Microsoft rolled out its Copilot AI assistant to its own sales force the way it had introduced other new tools. Employees experimented with the technology, then stopped using it.
"They didn't see it as useful in the flow of their work," Katy George, corporate vice president, AI and work transformation at Microsoft, said during Newsweek's October 21 webinar, "Overcoming Barriers to AI Transformation in Legacy Industries."
The company responded by sending engineers to shadow sales representatives and understand how they actually did their jobs. The collaboration led to more specialized tools, including Sales Chat, a set of prompts built on Copilot that helps account executives prepare for meetings, and agents that simplify processes and provide coaching.
The engineers worked directly with "the people doing the work, the ones who have the tacit knowledge of what's important," George said.
...
Joining George on the panel were Matt Sanchez, president and chief operating officer at Yahoo; Durga Malladi, executive vice president and general manager of technology planning, edge solutions and data center at Qualcomm; and Rodrigo Madanes, global next frontier technology and AI officer at EY. Gabriel Snyder, Newsweek's executive editor, enterprise, moderated the webinar, which was presented by Cognizant.
At Yahoo, Sanchez encountered a different obstacle. Teams were experimenting with AI on side projects, but the company's existing development tools and systems weren't ready for agents to work on its core software.
"It became a kind of go-slow-to-go-fast moment for the company," Sanchez said.
Yahoo ran pilots to test whether spending four to six weeks retooling its code and development tools would allow engineers to move faster afterward. The results persuaded business leaders to expand the effort across the company, one engineering team at a time.
Such preparation can extend beyond software, Malladi said.
"There's a lot of groundwork that needs to be done before you can start just throwing AI at every single problem out there," he said.
At the beginning of the session, Snyder cited research published by Cognizant illustrating how many companies have struggled to translate their AI investments into measurable improvements. In a survey of 1,200 senior business leaders, only 32 percent reported tangible gains in business productivity, while about one-quarter said their organizations had paused or discontinued AI deployments.
The study also found that investment priorities mattered. Among companies with less mature technology infrastructure, those pursuing broader, nontechnical AI investments were 60 percent more likely to pause or abandon a deployment than those concentrating on technical foundations such as computing capacity and data readiness.
George argued that companies should let business goals determine which systems and data deserve investment, rather than undertake extensive infrastructure projects without a clear purpose.
The problem of somewhat aimless AI investment is especially visible in software engineering.
"If you only speed up one piece of it using agents, you just shift the bottleneck forward or backward, and you don't actually get better results," George said.
Faster coding can leave engineers spending more time reviewing AI-generated code. At Microsoft, George said, teams that have made more substantial gains put greater emphasis on clear specifications, reusable expertise, defined boundaries for agents and new approaches to testing and validation.
Sanchez described a related development at Yahoo. Engineers are spending more time on software architecture, defining requirements and evaluating whether the output meets expectations. Faster code generation also makes deciding precisely what to build more important.
"Your customer can only absorb so much change in your product. And so you have to be really thoughtful about where it is that you're actually going to deploy those resources," he said.
George also pointed to Microsoft's use of AI to identify vulnerabilities in cloud data centers and help internal auditors assess risks proactively. Those capabilities, she argued, can create greater strategic value than one-time productivity improvements.
...
At Qualcomm, Malladi described applications in finance, sales, human resources and product management, including the complicated planning involved in coordinating multiple semiconductor programs and the intellectual property they rely on. Changes to one project can affect others, requiring teams to revisit their plans. AI can help employees compare alternatives and weigh the consequences.
Deciding where to run AI workloads raises another set of questions. Malladi said technology leaders must account for employees working without reliable internet connectivity, the processing requirements of different models, the cost of running them and restrictions on sensitive data.
Those factors may favor cloud computing for one task and on-device processing or company servers for another. Malladi described an approach in which technology leaders establish the relevant policies while the systems operate within those boundaries.
"It's the agents that eventually get to decide what runs where at any given point in time," he said.
The growing autonomy of those systems also makes oversight more consequential. Malladi emphasized guardrails and traceability, particularly the ability to investigate how an automated decision was reached, identify where an error occurred and determine what needs to be corrected.
Madanes described how a successful pilot in one country can encounter unexpected regulatory obstacles when a company expands it internationally. Data sovereignty, privacy and retention requirements can complicate the wider rollout.
"It becomes a showstopper or something really hard that they hadn't foreseen," Madanes said.
What appeared to be a straightforward expansion can force teams to revisit their architecture and reengineer work they believed was ready. The problem is especially difficult when compliance specialists haven't been involved early enough to identify the restrictions.
Madanes proposed two complementary levels of leadership. Strategic leadership should include business and finance executives, while an operating group brings together engineering, product, human resources, legal and compliance to address obstacles before they stall a project.
"You need to do both of them. Prove every dollar and move fast," Madanes said.
Yahoo followed a similar leadership model. Sanchez said he partnered with the company's chief financial officer to lead its efforts, supported by a cross-functional group of engineering and product leaders exploring new approaches and another dedicated to cultural and employee concerns.
Sanchez said the company also had to account for workers whose professional identities were tied to long-established routines.
"This can be scary or challenging or really emotionally fraught. You have to be very empathetic to that," he said.
Yahoo provided coaching, resources and colleagues who could help employees develop confidence with the tools. Some initially reluctant workers became strong contributors. The company also established a no-fault opt-out path for people who felt the changes weren't right for them.
Sanchez urged executives to "find your pirates," employees already experimenting with AI, and empower them to help others adopt new approaches.
Microsoft has taken a structured approach to that kind of peer learning. George described weekly meetings of roughly 15 employees in the same role, led by a colleague experienced with AI. Participants discuss how they applied the tools, what worked and what they plan to try next. Bringing employees directly into redesigning their responsibilities, she said, can help overcome anxiety and build confidence.
As agents take on more execution, George expects employees to need stronger business judgment, domain expertise and an understanding of how different parts of an organization fit together. Workers early in their careers, she said, will need opportunities to develop those skills sooner.
"We can't future-proof any one specific job, but we can help you future-proof your career," George said.
We hope you'll join us for two upcoming Newsweek webinars.
On Wednesday, November 4, at 11 a.m. ET, Dr. Ranjit Tinaikar will sit down with Firdaus Bhathena, executive vice president and chief technology and transformation officer at S&P Global, for "What Enterprise AI Looks Like When It Works." They'll discuss how companies can put trusted data to work, redesign workflows and turn AI investments into measurable business results.
Register for the webinar, for free, today.
And then on Tuesday, November 10, at 2 p.m. ET, Newsweek's Gabriel Snyder will moderate "The AI Election: How Artificial Intelligence Shaped the 2026 Midterms," featuring Nathaniel Persily of Stanford Law School, Brent Buchanan of Cygnal, Nick Penniman of Issue One and Andrew Mamo of The Bench.
Drawing on Newsweek's AI Policy Scorecard, the panel will examine how issues such as data centers, jobs and regulation figured in the elections and what the results reveal about the politics of AI.