Lenovo is building an AI assistant that ‘can act on your behalf’

While most attention in the AI race is focused on model builders and cloud platforms, Lenovo sits closer to millions of users than most companies. As the world’s top PC maker by volume, Lenovo ships tens of millions of devices every year. What it decides to ship, bundle, and integrate can directly shape how AI shows up in many everyday lives.

That’s what made Lenovo’s announcement today at CES notable. At a flashy event on Tuesday at The Sphere in Las Vegas, it introduced Qira, a system-level, cross-device AI assistant designed to live across Lenovo laptops and Motorola phones. It’s Lenovo’s most ambitious AI effort to date and a rare look at how a hardware giant with global reach is thinking about integrating AI more deeply.

Jeff Snow, Lenovo’s head of AI product, told me how Qira came together, why the company is deliberately avoiding a single exclusive AI partnership, and what he learned from earlier experiments like Moto AI and Microsoft’s Recall debacle.

Qira emerged from a quiet but meaningful internal reorganization less than a year ago, according to Snow. Lenovo pulled AI teams out of individual hardware units such as PCs, tablets, and phones and centralized them into a new software-focused group that works across the entire company.

For a company long optimized around hardware SKUs and supply chains, the move signaled a shift toward putting AI more front and center. “We wanted a built-in cross-device intelligence that works with you throughout the day, learns from your interactions, and can act on your behalf,” Snow said. He mentioned using Qira’s on-device model during his flight to CES to help him workshop how to talk about the news in meetings based on the notes and documents on his PC.

Qira is not built around a single flagship AI model. Instead, it’s modular. Under the hood, it mixes local, on-device models with cloud-based models, anchored by Microsoft and OpenAI infrastructure accessed through Azure. Stability AI’s diffusion model is also integrated, along with tie-ins to app-specific partners like Notion and Perplexity.

“We didn’t want to hard-code ourselves to one model,” Snow said. “This space is moving too fast. Different tasks need different tradeoffs around performance, quality, and cost.”

That stance runs counter to the push from major AI labs, many of which would happily become the exclusive intelligence layer for a company with Lenovo’s reach. Lenovo’s view is that optionality matters more, especially given its control over one of the largest consumer computing distribution channels in the world.

Snow previously worked on Moto AI, Motorola’s assistant, which he said saw high initial engagement. More than half of Motorola users tried it, but retention wasn’t good. He said that too much of the experience felt like prompt-based chat features people could already get elsewhere.

“That pushed us away from competing with chatbots,” Snow said. “Qira is about things chatbots can’t do, like continuity, context, and acting directly on your device.”

Lenovo also paid close attention to the backlash around Microsoft’s Recall feature. Snow said Qira is designed from the outset with opt-in memory, persistent indicators, and clear user controls. Context ingestion is optional. Recording is visible. Nothing is silently collected.

Cost pressures loom over all of this. Memory prices are rising as AI demand strains supply chains, and analysts expect PC prices to follow. Qira does not raise baseline system requirements for PCs, Snow said, but it performs best on higher-end machines with more RAM. Lenovo is working to bring local models down to smaller memory footprints, like 16 gigabytes of RAM, without watering down the experience.

Strategically, Lenovo sees Qira as both a retention play and a hedge against hardware commoditization. In the short term, it hopes that tighter integration between laptops and phones will nudge customers to stay within the Lenovo ecosystem. Over the longer term, Snow framed Qira as a way to differentiate Lenovo devices when specs alone are no longer enough.