Baidu CEO Robin Li urged companies to move faster from AI pilots to “agent” systems embedded in workflows, as firms weigh token costs and job impacts.
Companies are only at the beginning of adopting artificial intelligence in meaningful commercial ways, Baidu founder and CEO Robin Li said Thursday in Beijing, urging businesses to move faster in deploying AI “agents” even as executives worry about rising operating costs and the impact on jobs.
Li delivered his remarks as enterprises—particularly in finance and manufacturing—are reassessing how work gets done. Firms in these sectors are redesigning processes to integrate AI into everything from back-office operations to production planning, and their technology spending is rising as they experiment with large language models and agent-like systems that can complete tasks with limited human direction.
The Baidu chief’s comments came during the Baidu World 2026 conference, where the company showcased new computing infrastructure intended to support large-scale AI workloads. On display was “Tianchi,” a high-performance AI computing system built around what Baidu described as a 1,024-card “super node.” Systems of this kind are designed to run and train advanced AI models and to serve heavy real-world usage—both of which require vast parallel processing capacity.
Li argued that broader commercial uptake will depend on companies moving beyond pilots and chat-style demonstrations to put AI systems directly into workflows. In his view, AI agents—software that can take a goal, plan steps, use tools and data sources, and execute tasks—represent the next stage of adoption. Rather than being used only for Q&A or drafting text, agentic systems are positioned to automate multistep business processes and coordinate work across internal systems.
But the push toward agents is arriving alongside practical anxieties. One is the “token budget,” a shorthand for the consumption-based costs of using modern AI models, which often scale with the length and volume of inputs and outputs. As enterprises ramp up usage, those costs can expand quickly, forcing leaders to weigh productivity gains against monthly bills for model access and computing.
Another concern is the workforce disruption that could follow if AI tools reduce the need for certain roles or reshape job requirements. Li’s call for expanded adoption underscores a growing tension: executives see competitive pressure to deploy AI, while employees and policymakers scrutinize how the technology will affect hiring, training, and job stability.
Still, Li maintained that the current wave of enterprise experimentation signals only the early innings. With companies already reallocating budgets and revisiting operating models, he suggested that commercial AI—particularly agent-driven applications—has significant room to grow.
Why it matters
Li’s message reflects a broader shift from AI as a novelty to AI as infrastructure, with major implications for how industries like finance and manufacturing allocate capital and redesign work. If token-driven costs continue to climb, businesses will need clearer ROI metrics and better efficiency strategies. The next phase to watch is whether AI agents move from controlled trials into core operations—and how quickly companies can manage the resulting cost and labor impacts.