AI Pulse · AI Trends Pulse

Anthropic just flipped a switch that changes how Claude Code picks which model to use for each coding task. “Auto mode” used to be something you turned on. Now it’s the default, and you have to turn it off if you want fixed behavior.

What this means in practice: every time you hand Claude Code a task, it now decides on its own whether to use a faster, cheaper model or a slower, more capable one. That sounds helpful, and sometimes it is. But if you’re running agent workflows in production, you’ve just lost a layer of predictability. Your costs can swing based on decisions the system makes, not ones you make. Your response times vary. Your error rates might shift because a different model handled the same prompt yesterday versus today.

This isn’t theoretical. We run Claude Code workflows at Enterprise DNA, and the moment this rolled out, we had to audit every pipeline to understand what changed and whether we needed to lock behavior back down. If you’re using Claude in any repeatable business process, billing forecasts, SLA commitments, or compliance logs, you’re now managing a moving target unless you explicitly opt out.

The broader pattern matters more than this one toggle. AI providers are increasingly making decisions for you in the name of convenience, and those decisions have cost and reliability consequences. A command centre approach, where you route tasks through logic you control and can see, lets you decide when auto-selection helps and when it doesn’t. That’s the kind of thing we build into an AI command centre, so you’re not surprised by a blog post that changes how your tools behave.

If you’re using Claude Code in production, check your settings today. If you need consistent behavior, turn auto mode off. If you’re fine with variance, at least track what it’s costing you.