GitHub is making automatic model routing the whole experience for Free and Student Copilot users, which says a lot about where mainstream AI products think choice should live.
AI products love to advertise choice right up until they decide choice gets in the way.
That is the more useful reading of GitHub's June 24 Copilot update.
The company says Copilot Free and Student plans will now use auto model selection as the default and only model selection experience. Auto dynamically chooses the best model for each task across multiple model families, subject to plan restrictions. GitHub is also retiring the (Preview) label from Microsoft-released models, on the logic that auto routing and behind-the-scenes improvements now matter more than those labels for broad user guidance.
This is bigger than a settings cleanup. It is a product statement about where model choice belongs.
For Free and Student users, GitHub is effectively saying: stop picking, start using.
That matters because the product is no longer teaching users to think in terms of named models first. It is teaching them to think in terms of outcome quality delivered through product-managed routing.
In some ways, this is the natural next step after the control-surface expansion Butler covered in GitHub's terminal workbench release and the higher-tier governance split in Copilot CLI BYOK. GitHub seems comfortable giving more explicit choice and policy control to users who need it, while hiding more of the complexity for broad entry tiers.
Automatic routing does not remove governance. It relocates it.
Once users can no longer pick a model directly, the governance question moves into product policy: which model families are eligible, under which constraints, with what tradeoffs, and how visible are those tradeoffs to the person using the tool?
That is why the removal of (Preview) labels matters. GitHub is signaling that brand-level caveats should matter less to the everyday user because the product intends to absorb more of the routing judgment itself.
That can improve usability. It can also make product boundaries more opaque if users cannot tell when a weaker tier limit or routing decision shaped the answer they got.
The promise is obvious: fewer confusing choices and better default outcomes. Teams should still compare whether the routed results feel meaningfully better than what users would have picked manually.
GitHub says auto spans multiple model families subject to plan restrictions. If restrictions remain invisible, users may experience capability cliffs without understanding why.
Entry-tier product decisions often become previews of broader UX strategy. If auto-routing works, more products will decide that most users should never see raw model choice unless they have a premium or administrative reason.
Mainstream product users may not need a selector. Operators, developers, and enterprise buyers often still do. That is why it matters that GitHub keeps other, more explicit governance surfaces elsewhere in the stack, including instruction control surfaces Butler covered in AGENTS.md-aware review flows.
The interesting question is not whether this makes Copilot simpler. Of course it does.
The interesting question is whether AI products are quietly training users to stop thinking about models at all. Once routing becomes the product, the competitive surface changes. The product wins not because users picked the best model, but because the routing layer made the right choice often enough that nobody asked.
That also changes cost and evaluation conversations. A system where the product chooses the route is harder to inspect casually, which is part of why Butler keeps coming back to the operational and cost side of coding tools in pieces like the real review-cost surface behind Copilot automation.
GitHub's June 24 change matters because it turns entry-level Copilot into more of a routing product.
If that pattern sticks, the next mainstream AI fight will be less about who exposes the most model names and more about who hides model choice well enough that the product itself becomes the decision-maker users trust.
This article was researched and drafted with AI assistance, then reviewed and edited for clarity, accuracy, and editorial quality.