Microsoft CEO Satya Nadella has sharply criticized his company's own AI assistant in internal communications, telling engineering managers that Copilot's integrations with Gmail and Outlook "for the most part don't really work" and are "not smart." The comments, reported by The Information on December 28, 2025, reveal a CEO who has essentially demoted himself to chief product manager in an attempt to fix a product that has become central to Microsoft's growth narrative but stubbornly resistant to adoption.
The CEO in the trenches
Nadella's intervention goes well beyond the occasional executive pep talk. He now sits in a Microsoft Teams channel with roughly 100 of the company's top engineers, posting critiques when he believes products are falling short. He runs weekly hour-long meetings to monitor progress. He sends bug reports directly to product teams. One engineer on the email thread where Nadella delivered his criticism noted that Google's Gemini chatbot had pulled ahead in tasks like summarizing photo folders in Google Drive, a capability that should be table stakes for an AI assistant in 2025.
This hands-on approach follows an October 2025 reorganization in which Nadella appointed Judson Althoff as CEO of Microsoft's commercial business, explicitly to free himself for "laser focused" technical work. At the time, the move looked like a CEO positioning himself for the next phase of AI competition. Three months later, it looks more like triage.
Nadella has been personally calling potential hires and approving compensation packages competitive with OpenAI and Google DeepMind. Microsoft poached more than 20 employees from DeepMind in recent months, including Amar Subramanya, formerly head of engineering for Google's Gemini chatbot. The talent war is expensive, but Nadella is treating it as existential.
The numbers Microsoft won't discuss
Microsoft's official line is that Copilot adoption is growing and that "aggregate sales quotas for AI products have not been lowered." The company disputed parts of The Information's reporting. But the data leaking out from various sources paints a different picture.
According to a report citing internal sales materials, Microsoft had approximately 8 million active licensed users of Microsoft 365 Copilot as of August 2025. That's a 1.81% conversion rate against 440 million Microsoft 365 subscribers. Two years after enterprise launch, two percent adoption is, to put it charitably, not what anyone expected from the centerpiece of Microsoft's AI strategy.
The consumer numbers aren't much better. The Newcomer newsletter reported that Copilot's weekly active users have been stuck around 20 million for roughly a year, while ChatGPT approaches 900 million. Microsoft CFO Amy Hood reportedly showed executives a slide depicting Copilot's user growth as essentially a flat line.
Meanwhile, The Information reported that Microsoft has reduced sales growth targets for AI products in some divisions by as much as 50%, after fewer than 20% of salespeople in certain units hit their original goals. Microsoft denies lowering "aggregate" quotas but hasn't addressed the specifics.
Why adoption is stalling
The integrations Nadella criticized aren't minor features. Connecting an AI assistant to email is arguably the core use case for enterprise productivity. If Copilot can't reliably summarize your inbox or draft responses based on context from your calendar, what exactly is the $30 per user per month paying for?
User complaints have been consistent. A corporate trainer who spent more than 100 hours testing Copilot for enterprise deployment described the experience as "performative," requiring constant use of a tool that often slowed her down. The AI would draft emails she had to extensively rewrite, removing what she called Copilot's signature tics: passive voice, bullet lists, upbeat platitudes. Her manager would then return the emails rewritten by Copilot, re-adding everything she'd removed.
A financial analyst told Quartz that Copilot "lies sometimes, withholds information," and "doesn't do the greatest job helping me with things I hate about Excel." For a tool positioned as an always-available assistant for knowledge workers, inconsistency is fatal.
Enterprise customers are hitting similar walls. The Carlyle Group reportedly piloted Copilot Studio for tasks like meeting summaries and financial models, then cut spending after the software struggled to reliably pull data from other applications.
The competitive picture
Google's Gemini now has roughly 400 million monthly users, according to internal estimates revealed in court filings. The gap between Gemini and Copilot has been narrowing, with some analyses putting Gemini less than a percentage point behind Microsoft's 14% market share for AI assistants. ChatGPT remains dominant at around 61%.
What's galling for Microsoft is that this was supposed to be their race to lose. The company invested $13 billion in OpenAI, integrated GPT models before anyone else, and slapped Copilot on everything from Windows to LG televisions. First-mover advantage has evaporated. The technology works better for Google, and the standalone experience works better for OpenAI. Microsoft is stuck in the middle with a product baked into everything but beloved by almost no one.
Nadella himself acknowledged the competitive gap in the email to engineers, specifically calling out Gemini's Drive integration as superior to Microsoft's own efforts. That kind of candor from a CEO is rare. It's also a sign of how worried he is.
What happens next
Microsoft has until its next earnings call in late January to show progress. The company's stock trades at roughly 35 times forward earnings, a valuation that assumes AI will drive meaningful revenue growth. If Copilot continues to stall, investors will start asking harder questions about the $20 billion per quarter Microsoft is spending on AI infrastructure.
Nadella is reportedly reminding employees of Microsoft's past strategic blunders: losing internet search to Google, squandering its early lead in smartphones and tablets. The message is clear. This is the moment.
But fixing a product that doesn't reliably work isn't a matter of executive willpower. It requires engineering breakthroughs on integration quality, reliability, and context understanding that Microsoft hasn't yet demonstrated. Nadella can hold as many weekly meetings as he wants. The question is whether the product can actually get smarter.
Gartner predicted in June 2025 that more than 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs and unclear business value. Microsoft's Copilot may be the highest-profile test case for whether enterprise AI can justify its price tag, or whether it remains an expensive solution searching for a problem.
