Microsoft CEO Satya Nadella has warned that companies could lose control of their institutional knowledge if they depend entirely on proprietary AI providers. He said businesses should retain their prompts, usage metadata, context and memory so they can switch models or eventually train systems of their own.
Speaking on CNN’s Fareed Zakaria GPS, Nadella said a company that gives up this control has effectively “outsourced” its thinking. He argued that such businesses may struggle to remain independent as AI becomes more deeply embedded in their operations.
Companies Should Retain Their AI Metadata
Nadella said businesses should preserve the information generated whenever employees or software agents interact with an AI model. This can include prompts, corrections, workflow data, model responses and details about how a system completed a task.
Keeping that information could allow a company to fine-tune another model, train its own weights or move to an open-weight system later. Without it, the model provider may gain valuable insight into the organisation’s internal processes while the customer remains dependent on the provider’s technology.
The comments expand on Nadella’s earlier discussion of what he calls the “Reverse Information Paradox”. He argues that companies pay for AI twice: first through usage fees and again through the knowledge they provide while using it.
Harnesses Should Remain Separate From Models
Nadella also advised companies not to tie their entire AI operation to a provider’s coding agent or agent framework. Products such as Claude Code and OpenAI Codex combine a model with a harness that manages tools, memory, context and task execution.
He said organisations should keep the harness and context layers separate from the underlying model. This would allow them to use several models for different tasks and replace one provider without rebuilding their entire AI system.
Microsoft has been promoting this architecture through Microsoft Foundry’s model router, which selects between multiple models, and its AI Gateway, which sits between company applications and model providers.
Warning Also Supports Microsoft’s Cloud Strategy
Microsoft is a major investor and commercial partner of both OpenAI and Anthropic, but it is also developing its own models and infrastructure. A multi-model market benefits Azure because companies need cloud services that can manage models from several providers.
Nadella’s warning is not limited to costs or technical dependence. He said an AI laboratory that gains enough understanding of a customer’s workflows could eventually introduce a competing service.
The concern resembles a warning investor Jason Calacanis gave after OpenAI offered AI credits to Y Combinator startups. Calacanis argued that a model provider could study promising applications and later incorporate similar features into its own products, although no evidence was presented that OpenAI intended to do so.
Featured image credits: Wikimedia Commons
For more stories like it, click the +Follow button at the top of this page to follow us.