How AWS aims to make generative AI tools that ‘help remove toil’ for developers

Deepak Singh works on a newly formed team that sees coders as a ripe customer base.
article cover

Deepak Singh

· 5 min read

Can AI help coders build more AI?

That’s part of what Deepak Singh, VP of Next Gen Developer Experience at Amazon Web Services, is tasked with figuring out. The longtime Amazon exec was appointed head of a newly formed team last year that’s focused on building generative AI tools that aim to help developers better build products on top of Amazon’s cloud.

In 2023 alone, he oversaw the rollout of tools like Amazon CodeWhisperer, an AI-powered coding assistant; parts of Amazon Q, an enterprise-focused chatbot; and PartyRock, a “playground” for building AI apps on Amazon’s Bedrock platform with no coding needed.

“Our mission at a high level [is] how do we change how customers build applications, and more specifically, How do they build and run those applications within AWS?” Singh said.

These efforts come as Amazon works to keep up with Big Tech peers like Microsoft and Google in the AI arms race. Amazon’s CodeWhisperer is in direct competition with Microsoft’s Github Copilot, and AWS and Microsoft Azure are both trying to lure AI developers to their respective clouds.

Tech Brew spoke with Singh about his team’s strategy, the state of the AI race, and the future of AI-powered coding.

This conversation has been edited for length and clarity.

What have you learned since rolling out tools like CodeWhisperer about how AI can help developers?

What I’m seeing is most organizations start rolling it out in two ways: They either roll it out gradually—they start with one team and one team is successful, and they’ll start opening it up to other teams—or they roll it out to the whole org, and say, “Try it out. We’ll see where we end up,” because it’s so early. And I think different orgs have different skill sets and different cultures, and how they roll it out and use it varies.

[British Telecom] is a good example. They made it broadly available, and a good set of software developers started using it in less than four months. I think they’d written over 100,000 lines of code. And I think they had good acceptance rates, but I think what was useful was they were able to automate 10%–12% of things that they weren’t able to do before…And that’s typically what we see…I like using this metaphor of swimming pools. You go to any swimming pool complex, there’s usually three pools: a children’s pool, a regular pool that goes shallow to deep, and then there’s a swim team Olympics pool. Most of our customers are trying to figure out how to go from the children’s pool to the regular pool. A bunch are in the regular pool and are going into the deep end over the last year…And then there are some that go, “You know what? This thing is going to be transformational. We are going to make organizational changes and go right into the Olympic pool…What I really see people doing with things like CodeWhisperer is getting really strong at swimming laps and really understanding how these tools can help them, and using them while recognizing that they’re constantly proving and really quickly.

Keep up with the innovative tech transforming business

Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.

Why has Amazon focused so much on building generative AI tools for developers in particular?

Part of the reason to focus on developers is we've got so many who use AWS every day. So it’s a natural customer base for us. AI happens to be good at development tools, as well. I think the way I see it is to get to this point where your AI assistant—in this case, Q—we would like it to become a partner for you to help you get things done. And getting things done could be understanding documentation…all the way to helping you unblock writing tests, correcting bugs…How can we block and tackle for you while also helping you think through problems, which is the feature development part. That’s a big area of focus for us. We have lots of other ideas but right now, how we’re spending our time is: How can we help remove toil, and how can we help with problem-solving?

What are some of the next steps in terms of expanding how AI can help developers?

We’ve already seen that with models getting better and better, the problem-solving aspect gets more and more [adept] and the kinds of problems that it can help with get bigger and bigger. You can help more and be more collaborative, like almost make it part of the team.

You see examples of that in this product called CodeCatalyst. It’s like a DevOps platform where you can make Q part of your team and give it tasks that you don't want to tackle. And then we’ll go get them done for you, and then it’ll bring you in to make sure that the code reviews are done and all the right things are done. I actually think another area where Q is going to be very, very useful is troubleshooting and helping you get tasks done in the AWS console…it [helps] in a way that a developer can understand, and you don’t need to go running to your network engineer for every little thing. They can focus on the hard ones. I think that is going to be huge for a lot of people.

Then it’ll expand into other areas that are more advanced, which we can’t tackle right now. But as the models get better, as our ability to reason around them gets better, we are building a lot of underlying capabilities that will allow us to help you do more with your AWS resources, then, you’ll see a lot of that happen.

Keep up with the innovative tech transforming business

Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.