THE VERDICT
● Microsoft's bet: own the interface, rent any model. Copilot now ships OpenAI, Anthropic and xAI.
● Anthropic's bet: own the model, reach into any system. Claude is default across Salesforce.
● For a buyer: one gives you model choice and no depth. The other gives you depth and one vendor.
What each company did this week
12 September. Microsoft added xAI models to Copilot across Word, Excel, PowerPoint and GitHub Copilot, alongside the OpenAI and Anthropic models already there. Copilot stopped being a product with a model inside it and became a marketplace.
15 September. Salesforce in Claude entered open beta, the first shipping product of a partnership under which Claude is the default reasoning model across Agentforce, the Atlas Reasoning Engine and Slack.
THESE ARE NOT COMPETING PRODUCTS. THEY ARE COMPETING THEORIES
Microsoft is assuming models become interchangeable, so the durable asset is the software people open every morning.
Anthropic is assuming models do not become interchangeable, so the durable asset is being the one that reasons well enough to be worth integrating deeply. Only one of those can be true for long.
What each means if you are buying
| Marketplace model | Default model |
|---|
| Model choice | Yours | Made for you |
| Depth of integration | Shallow by necessity | Deep |
| Switching cost later | Low. Change the dropdown | High. 37 skills are built for one model |
| Data handling terms | Differ per model vendor | One set of terms |
| Who is accountable | Split between platform and model vendor | Clearer |
The switching cost row is the one people underweight. A model selector makes changing engine feel like changing a font. It is a change of data processor, with different terms — and that cuts both ways.
The evidence so far is mixed
- For Microsoft: Copilot is inside applications people already have open, and distribution has beaten capability repeatedly.
- For Anthropic: GitLab, Siemens and Legora deployed to roughly 7,000 sellers before open beta. Depth got adopted faster than choice.
- Against Microsoft: no model vendor owns the interface, which is worse for OpenAI — its largest investor now distributes its competitors.
- Against Anthropic: being the default inside someone else's product means the partner controls the surface.
Which one
| If you... | Lean |
|---|
| Expect to change models within two years | Marketplace. Keep the switching cost low |
| Need workflows that actually do the job | Deep integration. Shallow does not deliver |
| Have a procurement team asking about data | One vendor is easier to answer for than three |
| Are not sure | Neither bet has resolved. Avoid anything with a long contract |
FAQ
Which approach is winning?
Too early. Both landed within four days of each other and neither has produced a year of evidence.
Does Microsoft still favour OpenAI?
Microsoft remains OpenAI's largest investor and now distributes its competitors inside the same interface. Those facts sit together uncomfortably and that is the point.
What should I avoid?
A long contract on either theory before it has resolved. Both are a few weeks old.