How Microsoft tests its own developer tools to build an AI-first engineering organization
AI coding tools are among the most visible manifestations of AI in business so far. The startup Cursor has grabbed headlines with its rapid growth and $29 billion valuation. Meanwhile, GitHub Copilot from Microsoft is widely regarded by analysts like IDC as the leader in enterprise adoption of AI coding assistants, boasting deep penetration inside large organizations.
Amanda Silver, CVP and head of product for apps and agents at Microsoft, has been with the company for almost 25 years. She has had a hand in many of the major developer tools over that time, including Visual Studio, Azure Application Platform and GitHub Copilot, but recently shifted roles. Today, she is focused on bringing Microsoft’s application and agent platforms together to deliver a better developer experience.
But she doesn’t just work with external customers. She’s also in charge of helping internal teams act as a testing ground for these products and learn the best way to use them before they roll out to customers.
“So what we do is we take all of the products that we share with our external customers, and we host them, manage them, extend them, administer them for the Microsoft internal developer population building our first-party products like Excel, Azure, Office 365, Bing — all of the products basically,” Silver told FastForward. This enables her to get feedback quickly from internal developers, which she believes makes these products better as they roll out to customers.
This is a similar approach to the one Roger Barga, head of artificial intelligence and machine learning at Oracle, takes with product development. “Throughout my career, I have built services that first served our internal needs. You have a very active audience willing to give you feedback and engage with you very early on,” he told FastForward in a July interview.
Likewise, Silver has spent years on the front lines of both internal and customer developer needs, and she has a unique perspective on how AI is transforming the engineering role from writing lines of code to focusing on broader system architectural issues at companies of all sizes from startups to giants like Microsoft.
Reducing the grunt work
There are parts of every job that people don’t enjoy. AI at its best is intended to decrease those tasks we like least, helping achieve quality over quantity. Traditionally, programmer productivity was measured in lines of code produced, but AI is changing that, enabling them to focus on design, security, compliance and delivery of the best possible product.
“It used to be that everybody thought about developers as being cogs in a machine,” Silver said. “The value of developers was measured by how quickly they could output code, and the code quality, and then how quickly they could burn down bugs that were in their backlog.” And the tools really focused on things that helped them perform these kinds of tasks as fast as possible.
She says now with agents being inserted into the different phases of the development cycle, the script is being flipped. Instead of measuring productivity in lines of code, it’s about efficiency and automation. “Our focus now is on getting the most miserable, soul-draining parts of the job and transforming them with AI, as opposed to just generating AI code,” she said. “Because at the end of the day, developers really want to focus on the code that's unique to them, unique to their organization, unique to their scenario, unique to what they're trying to build.”
Shifting the culture
In many ways, any AI project is about organizational transformation. It requires changing the way you have been working and moving to something completely different. That level of change can be intimidating. You are asking professionals who have worked a certain way for years to alter their approach in a dramatic way.
Silver says there are different strategies to make this work. You can obviously mandate use of certain tools, but she prefers an approach where there are successful early adopters, who then carry that message to the rest of the organization.
Our focus now is on getting the most miserable, soul-draining parts of the job and transforming them with AI, as opposed to just generating AI code
“You really need to think about it from a systems perspective in terms of how you incent everybody, and how you think about who the beacon teams are. What are examples of hero teams that can adopt it as a tip-of-the-spear example that everybody then kind of models,” she said.
Part of that is sharing learnings so everyone can benefit, internally and externally. One example of this approach is the .NET Aspire team, which develops its projects in the open and lets anyone access the GitHub codebase. “David Fowler, the technical fellow for .NET Aspire, frequently shares updates on LinkedIn and Twitter, openly discussing the significant impact that the Copilot coding agent has had on the team's contributions in recent months,” she explained. It gives the broader developer community visibility into how the work gets done using the AI agent.
Building a feedback loop
While Microsoft developers help inform the AI coding tools, they can also learn from the tens of millions of users as well. “Microsoft developers get to benefit from having some of the best tools in the industry that we get to innovate on with fast feedback cycles from our internal developers back to our developer products,” Silver said.
That enables them to continuously improve and learn based on what they’ve built, incorporate that back into the products, then share those experiences and updates with external customers. “So I think in some sense, this is a luxury that a lot of organizations don't necessarily have, but every organization gets to benefit from because many of them are using the tools that we build."
Silver has been on the forefront of great change throughout her career, none more significant than the tectonic shift that AI has brought. As this has happened, Microsoft engineering groups have acted as a lab helping each other learn to use AI more effectively, while building what they have learned into the developer tools they sell, and also sharing their knowledge with the broader community.
Featured photo courtesy of Microsoft.
