Robots and AI - Cheap? Safe? - I don't think so

I read an article in the Telegraph a little while ago, and with the news this week about AI needing regulation there is something I keep thinking about and I wanted to share it with you.

It feels like we've all spent the last two years being told the same thing: get the humans out, get the AI in, and if you haven't automated something by now you're already behind. Boards of businesses bought it; specialists in bringing in AI and kicking the people out seemed to pop up, and inevitably layoffs and redundancies got announced with "AI" as the explanation. But two years in, two things are surfacing at once; AI is costing more than the people it was supposed to replace, and now the people actually building it are telling us, out loud, that it might not be safe either.

Forbes' Jemma Green wrote about it properly in July. Companies laying staff off to fund AI tools that then turn out to cost MORE than the salaries they just cut. Uber's the poster child - its CTO reportedly blew through the company's entire 2026 AI coding budget in four months flat, with most of its code now AI-written and, rather awkwardly, no one able to draw a straight line between that spend and anything actually useful to their users. Microsoft has been quietly pulling licences back in several departments for the same reason. And an Nvidia exec put it about as bluntly as you can: AI is now costing them more than their own people do.

The Telegraph's Andrew Orlowski made the point that should have been obvious all along: if the AI costs more than the person, why on earth are you replacing the person? It's the kind of line that sounds almost too simple, right up until you realise it undoes most of what businesses have spent this year being told to believe.

There's a China angle too. As Western firms are starting to work out how eye-wateringly expensive the AI models are to run properly, a lot of them have apparently quietly switched to cheaper Chinese open-source models instead. Usage has gone from next to nothing to the majority of the market in about a year. Make of that what you will politically, but in reality, even the companies building this stuff for a living are following the price tag, and the price tag is saying the "replace the humans" version of AI is currently overpriced and costs a lot more than the headlines suggest.

None of this makes AI the villain. Both pieces cited the same MIT/IBM study, finding only around 23% of workers' wages could actually be replaced by AI cost-effectively right now - not because the tech can't do the job, but because doing it properly, without a human quietly checking its work, still costs more than just paying someone to do it well in the first place.

Just as an aside, this cost isn't only financial. There's the data centre build-out, the water usage, the environmental impact - and that's a whole other conversation in itself.

This week, a researcher at Anthropic (the company behind Claude) said publicly that he thinks there's more than a 10% chance AI could kill everyone within the next decade. He said it after a former colleague resigned and accused the industry of racing towards more powerful, less controllable systems faster than anyone knows how to keep them in check. Geoffrey Hinton, one of the people the whole field credits with getting AI off the ground in the first place, has said much the same for a while now. Even Elon Musk, who has plenty of AI skin in the game himself, puts the odds at somewhere around 10-20%.

I'm not writing this to be alarmist, and I'm certainly not equipped to tell you whether 10% is a real number or a rounding error. But when the people who built the thing are the ones raising their hand and saying "we don't yet have a plan for this," it's worth sitting with for a second, not scrolling past.

Put the two things together and you get a strange but honest picture. The version of AI that's supposed to replace your people is, right now, too expensive to make sense. And the version of AI that's powerful enough to actually be dangerous is, by the admission of the people building it, not yet safe enough to hand real control to. Neither of those is an argument for "go all in." Both of them are an argument for "know exactly what you're asking it to do, and keep a human in the loop."

Sadly with my recruiter hat on, we all know and have seen that hiring in this country has got more expensive. Rachel Reeves has seen to that with higher employer National Insurance, a higher minimum wage and less room to justify a hire that isn't pulling its weight from day one. So it's no wonder agency owners are eyeing up AI and thinking "surely this works out cheaper."

Sometimes it does. Sometimes it very much doesn't. Something that answers out-of-hours valuation enquiries, or chases referencing paperwork through a lettings pipeline, is genuinely cheap, genuinely usefull and probably something you should look at if you aren't doing it already.

Asking software to replace a good negotiator's read on a wobbling chain - or a difficult vendor conversation, or a first-time buyer who needs a human to talk them through the process and guide them on such a massive decision.  It's a different job entirely. Right now, the evidence says that's still done better by a human - and quite possibly the riskier kind of AI to lean on, too.

The honest answer for property businesses isn't people versus AI, it's how they can work together to provide the best service at the right cost, with someone still firmly holding the wheel. It's about working out, role by role, which parts are best done by AI and which need a human - and having the guts to admit when the shiny new tool costs more than the person it replaced.

The firms that get this right won't be the ones who went all in one way or another; they'll be the ones who worked out where a person is worth paying a salary to and where a tool genuinely earns its keep, safely.