The People Building AI Are Asking to Slow Down

The People Building AI Are Asking to Slow Down. Real Estate Agents Should Pay Attention.

When the CEO of the company behind Claude says the AI industry needs to slow down, I’m going to pay attention. Over the weekend, Dario Amodei published an essay calling for a more measured pace of AI development, and both Sam Altman and Elon Musk expressed agreement. Altman also said OpenAI would match the commitment to give independent evaluators access inside the company. These are people competing to build some of the most powerful technology in the world, so their agreement deserves more than a quick reaction to a headline.

Amodei’s concern is that advances in capability could outpace the work needed to understand and control these systems. He wants more time for testing, safeguards, and independent review, along with coordination between companies and governments. He specifically says this does not mean stopping model training or technical progress. One of his concerns is AI becoming increasingly capable of helping build the next generation of AI, accelerating development further. That is a serious issue, and it deserves to be understood before we turn it into another argument about whether AI is good or bad.

My takeaway for real estate agents is that we need to get more deliberate about how we use this technology. I can already imagine someone reading the headline and deciding they were right to avoid AI, while someone else wonders whether they should stop using it altogether. I wouldn’t make either decision based on this announcement. I would look at where AI is being used in my business, what information it can access, what it has permission to do, and who is checking the work. Those are questions we should already be asking, regardless of what the leaders of these companies say.

I also don’t believe most agents need a more powerful model before they can get meaningful value from AI. There is plenty of work to do with what’s already available, especially if you’re still trying to figure out how to use it consistently. You can spend the next six months chasing every announcement, or you can spend that time learning how to apply the tools to problems you actually have. I care much more about whether an agent has improved their follow-up, prepared more thoroughly for an appointment, or reduced the hours they spend on administrative work. A new model announcement doesn’t accomplish any of that for you.

There’s another part of this I want to address because I’ve written about building systems at Incyte that learn and adapt. Amodei is discussing AI helping develop more capable AI, with the risk that progress becomes difficult to understand and control. Our work involves reviewing how a system handled a specific responsibility and using that feedback to improve the process. That could mean identifying a question Morgan struggles with or understanding why an agent corrected part of a market analysis. We’re developing ways to bring those issues forward, evaluate changes, and test whether they help. We are not giving Morgan permission to independently redesign the underlying AI model or expand its own authority.

That distinction matters, but it doesn’t excuse us from being careful. I don’t let AI delete an email from my inbox without my approval, and I’m certainly not comfortable letting it change how we communicate with clients without someone reviewing the change. Trust but verify and inspect what you expect apply to our technology just as much as they apply to the brokerage. We still need to check the information, review the results, and understand what the system is allowed to do. Parts of what we’re building are working today, and other parts are still in development. Reading an essay about AI safety doesn’t prove our approach is right, but it reinforces why that oversight needs to remain part of the work.

For an agent, the responsibility is very practical. If AI helps prepare your CMA, you need to understand and verify the analysis before you recommend a price. If it writes a listing description, you need to make sure the property actually has the features it describes. If it drafts a client message, read it and decide whether it reflects what you mean and what that person needs to hear. A confident answer can still be wrong, and putting your name on it makes the review your responsibility. The time you save preparing something should give you more room to think through it carefully.

I want agents to take this seriously without becoming so intimidated that they refuse to learn. Your clients still need someone who can understand their situation, explain their options, negotiate for them, and take responsibility when a transaction gets difficult. Technology should help you do those things better, and learning how to use it is part of developing your business. The agent across town who is doing that work is improving while everyone else argues about the headlines. You don’t need to become a developer, but you do need to understand the tools you choose to rely on.

A call to slow development also doesn’t mean the industry has already slowed down or that anyone has guaranteed us extra time. I wouldn’t build my plans around that assumption. What we can control is how seriously we take our own learning, the standards we set for the work, and the attention we give the people trusting us. That’s where my focus remains at Incyte, and it’s what I want our agents focused on too. If you want to understand how we’re approaching it, come spend an afternoon with us and see what we’re building, how we review it, and where there’s still work to do.