Multi-Model AI Coding Platforms Absorb the Developer Tool Stack

What happened

For the past year, the standard advice for AI-assisted development has been to mix and match: one tool for autocomplete, another for large refactors, a third for review. That pattern is now being pulled inside the platforms. GitHub Copilot and Microsoft's broader Copilot products have become multi-model shells that route work to Anthropic, OpenAI, Google and Microsoft models from one interface. At the same time, Microsoft is reportedly pushing customers off a standalone rival, Claude Code.

The result is a shift in where the multi-tool decision gets made. It is moving away from the individual developer and toward the platform vendor.

Copilot becomes a model picker

The clearest change is in GitHub Copilot itself. One hands-on comparison describes it as a multi-model layer inside the IDE. Its model picker spans Claude Opus 5 and Sonnet 5, GPT-5.6 Sol, Gemini 3.1 Pro and Microsoft's own models, so users choose an engine per task.1 In that framing, the old "Claude vs Copilot" question partly dissolves, because Copilot users can run Claude without leaving Copilot.1

Microsoft 365 Copilot, a separate product from GitHub's coding assistant, has gone further toward automation. One consultancy reports that Claude has been enabled by default there since January 2026, alongside GPT-5, Gemini 2.5 Pro and Microsoft's small Phi-4 model.2 Rather than relying only on a user-facing picker, the system reportedly routes each request based on task complexity, latency and cost. A routine email summary might land on Phi-4, while heavier document analysis might go to Claude.2 The same report frames this as the end of single-model enterprise AI. It also flags a new governance burden: tenants now have to manage several AI vendors, some running on different cloud infrastructure.2

The sources do not agree on model names. The model list for the Microsoft 365 roster refers to "Claude 3.5 Opus," while the GitHub Copilot comparison cites Opus 5 and Sonnet 5.12 The two may describe different products at different points in time. Either way, the specific versions should be treated as fluid.

The Claude Code cancellation

The sharper development is reported by AI Business Review. Citing The Verge and TechCrunch, it says Microsoft has discontinued enterprise licences for Claude Code that customers obtained through Azure. Affected organisations were reportedly told support ends within 90 days, leaving them to move to GitHub Copilot or contract with Anthropic directly.3 The outlet calls this one of the most direct examples yet of a cloud vendor blocking a competitor's AI tool to consolidate its position.3

This report rests on secondhand attribution, and the available details are thin. Still, it fits the broader pattern. Microsoft is happy to offer Anthropic's models inside Copilot, but appears less eager to resell Anthropic's own agent as a separate product.

Why developers were juggling tools in the first place

The fragmentation that platforms are now absorbing is well documented. JetBrains' January 2026 AI Pulse survey found no dominant assistant at work. GitHub Copilot led with only 29%, and ChatGPT, Cursor and Claude Code were close behind.4 An industry roundup cited in the same analysis found that 73% of senior engineers regularly use at least two AI coding tools. A typical split was Copilot for inline suggestions, Claude for refactors and ChatGPT for review.4 The costs of that setup include context-switching overhead and lock-in risk, made worse by a benchmark leaderboard that keeps reshuffling.4

A multi-model Copilot pitches itself as the fix: keep the model variety, drop the tool sprawl.

Agent versus shell

The sources suggest that model access and product experience are now separate things. The comparison in source 1 still distinguishes Claude Code as "the autonomous agent you delegate to" from Copilot as an IDE-embedded assistant, even when Copilot is running Claude underneath.1 Picking Claude from a dropdown does not reproduce an agentic workflow built around that model. For teams that relied on Claude Code's delegation style, the reported licence cancellation is therefore more than a billing change.3

This points to a gap in how "code review" and refactoring are bundled into these platforms. A routing engine can choose a model, but it does not decide how much autonomy the tool gets or how its output is checked.

Our reading

The multi-tool stack is not going away. It is being repackaged as a vendor feature. Model choice is becoming commoditized inside Copilot, while competition shifts to the surrounding agent experience, governance and distribution.12 Microsoft's reported move against Claude Code shows that distribution can be used as leverage.3

For engineering leaders, the practical question has changed. It used to be "which model is best?" Now it is closer to "who controls the routing, and what do we lose if they change the menu?" Given how quickly rankings shift, keeping a direct relationship with at least one model vendor looks like a sensible hedge rather than redundancy.4