Lenovo Brings Tianxi AI to Alibaba's Yunqi Conference, Pitches 'Human Plus Agent' as the New Unit of Work

Liu Linlin, head of Lenovo's Tianxi AI strategic ecosystem department, made the case in a speech titled "A New Paradigm for On-Device AI: Landing and Co-Creating Super Agents" at the conference's Qoder forum. The conference was themed "making intelligence useful."

According to Liu, the largest cost of an organization in the AI era is no longer labor but coordination itself. Once individual output rises by an order of magnitude, waiting becomes the only bottleneck, and the minimum production unit becomes one person working with a group of agents, with the person responsible for the goal and the agents responsible for the path. Lenovo said it has tested that arrangement internally: three people took a product from design to release in 30 days without holding a single meeting.

The AI host plays a central part in that model. If a notebook is shut down and its owner has gone home, the agents stop working, which Lenovo describes as the hard bottleneck of the super organization. Its AI host runs around the clock and takes on long-running, compute-heavy tasks so results are ready the next day. The company defines the device internally as an "AI execution team," or the office where a person's agents work.

Liu described five changes taking shape across the industry: models are converging, so the harness matters more and the barrier shifts to engineering capability in context, orchestration, verification and security; the minimum unit is one person with a group of agents, turning headcount thinking into how many agents each person can command; agents will become the first users of software, moving interfaces from graphical screens to MCP, skills and CLIs; terminals are the last mile of the super organization, and device-cloud is a long-term balance rather than a transitional form; and trust infrastructure such as identity, permissions, approvals, audits and rollbacks sets the ceiling on how much authority companies will delegate to agents.

"The era of competing on models is ending, and the era of competing on 'human plus Agent' organizational models is beginning," Liu said. "What separates the field in the next round is not whose model is stronger, but who gets the 'human plus Agent' model working first."

On Tianxi AI 4.3, the upgrade centers on a "Super Mode" that unifies entry points across devices and links their data, with task execution visible on screen. It adds self-developed full-duplex natural voice dialogue that can be interrupted mid-sentence, and debuts the Tianxi AgentCore on-device agent framework, which plans and executes tasks autonomously and generates results, and allows a phone or tablet to remotely control a PC while data stays local. Lenovo said it also worked with Alibaba's Qoder to improve the user experience.

Lenovo said the Tianxi software ecosystem has 280 million monthly active users and 4 billion distributions a year, with 39,000 pieces of AI ecosystem content covering AI PCs, AI phones, AI tablets and AIoT. For developers, its Cangqiong Plan offers one trillion tokens, exposure slots and incentive funds. The company described its own role as commander-in-chief, main entry point and dispatcher, saying developers need not maintain complex device-cloud rules themselves and that a single integration reaches multiple devices.

Lenovo presents the lineup as part of its hybrid AI strategy landing on the device side, built on device-edge-cloud collaboration. It said it will keep investing in on-device AI and work with developers and industry partners to get the "human plus Agent" model running on terminals first.