SerpApi.Org
- A Microsoft engineer says AI has made software development dramatically faster.
- He says AI still hallucinates, but it's made deadlines much less stressful.
- He shares why he thinks AI won't replace engineers who keep learning.
This as-told-to essay is based on a conversation with Rahul Devikar, a 34-year-old senior software engineer at Microsoft who lives in the Seattle area. It has been edited for length and clarity.
I've worked as a software engineer at Microsoft since 2018. Over the years, I've had a front-row seat to how AI has transformed the way software gets built.
One of the biggest projects I've worked on is Microsoft 365 Agents SDK, an open-source library for building AI agents. It was my first open-source project at Microsoft, and we took it from idea to public preview in just a few months.
While projects like this require a lot of work, I haven't experienced much of the AI fatigue that I've heard other engineers talk about. That's because overall, I think AI has made it much easier to get projects done.
That said, AI comes with its own challenges. Sometimes an AI agent hallucinates, and figuring out why can be frustrating. As long as you're checking what it produces along the way, though, I think AI helps more than it hurts.
The biggest AI time savings isn't writing the initial code
In my experience, it used to be much more common to work long hours just to hit a deadline. There was often a concern about whether we'd finish a project on time. Now, we can usually build the first version much more quickly and spend more time refining and improving it instead.
Some projects that used to take about a month can now be finished in three or four days — or at most a week. Using a combination of internal tools like Copilot and external ones like Claude, we have a lot more opportunities to build proofs-of-concept, and AI has dramatically sped up development.
For bigger product releases, though, the biggest time savings haven't come from writing the initial code. They've come from making changes after the first version is built.
When you design something, show it to other people, and realize certain things need to be improved, you have to go back and make those changes. Back in 2020, for example, that process might have taken another month or two. Today, it's often a matter of days or weeks.
Business Insider is speaking with tech workers affected by recent workforce shifts, including layoffs, AI adoption, and changing workplace expectations.
Share your story by filling out this form.
The biggest challenge is knowing when AI is wrong
Despite the benefits of AI tools, they aren't perfect and can make mistakes.
I've seen AI agents hallucinate when asked to complete engineering tasks. The biggest challenge isn't that they hallucinate — it's recognizing when they've hallucinated. If you don't already understand the task yourself or know what to expect, it can be difficult to spot when the AI has made a mistake. I've found that the better you are at catching those mistakes, the more successful you'll be using AI.
Overall, I think AI should be thought of as an assistant, not a replacement. As a user, you have context that the AI doesn't always have. One of my biggest concerns is that people will become too reliant on AI without developing the underlying knowledge themselves. I think that could lead to more "AI slop."
Now is the time to start experimenting with AI
I think there's a comparison to be made between AI and the internet. When the internet first took off, there was a lot of hype around it. Today, it's such a normal part of our lives that we can't imagine living without it.
I think AI is on a similar path. There's a lot of excitement around it right now, but eventually it will just become part of everyday life. People won't think about using AI any more than they think about using the internet today.
That's why I think it's a good idea for people to experiment with AI tools and try vibe coding. If you have an idea, try building it. I think many people would be surprised by how much they can accomplish.
When it comes to working with AI tools, I think it's helpful to remember that you're working with an AI, not a human. The tool might not understand what you're trying to say, even if another person would.
You usually need to be much more descriptive about what you want. I've found it's more effective to break a task into smaller steps instead of asking the AI to do everything at once. That approach tends to make the whole process much more efficient.
I'm skeptical that AI will replace software engineers
Personally, I experiment with AI tools like Agents365, Claude, and GitHub Copilot to get more hands-on experience with the technology. During tax season, for example, I built an agent to help me understand my taxes.
Despite the benefits of using AI I've seen over my years at Microsoft, I'm skeptical that it will lead to fewer software engineering jobs. While there have been tech layoffs in recent years, this isn't the first time the industry has gone through layoffs.
I'm sure AI has played a role in some layoffs, but I also think engineers who keep learning new technologies will be in a much better position going forward. Software engineering has always required constant learning, and I don't think that's going to change. I think it's going to create more opportunities to build new products, and at the end of the day, engineers will still be needed to design and develop them.
AI can recommend different approaches or generate code, but it's still up to engineers to decide what technologies to use and how to build the product.
Do you have a story to share about learning AI or working in tech? Reach out to the reporter via email at [email protected], or via Signal at jzinkula.29.
If you enjoyed this story, be sure to follow Business Insider on MSN.
