AI News · Edition 69 ·
Your AI tools can change engines without telling you
Replit's switch from Anthropic to OpenAI is a nudge to ask your AI vendors three plain questions. Plus: sites blocking agents and Meta's fine print.
The one that matters
Replit moves its default AI from Anthropic to OpenAI, a reminder your tools can change engines quietly
What happened
Replit is a widely used service for building software apps, and some of its users have never written code. Its CEO says it now uses OpenAI's models by default instead of Anthropic's. At least one other report says the split is disputed, so the exact deal between the companies isn't public. The simple part is clear: a popular product changed the AI engine its customers get unless they go looking for a setting.
What it means for your business
You probably don't use Replit. But you very likely use, or are being sold, something built the same way. Think of a website chatbot, an AI receptionist that books appointments, a tool that drafts report-card comments, a WhatsApp assistant that takes orders, or an app that reads supplier invoices. Most of these products come from smaller companies that run on top of a model from OpenAI, Anthropic, Google or another lab. The company you pay is often not the company doing the thinking.
So three things can change while your contract stays exactly the same:
- Behaviour. A different model writes differently, follows instructions differently and makes different mistakes. Say a 600-student school uses a tool that drafts parent messages in a careful, warm tone. That tone could shift overnight without anyone touching a setting.
- Where your data goes. A new engine can mean a different company processes your text, possibly in a different country. Picture a 30-bed clinic that runs patient notes through a summarising tool. For that clinic this is not a technical footnote. It is a legal question under India's Digital Personal Data Protection Act or Canada's federal and provincial privacy laws.
- Price. What the engine costs the vendor changes over time, and what the vendor charges you can follow at renewal, up or down.
Even a vendor that never chooses to switch will eventually have to, because older models get retired on a schedule. One such retirement is in today's briefing.
What it does NOT mean
It does not mean OpenAI has “won”, or that Anthropic's models have suddenly got worse. Commentary that calls this a decisive round in a model war is written for investors and tech watchers, not for you. Replit's choice of default reflects its own costs, deals and product plans, and outsiders can't see any of those. It says nothing about which AI is best for a clinic's discharge summaries or a shop's product descriptions. For everyday office work like drafting, summarising and sorting, the leading models are close enough that the tool built around the engine usually matters more than the engine itself. That means how the tool handles your data, your workflow and your languages. And this is no reason to cancel a subscription or rush to back a winner.
What a sensible owner should do this month
- List your AI tools. Spend twenty minutes with whoever manages your software and write down every product that uses AI. Include AI features inside software you already pay for, such as your accounting, school management or clinic booking system.
- Ask each vendor three questions in writing. Which AI model do you use? Will you tell us before you change it? Where is our data processed, and is it used to train anything? If a vendor can't answer within a week, that tells you something.
- Keep a small test set. Save 15 to 20 real examples with personal details removed, such as typical parent emails, sample invoices or common customer questions, along with outputs you were happy with. When a vendor announces a change, or results start to feel off, run them again. It takes one staff member about an hour and costs nothing.
- Ask for notice in the contract at renewal. Asking for a clause that requires advance notice of any change of model or data location costs nothing. If you handle children's or patients' data, insist on it.
- Don't change anything because of this headline alone. If your tools work and your vendors give straight answers, you are fine. A 40-person workshop using an AI tool to write job cards doesn't need to act on Silicon Valley's partnership news. It needs to know who to call when the output changes.
Also worth knowing
How this briefing is put together
Every morning we read the day’s AI announcements and reporting from the companies themselves and from the technology press, then pick the handful that actually change something for a working business. The analysis is ours and it is written for owners and managers, not engineers. Every story links to its original source — read them, and disagree with us where we’ve got it wrong.