When you walk into the offices of a major technology company, you might expect to see software developers and engineers hunched over keyboards, eyes glazed over as they stare at each line of programming code.
But what is the reality today? Much of the code writing has been taken over by AI.
Spotify’s senior engineers haven’t written a single line of code since December, co-CEO Gustav Söderström said on an earnings call last month. Anthropic reportedly uses AI to write 70-90% of its code. Google leadership said in October that AI agents write half of all code.
Ryan J. Salva, Google’s senior director of product management, said that number is “much higher now.”
“The way the industry builds software is undergoing a fundamental shift,” Salva told Business Insider. The 2025 report from Dora, Google Cloud’s research program, surveyed 5,000 technology professionals around the world and found that as of September, 90% of software development workers were using AI in the workplace, a 14% year-over-year increase.
So, as AI becomes more prevalent in the coding field, what exactly are software developers at the country’s top technology companies doing?
Julian Togelius, a professor of computer science and engineering at New York University, said developers and engineers are moving from programming and syntax to design and management. They’re taking on roles that rely more on judgment than JavaScript, effectively rewriting what it means to be a software development 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 looks back five years ago, developer value was rooted in programming languages like Python and JavaScript. Their daily work revolved around opening a code editor and writing “if-then” statements.
“Today, that is no longer the case,” Salva said. The value of developers, he said, is in deciding what to build, thinking about software at an architectural level, and anticipating potential problems. At Google, Salva often asks his teams to use critical judgment and discretion over what features to build and which bugs to fix. Instead of writing code by hand, Salva urges teams to “exert more autonomy, discretion, and judgment.”
As this shift occurs, Togerius believes certain types of people with talent management experience will excel. By working with multiple AI coding agents, they are using skillsets similar to those needed to supervise a team, including frequently context-switching and creating high-level documentation and instructions to provide to agents.
“This is a completely different skill than just writing code,” Togerius said.
Managing multiple agents can make developers feel “very powerful,” Togelius added. In fact, Dora’s 2025 report found that 80% of software development professionals feel that AI has increased their productivity.
However, going back and forth to check the status of different agents can also lead to burnout.
“Suddenly you don’t have complete control over your time,” Togelius says. “It changes your relationship with work and, in a strange way, you actually lose agency.”
When a developer prompts a model, he or she may write lines of code by observing the model in action. The experience provides a dopamine hit similar to scrolling through TikTok. But while they are watching the agent’s work, they are not exactly doing the work themselves. This creates a disconnect where developers may feel like they’re more productive, but they’re actually not, Togelius said.
Keep up with changing technology
Salva warned that too much technology change across an industry can create challenges for change management on the human side. Salva said the team is now focused on keeping up with any changes in engineering, as the day-to-day focus has shifted away from writing code. “We need to make sure we continue to create time and space for engineers to learn new ways of doing things,” he says.
One way Google accomplishes this is by appointing hundreds of employees embedded in engineering teams. Their responsibilities include staying up to date with new tools and features. Then hold workshops or office hours where colleagues learn how to use the tools effectively.
“We can all laugh about where the tools are still rough, but we can also share tips and tricks about what’s actually working well,” Salva said.
Coding is just the tip of the iceberg when it comes to the potential of AI in the software development lifecycle. Dora’s report found that while writing new code is the most common use case for AI, more than half of developers use the technology to write test cases, analyze data, and debug software. Togelius said the model is continually being improved, making fewer errors and allowing it to work longer.
Beyond writing code, Salva sees huge opportunities in development and operations. This year, his team is particularly focused on using AI to maintain and scale applications deployed to end users.
“That’s the next frontier of AI,” Salva said.
