Walking into a major technology company’s office, someone might expect to see rows of software developers and engineers hunched over their keyboards, going bleary-eyed from staring at line after line of programming code.
But the reality today? AI has taken over the bulk of code writing.
Senior engineers at Spotify haven’t written a single line of code since December, Gustav Söderström, co-CEO, revealed during an earnings call last month. Anthropic reportedly uses AI to write 70-90% of its code. In October, Google’s leadership said AI agents are writing half of all code.
The figure is “much, much higher now,” said Ryan J. Salva, senior director of product management at Google.
“We’re going through a radical change in the way that the industry builds software,” Salva told Business Insider. A 2025 report from Dora, a Google Cloud research program, surveyed 5,000 tech professionals around the world and found that 90% of workers in software development were using AI at work as of September, a 14% increase over the previous year.
So what exactly are software developers at the nation’s top tech firms doing, as AI becomes ubiquitous in coding?
Developers and engineers are moving away from programming and syntax toward design and management, said Julian Togelius, professor of computer science and engineering at New York University. They’re taking on roles where judgment is more important than JavaScript, effectively rewriting what it means to be a software developer professional. This transition comes with its own set of pressures and must be handled carefully from a change management perspective.
From programmer to manager
When Salva thinks back five years, a developer’s value was rooted in programming languages like Python or JavaScript. Their day-to-day work focused on opening a code editor and writing “if-then” statements.
“That is no longer the case today,” Salva said. Developers’ value, he said, lies in deciding what to build, thinking about software at an architectural level, and foreseeing things that could go wrong. At Google, Salva often asks his teams to exercise critical judgment or use their discretion on which features to build or bugs to fix. Instead of manual code authoring, Salva calls on his teams to “exercise more autonomy, more discretion, and more judgment.”
As this transition occurs, Togelius sees a particular type of person excelling: those with experience in people management. By handling multiple AI coding agents, they’re using somewhat similar skillsets required to oversee teams, such as frequently switching contexts, or writing high-level documents and instructions to provide to agents.
“This is quite a different skill from just writing the code,” Togelius said.
Managing multiple agents can make a developer feel “super powerful,” Togelius added. In fact, Dora’s 2025 report found that 80% of software development professionals feel AI has increased their productivity.
But flipping back and forth to check on the status of various agents can also lead to burnout.
“Suddenly you’re not really in control of your own time,” Togelius said. “It changes your relationship to work and actually gives you, in some bizarre way, less agency.”
When developers prompt a model, sometimes they watch it work and draw up lines of code. The experience delivers a dopamine hit like scrolling TikTok. But while they’re watching the agent work, they’re not exactly working themselves. It creates a disconnect where developers may feel more productive but actually are not, Togelius said.
Keeping up with the changing tech
Salva warned that too much technological change across the industry can pose challenges for the human side of change management. Salva said as the day-to-day focus shifts away from writing code, teams are now focused on keeping pace with all the ways engineering is changing. “We need to make sure that we’re continuing to create the time and the space for engineers to learn the new way of doing things,” he said.
One way Google does this is through appointing hundreds of employees embedded within engineering teams. Their responsibilities include keeping up to date on new tools and capabilities. Then they hold workshops or office hours where coworkers learn how to use the tools effectively.
“They can all laugh about where the tools are still a little bit rough around the edges, but they can also then share tips and tricks about what is really working well,” Salva said.
Coding is just the tip of the iceberg when it comes to AI’s potential in the software development lifecycle. While Dora’s report found that writing new code is the most common use case for AI, more than half of developers are using the tech to create test cases, analyze data, and debug software. Togelius said the models are continuously improving, with fewer errors and the ability to work longer.
Salva sees a great deal of opportunity beyond code authoring into development and operations. This year, his team is especially focused on using AI to maintain and scale applications once they’ve been deployed to end users.
“That is the next frontier of AI,” Salva said.
