As CIOs pin down plans to make 2026 the year their companies capitalize on AI, they find themselves at an inflection point. After years of experimentation, they must now move beyond AI pilots to operationalizing the technology.
For some CIOs, that shift entails a reframing of priorities. Rather than launching new AI initiatives or adding new layers of technology, their focus is turning to integrating what they have already purchased and built — putting execution, not experimentation, at the top of the AI agenda.
SAS CIO Jay Upchurch has described this moment as the rise of the “Chief Integration Officer,” when CIOs need to do the hard work of weaving their AI technologies into core enterprise systems — selecting a foundational AI system, aligning AI agents from vendors with technologies built internally and working out the data flows between SaaS platforms. Demonstrating ROI, he has said, will ride less on acquiring new AI capabilities and more on proving that the existing ones can deliver value at scale.
On-device AI adds new layer to the stack
The integration-first mindset, however, is being tested by a new wave of AI technologies that promise to deliver value by adding another layer to the technology stack. One recent example is Lenovo‘s launch of its embedded “Personal Ambient Intelligence” AI assistant, Qira.
For CIOs, the issue isn’t whether on-device or “ambient” AI has potential, but whether adding new system-level AI capabilities helps solve the problem of AI integration — or makes it worse.
On-device AI represents an opportunity to shift how enterprise PCs are used — from passive endpoints and portals to the cloud to more active agents. Embedded or on-device AI technologies make it possible to combine system-layer AI with an organization’s use of browser-based large language models, reducing dependence on the cloud while potentially increasing AI use cases.
Charlie Dai, vice president and principal analyst at Forrester, explained that the shift is being enabled by the use of neural processing units (NPUs), which support embedded AI and machine learning operations directly on the device. “The collective goal is to redefine the device from a passive tool to an active, context-aware AI agent,” Dai said.
He noted that Lenovo is not alone in that push, but part of an “industry-wide race.” PC makers including Dell, HP and Samsung (with its Galaxy AI) have launched AI-enabled PC lines with dedicated NPUs. Platform providers such as Apple (Apple Intelligence) and Microsoft (Copilot) are also embedding on-device AI capabilities directly into their operating systems and productivity tools.
Tradeoffs of embedded AI
Demand for AI-enabled devices is also being shaped by a practical event, Dai said — the ending of Windows 10 support. With support ending in October 2025, many enterprises are planning PC refresh cycles. Dai said CIOs see embedded AI as a way to future-proof their workforces in a secure and efficient manner.
CIOs can reap a number of benefits from deploying devices with embedded AI, including improved data privacy via local processing, lowered cloud costs and reliable offline functionality, Dai explained. In Lenovo’s case, its on-device AI assistant is a hybrid model that combines cloud-based AI with on-device processing through the NPU.
Still, those benefits come with tradeoffs. Dai cautioned that CIOs should weigh the risks of technical debt, vendor lock-in, and a still-nascent ecosystem of “truly transformative AI-native apps.” CIOs also have to consider the complexity of managing and governing numerous AI technologies across the enterprise.
For Brian Greenberg, CIO at RHR International, these concerns are central. In considering using an embedded AI like Lenovo’s Qira, Greenberg said he would want to know more about the AI’s security and data governance capabilities, how it would interoperate with his organization’s existing Google Gemini investment and Qira’s performance versus a browser-based AI technology.
“Your workspace is a browser, for the most part, for so many companies. To start to depend on hardware for anything other than acceleration seems interesting, but I don’t see a need yet,” Greenberg said.
Others urged a similarly pragmatic approach. Runar Bjorhovde, a research analyst at Omdia’s Canalys, said that in assessing any new AI tool, CIOs will want to examine whether it meets compliance standards, is secure and scalable.
“For many companies, it’s about optimizing the infrastructure you have and what people actually need,” Bjorhovde said.
