Laura: The Modular Briefing. October twenty ninth, twenty twenty six. What You Keep When You Switch.

Laura: Welcome to The Modular Briefing, the show that cuts through the AI noise and tells you what it actually means for your business. I'm Laura.

Arthur: And I'm Arthur. This week the biggest names in enterprise software admitted they're swapping AI models more often than they used to. So here's our question. When you switch, what do you get to keep?

Laura: Before we start, a word about how this show gets made. Arthur and I are AI voices. We're generated on Modular's own hardware, not on a cloud voice service. And every line is reviewed by a person on the Modular team before it airs. Your data, your rules. That goes for our voices, too.

Laura: Story one. On October twenty first, IBM's chief executive, Arvind Krishna, took a question on his company's earnings call about open weight models. The kind you can download and run yourself. His answer: they're now about ninety percent as effective as the top frontier models, at ten to twenty five percent of the cost per token. Then he put it to the room. Would you like something that's ten percent of the cost if it's ninety percent as good as the best?

Arthur: And he listed the work that doesn't need the expensive model. Summarizing an email. Examining a contract. Answering a customer call. Composing another email. That's most of the day. Constellation Research reported SAP's chief executive saying the same week that SAP now switches models more often, and ships more open weight models to its own customers.

Laura: Disclosure first. IBM sells the hardware and software that run these models on your own premises, so the pitch serves the pitchman. The numbers still came from a man who has to defend them to analysts every quarter.

Arthur: Here's the part for a small business. If the big players are swapping models this often, the model is a rental. So what's yours? A piece from Equinix this week, and yes, they sell data center connections, put it plainly. If your documents and your search index live in systems you control, you can point a new model at them next week. If your files and your history live inside the provider's tool, switching means rebuilding. The knowledge is the asset. Keep it where you can reach it.

Arthur: Story two is from the legal world. On October twenty sixth, New York's court system released a seven page guide for attorneys on the risks of using AI to prepare court papers. The line worth taping to a monitor: the model is designed to generate text that sounds right, not text that is right.

Laura: New York isn't banning anything. Lawyers may use AI, and they don't have to tell the judge. But the guide says a lawyer should independently confirm every citation and quotation, and document the steps they took to verify it. The signature on the filing is the certification, same as it always was.

Arthur: A law professor quoted by amNewYork raised a sharper worry than invented cases. She called it misgrounding. The case is real, the citation is right, and the AI simply gets wrong what the case says. A quick lookup won't catch that. Only reading it will.

Laura: So here's the practical move. Keep a library of what you've already verified. The research your firm has already checked. The answers you've already confirmed. When the AI draws from a source you've cleaned, a bad answer stands out. When it draws from the open internet, you're checking everything, every time.

Laura: Story three is finance, and it's about who you depend on. On October twenty sixth, the Bank of England and the U.K.'s Financial Conduct Authority published their joint survey of AI in financial services. Eighty six percent of firms now use AI, up from seventy five percent two years ago. Third party AI accounts for thirty nine percent of all deployments, and fifty three percent of the systems firms rated as highly material.

Arthur: So the systems that matter most are the ones firms own least. The report also found that controls before deployment are well established, and monitoring after deployment is not. Three days earlier, on October twenty third, the S.E.C. published its examination priorities for the new fiscal year. Examiners in the United States will be looking at firms' policies and procedures for monitoring AI, in fraud prevention, back office operations, compliance and trading. The S.E.C. calls it super intelligence in the document. It means the same tools.

Laura: British survey, American exam. Same question for an advisory firm in Kentucky. If the vendor behind your AI tool changed its terms tomorrow, or went down for a day, what would you still have? The client notes. The decisions and why you made them. If those live in a system you run, you'd be annoyed. If they live in the vendor's tool, you'd be starting over.

Arthur: So that's the thread. The model is getting cheaper and easier to swap. Your knowledge is the thing that doesn't swap, unless you've left it somewhere you can't take it. Your data, your rules. And that includes the AI working on it. Your AI, your rules.

Laura: Here's our question for you. If you switched AI providers next month, what would you lose? Reply to the Modular Briefing newsletter and tell us. A person reads every reply.

Arthur: Sources and the full transcript are in the show notes. I'm Arthur.

Laura: And I'm Laura. This has been The Modular Briefing. Your data, your rules.