🔍 Read the full analysis: What You May Need To Change When Switching From Claude on ThorstenMeyerAI.com

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get the latest gadgets delivered free — and shop member deals

• Fast, free delivery on millions of items
• Access to Prime Big Deal Days deals on October 6–7
• Prime Video, Amazon Music and more included

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Meta and Microsoft are reducing some employees’ internal use of Anthropic’s Claude tools and steering them toward alternatives, according to a report by The Information on Oct. 5. The reported changes matter beyond those companies: they show how access to existing tools, spending limits and the engineering cost of changing systems can shape enterprise AI choices, without establishing that Claude performed worse.

The Information reported that Meta cut the number of employees using Claude Code from about 60,000 to about 30,000 compared with earlier this year. It said Meta was encouraging staff to use its own coding tools, MetaCode, which reportedly had more than 30,000 internal users, and Muse Code, with more than 6,000. These figures and the reasons for the change are attributed to the report; they are not evidence that all Meta teams have stopped using Claude.

Microsoft had reportedly projected more than $1 billion a year in internal spending on Anthropic technology, including Claude Code and Claude models in Copilot. The report says the company cut that projection by more than a third and redirected employees toward GitHub Copilot and OpenAI models. It also says Microsoft continues to use Anthropic models in customer-facing Copilot features, and that customer spending on Claude through Microsoft platforms is growing.

The source material identifies rising token costs and tighter spending controls, alongside the availability of in-house or affiliated alternatives, as reported drivers. It does not report either company saying Claude was inferior. A separate reported detail says some Microsoft team budgets fell from around $100,000 to around $10,000 per month, but that figure comes from a single account cited in the report and should not be treated as a company-wide policy.

Switching Takes More Than a New Model

For companies considering a change, the sticker price of a model is only one part of the calculation. A new system may require teams to rerun evaluations, adapt prompts and tool definitions, and rebuild integrations. Employees also need time to learn a different tool. A move that lowers token costs can still add expense if it creates more review, rework or errors.

The consequences depend on what a company has already built. Meta and Microsoft reportedly had alternatives available at scale, so they could redirect work rather than start from scratch. Other organizations may lack tested substitutes, representative evaluation tasks or software layers that make models interchangeable. In those cases, switching costs can narrow or erase expected savings; the available reporting does not establish a universal break-even point.

The practical takeaway is not that businesses should leave Claude or commit to it. It is that they should know how their applications behave on more than one model and measure the full cost of accepted work, including human review. A second provider used on a limited but real share of tasks can make future choices easier to assess, though running multiple systems also has costs.

Internal Use Is Not Customer Access

The reported changes concern Meta’s and Microsoft’s own employees. They do not amount to a statement that either company has ended customer access to Claude. Microsoft’s reported continued use of Anthropic models in Copilot features is a separate part of the picture, as is reported customer spending through its platforms.

Both companies also have alternatives tied to their own businesses: Meta develops models and coding tools, while Microsoft operates GitHub Copilot and backs OpenAI. That can make internal adoption decisions different from those of a customer with no comparable tools. A company redirecting its own staff toward products it builds or supports is not, on its own, a verdict on a supplier’s model quality.

The account also points to the operational work behind a model change. Teams may need to check whether the replacement meets task-specific standards, revise software built around the previous model, and account for changes in caching and usage costs. Those considerations are particularly relevant to coding agents, where value depends on connections to editors, repositories and team practices. The report supplies no independent comparison of model performance on Meta’s or Microsoft’s work.

Performance Evidence Is Not Reported

The report, as summarized in the supplied material, does not provide a controlled comparison of Claude against the replacement tools, nor does it establish how much of the reported spending reduction came from lower usage, budget limits, pricing or other changes. It is also unclear how the reported user counts were defined, whether they refer to active users or access, and how consistently the changes were applied across teams.

The reported figures should be treated as estimates attributed to The Information, not audited company-wide totals. The account of Microsoft team budgets is especially limited because it comes from a single report. No details are provided about the quality of the replacement tools on specific workflows or the productivity effects of the changes. Those unknowns mean the figures cannot show whether switching improved overall value.

Measure Costs Before Switching

The next useful evidence would be clearer information from Meta and Microsoft about the scope of the internal changes, their reasons and how the alternatives perform in daily use. Until then, the reported moves show that large buyers are redirecting some work, but not the full outcome of that decision.

For other organizations, the immediate step is to compare models on representative tasks before moving critical work. Teams can record pass rates, accepted results, review time and rework, then compare those measures with token and subscription costs. Keeping prompts, business rules and tool definitions in a company-controlled layer may reduce future migration effort, but it does not remove the need to test integrations and user workflows.

Any switch should be judged against its total operating cost and the risks of weaker results on particular tasks. A limited trial on real work can help a team learn whether savings hold up after evaluation and adaptation. The timeline and outcome of any broader Meta or Microsoft changes remain unclear.

Key Questions

Are Meta and Microsoft ending their use of Claude?

No such company-wide exit is established. The Information reported reductions in some internal use. The same account says Microsoft continues to use Anthropic models in customer-facing Copilot features and that customer spending on Claude through Microsoft platforms is growing.

Did the companies say Claude was performing poorly?

The supplied reporting does not say either company blamed poor model performance. It attributes the changes to factors including cost controls and the availability of alternatives. It does not provide comparative performance results.

Why can switching AI tools be expensive?

A move can require new evaluations, prompt and integration changes, employee training and additional review of model outputs. A cheaper token rate alone does not show whether the replacement costs less overall.

What should a company test before changing models?

Use a set of representative tasks and defined pass criteria. Compare quality, review time, rework and usage costs on the same work, and check that the replacement integrates with existing applications and team processes.

Source: ThorstenMeyerAI.com