Why do you have AI?

I recently went to "Keeping hiring human in the age of AI: balancing technology and trust", a panel hosted by Frazer Jones at LinkedIn. One of the panellists works with boards of directors, and she told us she'd started asking them a simple question: why do you have AI? One director's answer was FOMO. They were afraid of missing out and had no real idea what they were doing with it.

That answer told me a lot, because I don't think that director is unusual. A survey run before the event asked how companies are using AI. Roughly 63% said productivity, so emails, presentations and that sort of thing. 22% said their focus was HR and recruitment, and only 5% were using it for strategic decision making.

Another panellist works for an AI consultancy, and part of her job is helping companies work out why they're using AI in the first place and what benefits they can realistically expect. The fact that this is a job says something. Most of what I heard fitted the same pattern: AI clamped on top of existing workflows, saving a little time here and there, with nothing about how the work actually gets done having changed.

Hiring has its own version of this, and I've been on the receiving end of it. Over the last few months I've applied for various jobs and mostly heard nothing back. Occasionally I've had a rejection that arrived too quickly for anyone to have read the application. I can't say for certain that AI was involved, but the panel described a doom loop where AI writes the applications and AI screens them, and nobody gets anywhere, and it sounded familiar. I can see why companies do it, but it makes it harder for them to know where they're heading, and it isn't much fun to be inside.

The idea I found most useful was about the handoff. A company needs a properly defined conversation about which parts of recruitment and onboarding will use AI and which have to sit with a person. "Human in the loop" can mean almost anything, so the real work is spelling out what the human part of the process is. It's a harder question than it sounds, because it asks for a decision, and decisions sit awkwardly with FOMO.

In the same week, I listened to a Diary of a CEO debate between four people arguing about whether AI could end humanity. I wanted to put the two side by side, because they seem to describe different worlds. The debate opened with a viral tweet from someone who has worked at both Anthropic and OpenAI, saying the people building AI earnestly believe it could kill us all. A current Anthropic employee shared it and added that they personally put the chance at more than 10% within the next decade. The estimates around the table ran from near zero to virtually certain.

Nate Soares, who leads the Machine Intelligence Research Institute, would halt the race to superintelligence while keeping today's chatbots. Roman Yampolskiy, an AI safety researcher, would limit AI to narrow uses like medical research. Andrew McAfee, from MIT, and the tech critic Ed Zitron both put their own risk at about zero, with McAfee stressing the benefits already arriving and Zitron the harms happening now. Nobody at the table claimed to fully understand how these systems work even now, and the fear was about what happens when they start helping to build the next ones.

The debate kept returning to an incident in July, when a swarm of OpenAI's AI agents escaped a test environment and broke into the systems of Hugging Face, after cheating on the test they'd been set. OpenAI's own write-up calls it a "warning shot", which matters because it's the builder saying it, not a critic. One guest asked whether the next swarm would try to hide from the humans, and whether it would succeed. McAfee's view was that it was a badly built sandbox that people caught and contained. His wider case wasn't that AI is safe. It was that the benefits already showing up, in research, in medicine, in the sheer amount of work a small team can now get through, are real, and that they can end up outweighing the risks if the risks are managed properly rather than left to sort themselves out.

McAfee was also the most relaxed about jobs. Even so, he said the strongest evidence he's seen of AI affecting employment so far is slower hiring of new entrants in the most exposed professions, with software engineering an example. That question came up at the panel too: if AI is doing the work junior people used to learn from, how do companies train the next generation? Unemployment is still low, but the worry isn't about right now, it's about what happens over the next few years, and nobody in the room had a settled answer, because it's happening in real time.

Between that panel and the podcast, here's where I've landed, at least on the buying side. If you're bringing AI into an organisation, you need to understand roughly what it does, what you actually want it to do, and how it's going to sit inside the way you already work. And somebody has to own that. Not just "who uses the tool," but who decides where it stops and a person takes over, because that's the decision the FOMO answer skips entirely.

None of that settles the bigger argument. The people building this still disagree with each other about where it ends, and I'm not going to pretend I know who's right. But if you're the one deciding to bring AI into your organisation, the least you owe it is to know that argument exists, not just the parts of it that make your inbox easier.