The hard part of AI isn’t the code, it’s the company itself.

Lorraine BardeenHe, who leads Microsoft’s commercial business AI strategy, said this recently at the Stanford Economic Policy Institute. meeting Global companies are entering a complete restructuring cycle.

“What they are working on now is rebuilding the entire company,” she said. However, this shift is not about layoffs, but rather about expanding customer reach and increasing throughput by combining human employees with agents. “Are they restructuring the company to lay off people? No, they’re not.”

Bardeen said companies are settling on three modes. Teams incorporate agents into their daily workflows. The entire function then moves to autonomous operation.

One example is already in operation within Microsoft. Sales agents respond to small business leads without human intervention. “They never hear a word from humans,” she said. That agent creates outreach emails, logs activity, and records revenue.

What is difficult is restructuring the company structure.

Acuity Brands CEO also participated in panel discussion at Stanford conference Neil Ash said The biggest hurdle is not AI performance, but organizational capabilities. “It’s not the technology that’s hard. It’s changing parts of the company that are hard,” he said. Acuity manufactures lighting and architectural systems used in commercial spaces. Ash said one of the company’s major engineering modernization programs used to take up to 10 years. By using agent tools, the company resolved the issue within 30 days.

With this breakthrough comes new challenges. To take full advantage of the benefits, the rest of the workflow needed to be redesigned. Ash said this dynamic widens the performance gap across the industry. “There will be haves and have-nots,” he said. Companies that can effectively redesign operations at scale while managing risk and resistance to change are well-positioned to capture productivity gains and sustain AI-driven growth.

Ashe said AI will also reshape early career jobs. Acuity continues to hire young talent because judgment becomes more important when information becomes abundant. Rather, it puts pressure on managers to rethink what their initial role should be in an AI-enhanced environment.

Data bottlenecks slow change

Snowflake AI Vice President of Engineering and Research Dwarak Rajagopal said Fragmented data remains one of the biggest barriers to scaling AI. “Data exists everywhere within the enterprise,” he said. Pilots are often successful within a single domain, but agents fail when information spans systems with inconsistent governance rules.

Snowflake has built an internal enterprise agent that answers questions about enterprise data in natural language. Mr. Rajagopal said employees are asking “almost 12,500 questions a week,” saving “about 15 minutes per question.” However, when employee behavior changes, it becomes difficult to measure productivity. Employees now ask follow-up questions that they previously avoided due to too much manual work.

AI models will continue to improve, but the economic rewards will depend on how many companies can reorganize fast enough to use them. Reassure your audience that demonstrating clear ROI is important, and that strategic repositioning and proof of value are essential for sustained AI investment and success.

CFOs are also tightening their expectations as earnings remain uneven. new PYMNTS Intelligence The survey found that only 26.7% of CFOs plan to increase their generative AI budget over the next 12 months, down from 53.3% in the same period last year.

This backlash signals a shift from experimentation to disciplined, results-oriented spending. The gulf is clear. Half of companies reporting very positive revenues plan to expand their budgets, while only 16.7% of companies with modest revenues have similar plans. This number shows that future AI investments are directly tied to proof of financial or operational value.

For all of our coverage of PYMNTS AI, subscribe to our daily subscription AI Newsletter.