Artificial intelligence has turned software development at Microsoft upside down, and the changes reflect how technology is accelerating a wholesale reconsideration of the way the entire company functions.

On Thursday, Microsoft is releasing a playbook detailing how it is navigating the AI upheaval. It includes lessons that employees like Julia Gao, a 23-year-old software engineer, have had to learn quickly as their roles have radically changed in a matter of months.

"When I first joined Microsoft last year, we were using AI, but it wasn't to the extent that we are now. I was still writing code by myself," Gao told me last week over lunch at Microsoft headquarters in Redmond, Wash.

"After a few months, AI, it became rapid and there were a lot of changes, and now, I don't really code anymore," said Gao, who works on the Azure Chaos Studio, a managed service for chaos engineering and resilience testing of the Azure cloud computing platform.

The California Institute of Technology graduate's role was swept into a companywide transformation as AI lent urgency to change in almost every aspect of the business, from the definition of roles to the rethinking of longstanding workflows.

As AI labs pushed the envelope with ever more powerful frontier models, Microsoft was developing a parallel concept of the "frontier company," built on principles that reflected AI.

Microsoft aggregated those ideas in a "Frontier Playbook" that it is making public Thursday with the goal of helping other organizations evolve with AI. The company emphasizes lessons from its experience, including the need to start with the desired business outcome, not a technology.

During a series of interviews at Microsoft, I delved deeper into some of those lessons, including how the company is rethinking traditional organizational structures and roles and how it is using AI to design end-to-end workflows for a complex enterprise with about 223,000 people.

These changes were notably evident in the software engineering organizations, although they are taking root in other areas including sales, finance and human resources. I spoke with early career professionals such as Gao as well as senior leaders.

For starters, AI must support a business goal, according to Kathleen Hogan, executive vice president and chief strategy and transformation officer. Instead of measuring how many salespeople in a company use an AI Copilot, leadership should focus on identifying the agents and workflows that increase revenue per seller.

"One of the best CEOs that I met with said, 'I'm not implementing AI.' I thought, 'Oh, my God, you're the first CEO I've met that's telling me that,'" Hogan recalled. He quickly clarified that he wasn't avoiding AI. Rather, he was implementing his business strategy by using AI, putting the emphasis where it truly belonged.

"That is the shift everybody just has to implement," Hogan said. Microsoft's most significant error, she said, was leading with technology rather than focusing on people and process. "A bad process with AI is still a bad process," she said.

The overhaul of Microsoft's inner workings is integral to the performance of the company overall.

Microsoft isn't at the bleeding edge of frontier model development, but the business is showing strength and leverages OpenAI's frontier intellectual property, according to a July 29 report by Goldman Sachs analyst Gabriela Borges. Its fourth quarter was a "meaningful step in reversing the stock's multiquarter period of underperformance," she wrote. Borges, with a buy rating on the company, cited Azure cloud acceleration and improving AI unit economics. She also cited "increasing evidence of Copilot monetization," referring to the AI assistant launched in 2023.

A flatter and faster organization

"The entire profession of software engineering has completely changed in the last 12 months," Charles Lamanna, executive vice president of Copilot, agents and platform, told me. His product-building team has 20,000 people across software engineering. "The era of being a coder is kind of over," he said.

Lamanna says he now runs a flatter organization with an emphasis on player-coaches. The number of layers has been reduced over the past year or so to about five, down from 10 or 11. Managers who once oversaw seven people now function as player-coaches to groups of 15. Some teams work in person, collaborating face-to-face in a single room to make interaction easier and more spontaneous.

The new structure operates at a faster pace. A nine-person core team shipped the initial release of Copilot Cowork in 35 days. The AI assistant for complex tasks operates on Microsoft 365 apps after users give it a goal. In comparison, the low-code platform Copilot Studio took several years and hundreds of people to reach a similar scale, according to Lamanna.

For generations, companies have been organized around vertical roles and clearly defined departments and functions. AI is enabling a more ad hoc structure based on workflows. Increasingly, those workflows have an end-to-end character that moves through the more rigid walls of the traditional structure.

Introducing AI into a company's workflow poses a challenge for people, given that the human capacity for work obviously can't scale as fast as technology. To address that bottleneck, the company created an immersive training program called Camp AIR, in which teams put aside their regular work for a time and learn by using AI tools to create something together, according to Tim Bozarth, corporate vice president for engineering, core AI.

"You have to actually be willing to detach your people that are currently driving your business...and drop them into something else where they pick up the tools. When you're in that learning period, the thing you need to be focused on is learning," Bozarth said. Setting up classes for a few hours doesn't lead to a change in behavior, he found.

Software engineer Sara Young, a veteran of the program, says she has learned to break the review of code into smaller, more modular chunks, since AI's coding output is too massive to review all at once, which was the more traditional method.

"The team needs to build resiliency for how they adopt these new capabilities, and that is what Camp AIR does," said AI design strategist Andy Klein, who went through the program early and helped shape it.

Rethinking workflows

These new workflows have had an impact on sales, according to Lorraine Bardeen, corporate vice president for AI strategy and transformation.

When Microsoft co-invests with a customer on a transformation initiative, the approval process spans sales, finance, legal and procurement. It traditionally took 28 days to complete. The use of AI allowed the company to design a new workflow that allows a turnaround in less than six days.

The friction was almost always tied to a person who needed to click "approve." If that person went on vacation or had other priorities, the process bogged down. By using AI to instantly verify policy adherence whenever possible, it became easier to retire the old bureaucracy, according to Bardeen.

"In my 24 years, this is the most significant change in how we work, what we can accomplish individually and collectively by far," she said.