There’s no shortage of AI ideas right now. If anything, there are too many. Everyone’s got one. Every pitch has it. Every product has something vaguely “AI-powered” wedged into it somewhere.
You can generate ten ideas before lunch and probably build something that looks like a product by Friday, which feels impressive until you realise most of it isn’t actually doing anything useful.
On the surface it all looks like progress. Lots of movement. Lots of noise. Lots of new things being launched every week. But when you scratch a bit deeper, a lot of it feels half-formed, rushed, and slightly desperate to prove it’s keeping up.
Even the big players aren’t getting it right (yet)
One of the more interesting things has been watching the bigger companies move at speed. You’d expect them to lead here, but in reality they’re often just reacting like everyone else, just with bigger teams and better PR.
I tried using Google’s Gemini inside Google Drive during our M&A process, thinking it might actually be useful for once. We had loads of documents, around 150 due diligence questions, and it felt like the perfect use case.
In my head it was going to pull things together, summarise properly, maybe even organise files in a way that saved me time.
It couldn’t really do any of it. Not properly anyway. It struggled to summarise, couldn’t reliably group related files, and definitely wasn’t doing anything clever like pulling invoices together into something useful.
For a company of that size, it was a bit… underwhelming, which probably says more about where things are than anything else.
The pattern I keep seeing
That same pattern keeps showing up everywhere else. Startups adding “AI features” to recruitment platforms. CRM systems that help you draft messages that all sound like they were written by the same slightly awkward robot. Tools that promise efficiency but mostly just add another layer of noise.
I’ve sat through plenty of sales calls over the last 18 months where you can see what they’re trying to do, but it’s just not quite there. Even in sectors like MedTech, where you’d expect more rigour, some of it feels second rate once you get past the demo.
The common thread isn’t the tech. The tech is actually impressive. It’s the thinking behind it that’s often shallow, starting with “where can we add AI?” instead of “what problem actually needs solving?”
Where AI actually works well
To be fair, it’s not all bad. There are areas where AI is genuinely well thought through and actually useful, and you can feel the difference straight away because it’s solving a real problem rather than just adding a feature for the sake of it.
Predictive maintenance is a good example, especially where you’ve got large data sets and sensors everywhere. Things like machinery, vibrations, performance over time. When AI is used to spot anomalies and patterns before something breaks, that’s properly valuable. It saves money, reduces downtime, and makes sense of data at a scale humans just can’t.
It’s doing what it should be doing. Working in the background, spotting signals, quietly improving outcomes. Not shouting about itself, just being useful.
Recruitment is another interesting one, but only when it’s done properly, which it rarely is at the moment. Most are in danger of turning a people-based business into something that could easily backfire if not positioned and delivered carefully.
Good recruitment isn’t just matching a CV to a job description. It’s about aligning three things: what the candidate actually wants for their future career, what the company genuinely is, and whether the role and stage of the business fits where they’re trying to get to.
Most processes only match one or two of those, which is why people all too often move on quickly.
If AI can help spot those deeper patterns across conversations and behaviour, brand and role progression, things that would normally take loads of interviews to uncover, then that’s valuable. Better matches, longer tenure, and a process that actually works for both sides.
The common thread in both of these is intent. The AI is grounded in a real problem, backed by data, and designed to improve something that already exists rather than pretending to be the product itself.
When AI is doing something quietly useful, it tends to work. When it’s the headline feature, it often doesn’t.
AI makes it easier to build. Not easier to build something good.
This is the bit that gets lost in the hype. AI makes it much easier to start things. To prototype, generate, automate, and move quickly. You can get something up and running in a fraction of the time it used to take, which is genuinely powerful.
I’ve been enjoying using vibe coding tools like Loveable and Base44 (more on that in a future article). But pace and tools don’t improve judgement. They don’t tell you whether the problem is worth solving, whether anyone actually cares, or whether the thing you’ve built fits into the real world in a meaningful way.
If anything, it just means you can build the wrong thing faster, which isn’t really progress. It’s just more efficient disappointment.
The temptation to move too quickly
There’s a bit of pressure creeping in. No one wants to feel like they’re behind. No one wants to be the company that “isn’t doing AI”. So you get this rush to add features, ship things, say something, anything, just to show you’re in the game.
From the outside it can look like innovation. Lots of updates. Lots of announcements. Lots of activity. But inside, it often creates more complexity than value. More features without much thought about whether they’re actually helping anyone.
It’s the classic case of doing more, not necessarily doing better.
A more considered approach
What I’ve found, both in my own work and just watching how things play out, is that holding back slightly is often the smarter move.
Taking a bit more time to understand the problem properly, testing things in a grounded way, and not rushing to ship just because everyone else is. It’s not as exciting. It doesn’t give you a flashy LinkedIn post every week. But it tends to lead to something more useful, more coherent, and ultimately more valuable.
There’s also something quite powerful in making that part of how you show up. Not chasing every trend. Not jumping on every wave. Being known for doing things in a considered way, even if it means moving a bit slower.
There’s a bigger shift coming here as well, and it’s probably a topic for another article, but I suspect we’ll see AI app layers or agent ecosystems start to replace the need to bolt AI into every SaaS product.
We’re at the start of a user experience shift where clicking filters and filling in forms could be replaced by agents acting on our behalf. Not agents embedded into every individual product, but ones that sit closer to us, connected to our calendars, bank accounts, preferences, and able to interact with services through APIs in a much more natural way.
If that’s where things are heading, then holding back and thinking more carefully about where and how you apply AI makes even more sense. There’s no point getting ahead of the curve in the wrong place.
Where judgment comes in
This is where the real gap is opening up... Ideas are now cheap. Tools are accessible. Execution is faster than it’s ever been. The barrier to entry has dropped massively, which is brilliant, but it also means there’s a lot more average stuff out there.
What hasn’t changed is the importance of judgement. Knowing what’s worth pursuing, when to move, when to wait, how far to take something, and when to stop adding things that don’t need to be there. AI is powerful, no question, but it’s still just a tool.
The businesses that do well won’t be the ones shouting about it the loudest. They’ll be the ones quietly using it with intent, solving real problems, and not getting distracted by every new shiny thing that comes along.
Just because you can, doesn’t mean you should. AI needs to be designed into the system, aligned with your values, your brand, and your customer, not layered on top of it.
