It has been a whiplash of a week for AI. Days ago, some of the biggest names in the industry called for a slowdown, warning that frontier models are advancing faster than the guardrails. Then President Trump dismissed those warnings, saying the only guardrail AI needs is him. So which way is the industry actually leaning? Conor Twomey is Co-Founder and CEO of AI One. Conor, welcome back to the show.
Thanks, Johnny. Delighted to be back.
AI companies are saying slow down. The White House is saying speed up. How do you break down this week for someone in your position?
There is a lot of noise out there and a lot of it is fabricated. When we think about AI, there are two modes for the frontier models — building new models, which is model training, and the use of those models, which is inference. For all the practitioners in the audience, the slowdown talk has no impact on how you are using AI on a day-to-day basis. This is a commentary on what happens next. And what we say to customers is you need to put the guardrails in place. You cannot depend on AI blindly. Think of it like a genius trader — you cannot bring one in and just let them do whatever they want. You need the right guardrails, the right limits, the right checks and balances.
AI stocks and chip stocks fell on the slowdown talk. How does that change how much companies are spending on AI?
It is a CapEx trade versus an OpEx trade. CapEx is bigger data centres, more chips for model training — all about the frontier. If there is a slowdown, we might need fewer chips, smaller data centres, less energy. That is the CapEx sell-off we saw. But on the OpEx side — the actual usage of AI models — spending is going through the roof. One of our customers is effectively doubling their bill every 45 days. There is no slowdown. A lot of redundant spend is happening though — it is very similar to the transition from steam technology to electricity. A lot of people are plugging this new probabilistic technology into the old way of working and it is utterly redundant. Learning to harness this new type of technology is the biggest challenge.
Is it safe for a bank to be using AI right now? When the slowdown warnings came out, what did customers say?
People in general are worried. We are seeing a lot of AI anxiety in society — it is becoming a polarising issue heading into the midterm elections. But most of our customers are still at steps one, two, and three of a ten-step journey. When people talk about slowdown and existential worry, that is step ten. Studies have shown that productivity has not materially increased yet. Very little has transcended from personal productivity into actually changing how businesses operate.
Where do banks and companies stand now in terms of AI provider dependency?
The biggest lesson from everything we have seen is that you need to be able to reverse any key decision. If you are using AI as part of mission-critical operations inside a bank — supporting the desk, managing risk, creating research — you cannot be beholden to one provider. Build your systems in a way that does not depend on any single model. As the models change, you should be able to swap them in and out and reverse decisions without needing to reactivate the overall system. The people who are ahead versus the laggards — that is the number one determinant.
What is your single piece of advice for someone watching this?
Find a trusted partner. There are 50,000 companies on planet Earth purporting to be AI companies. Trust is the one thing AI cannot replicate. Find a trusted partner and lean on that partner to separate signals from noise on your behalf.
Thank you so much, Conor.
Correct. Thanks, Johnny. Thank you for being here.
