Jacob Andreou ordered a burger with Copilot Tasks in two months. Nadella just bet on it

Jacob Andreou is the kind of Microsoft executive who doesn’t just talk about AI agents. He was hands-on with software developers to ship Copilot Tasks, a homegrown AI tool, and the proof came fast: he set out to use it to order a McDonald’s cheeseburger to his apartment across from Microsoft’s Silicon Valley campus. By the time he got home, the burger had arrived.

That detail matters because it is not just a cool demo. Microsoft Chief Executive Satya Nadella, who has placed “great trust” in Andreou to guide Copilot, is leaning on that kind of execution to fix a product that has struggled to keep up with competitors. Microsoft fell behind early in the AI race and is counting on Copilot, and on Andreou as its product executive vice president, to pull the company back into the lead while moving quickly without breaking the “trusted business relationships” enterprises rely on.

There is a reason Nadella’s bet feels so urgent. Copilot’s adoption and competitive position have not matched Microsoft’s ambition. Only about 4.5% of Microsoft’s 450 million Microsoft 365 customers pay for Copilot features, and Microsoft’s free consumer version trails far behind OpenAI’s ChatGPT. The market reaction has been just as unforgiving: Microsoft’s shares are down double-digits in the past year as investors worried about AI’s impact on software, Microsoft’s reliance on OpenAI, and the company’s heavy data center spending.

And yet, the counterweight is real. Nadella posted on X that England’s National Health Service rolled out Microsoft 365 Copilot to more than 500,000 staff, signaling that massive, regulated enterprise customers trust Microsoft to deliver AI securely at scale. That kind of enterprise validation is the whole game. Consumer AI can be fun and fast, but regulated work is where budgets sit, contracts get renewed, and compliance teams make or break deployments. Copilot has to be good enough to win, and dependable enough to be allowed.

So Microsoft is resetting how Copilot works and how it charges. Andreou eliminated extraneous versions of the AI tool and merged teams focused on consumer and enterprise versions. Seat licenses still matter, but Microsoft also tweaked its pricing, introducing a consumption-based model that is becoming common in the AI era as investors ask for clearer returns. Copilot Cowork, its autonomous agent platform, bills by usage based on factors like model use and runtime, though it still requires a 365 Copilot license. The subtext: Microsoft wants customers to feel like they are paying for outcomes and intensity, not just checkboxes, while also giving investors a more measurable path to monetization.

Behind that product overhaul is a culture shift that Andreou represents. After only a year at Microsoft, he was promoted in March to one of the company’s most important roles, an unusual move that signaled how fast Nadella wanted Copilot to improve. Andreou is also now overseeing more than 11,000 people. In Fortune’s account, he is a “talkative” executive who relishes intricate details of AI models, formerly led product at Snap, and is known for getting into coding minutiae himself. He told Fortune, from Microsoft’s Mountain View, Calif. campus, that the moment tech is in is about focus and urgency, and that because the technology is moving so quickly, a six to twelve month roadmap “doesn’t really exist in the way it used to.”

That urgency comes with methods that look more like startup life than classic enterprise software. Employees described AI lab teams locking themselves inside offices to hack all day, and Microsoft hosting developer hackathons near Seattle and San Francisco. Some changes are modest, like using Slack for many communications, even though Microsoft said Andreou’s staff all use Teams. More ambitious is the emphasis on so-called “10x developers,” workers described internally as ten times more productive than peers. Andreou said world-class AI products require “surge” periods, but teams also need sustainability, including runs where staff go hard Monday through Thursday and then take Friday for home duties.

But sustainability is where friction shows up. Current and former employees said the new order can burn people out, with some teams effectively requiring 12-hour days. Others described a daily panic to keep up with Anthropic and other labs, and they worry that the speed of Copilot teams risks shipping products out of compliance with internal standards. Some critics also say Andreou can be overly confident even when it’s not clear he’s correct, and that he still has to prove himself in enterprise software. In other words, Microsoft is trying to win an AI race without reintroducing the one thing enterprises punish: unreliability.

This is happening while Microsoft restructures its AI leadership. When Nadella consolidated Copilot teams earlier this year, he elevated Andreou and shifted Mustafa Suleyman, the former consumer Copilot head, to focus on proprietary AI models. Andreou’s Copilot organization and Suleyman’s lab are separate but work closely. In April, the company offered its first voluntary buyouts, part of a broader push that also includes major changes in units such as Xbox. Meanwhile, the company’s leadership has told employees that the barrier to entry in AI has dropped, allowing smaller companies to ship rapidly and threaten Microsoft.

The Build conference earlier this month added fuel. Microsoft unveiled new AI models to showcase competitiveness and teased a purpose-built agentic platform for devices, including a desktop device and a wearable badge for interacting with agents. Nadella has warned against industry reliance on only a few AI models and said in the past he is focused on making Copilot an accessible AI assistant. Microsoft is also bringing more compute capacity online to train Copilot, reinforcing the idea that this reset is not only product and culture, but also horsepower.

For executives watching from the sidelines, Andreou’s burger test captures a bigger strategic shift. Microsoft is trying to turn agentic AI into a reliable enterprise system while keeping up with labs that move at startup speed. That requires organizational speed, compliance discipline, and monetization clarity all at once. Andreou’s job is to deliver on all three, fast, without breaking trust. In an environment where a six to twelve month roadmap “doesn’t really exist,” the winners will be the companies that can ship and still pass the enterprise reality check.