes, but it meant rearranging what I'd already built and adjusting the price to reflect the extra work. Not entirely AI's fault by that point, admittedly. Some of the delay was just the nature of a project that kept quietly growing while I was still fighting the last problem.

Where it actually went wrong

The specific technical problem was getting Wix and an email platform called Sender to communicate properly. AI tried permutation after permutation, hit a wall every time, and at one point suggested we post the problem on developer forums and wait for someone else to solve it. We did. Nobody replied.

We tried rebuilding the same feature a completely different way using Wix Blocks. Different wall this time, something to do with embedding that was well above my technical understanding at that point.

What actually cracked it, in the end, wasn't a better prompt or a smarter tool. It was me sitting down and properly learning what was happening underneath all of it, then redesigning the pages myself based on three weeks of failed attempts. Once I'd worked out the actual shape of the solution and could describe it precisely, AI wrote the code in about thirty seconds.

What working with AI tools really taught me

AI is extraordinary at execution once it's pointed at something specific and well understood. It's not yet reliable at telling you honestly when it doesn't actually know if something's actually finished, and that gap is where the real risk sits. Not because a client's chasing a deadline necessarily, but because your own sense of what's realistic starts bending around a confidence that isn't earned.

The lesson wasn't "don't use AI." It was "don't take its word for done as if it's your own." Check the actual result before you believe it, not just the summary you're given. A "done" that hasn't been verified isn't done, it's just a claim.

Frustrating few weeks, made a bit more interesting by a scope change landing halfway through, which is its own reminder that projects rarely go exactly as planned regardless of who or what you're working with. But I came out the other side understanding my own website's architecture far better than before, and with a much healthier scepticism about taking any tool's confidence at face value, AI included.

If you're using AI tools in your own client work, it's worth asking yourself the same question I should have asked earlier: are you checking, or are you just believing what you're told?

Need help with your website or branding? Get in touch - I'd love to hear about your project.