How Amazon’s generative AI tool for developers is saving 4,500 years of work, $260 million annually
By Cloey Callahan • August 22, 2024 •
Ivy Liu
This story was first published by Digiday sibling WorkLIfe
Developers don’t want to find themselves spending a significant amount of time on repetitive, thoughtless tasks. GenAI is helping them get out of that spot.
Specifically, at Amazon, developers previously reported that 70% of their time is spent on tedious and repetitive tasks rather than coding. That’s why they are leveraging new tools like the AI-powered software development assistant Amazon Q Developer to remove the grunt work.
The tool helps developers improve productivity and code more securely, while also helping junior developers ask more questions and learn by example from the 24/7 assistant.
The proof is in the pudding. Amazon CEO Andy Jassy disclosed in the company’s Q2 2024 earnings call that with Q Developer’s agents for code transformation capability, Amazon has migrated 30,000 product applications from Java 8 or 11 to Java 17, which is saving over 4,500 years of development work and $260 million annually from performance improvements. It’s been available to the general public since April 2024. Just a few months later, it’s easier to understand the impact it has had on developers.
So how exactly does it work and in what ways is it improving workflow?
Amazon Q Developer’s capabilities
Amazon Q Developer generates highly accurate code and can have a conversation about that code. Those conversations include filtering out code suggestions that might be biased or unfair, implementing new code generated from developer requests, and debugging and troubleshooting.
“A big part of it is how much of a developer’s daily work is not that interesting,” said Doug Seven, GM and director of AI developer experiences at Amazon Web Services. “It’s a lot of mundane work. It’s all necessary, but it’s not super cognitively interesting or challenging. If AI can do some of that for the development team and make them put their cognitive power into something more interesting and novel that has more business value, that’s great.”
Early indications signal Amazon Q could help an organization’s employee become more than 80% more productive at their jobs. When Amazon Web Services ran a productivity challenge, developers who used Amazon Q Developer were 27% more likely to complete tasks successfully. That’s in part because it’s simple to use and doesn’t require training to understand.
Seven compares it to another developer sitting next to you watching the work you’re doing and providing suggestions. He says it’s like super autocomplete. But the developer can also ask questions about where the possible problem is in the code.
Another way to use Amazon Q Developer is to have an AI agent. For example, code transformation from Java 8 or 11 to Java 17 is tedious and can take up to two days per program. Instead of the developer doing this themself, they can assign that task to Amazon Q.
“In some ways, it can be used like another engineer on the team that would do work and then come back for peer review and see if it’s up to snuff,” said Seven.
And when a developer really starts to understand the best work to delegate to AI, it can set them up for some real success.
“I like to say that any individual developer can now become a team of developers,” said Seven. “I can dispatch my work to different agents and I can do other things while the agents are doing that work.”
The ripple effect
With that much change in productivity thanks to this new tool, some people might wonder what’s left for the developer to do. It’s been a big question with the boom of AI, but most people argue that it will allow workers to do more fulfilling and important work instead. Plus, a person will always have to be in the loop.
“Asking an AI assistant to write some code is not very different from asking it to translate English text into French with the correct semantics,” said Jason Andersen, vp and principal analyst for Moor Insights and Strategy, where he covers application development and platforms. “In both cases, the requestor still needs to be in the loop and provide the context of the situation to request the work.”
There are also other gaps that developers need to account for when leveraging AI. For example, Andersen says that AI today is not smart enough to completely understand the nature of an application that is composed of thousands of files spread across many systems and locations.
“Given the complexity, an AI agent will also not truly understand the context of the previous or potential decisions and constraints,” said Andersen. “So, a developer must be involved to review the AI recommendations and decide if they are useful.”
While Amazon Q Developer has guardrails to ensure an AI agent is acting responsibly, not everything from AI is going to be correct all the time. That’s where the human oversight to ensure that things are running smoothly comes back in.
“I think I would have to say that the developer role has been evolving for a long time,” said Andersen. “The range of skills, including collaboration and coordination keeps becoming more advanced. So, AI not only would have to catch up to what developers do now, but also evolve with the future evolution of the role. I think this is in contrast to other jobs that AI could replace someday.”
https://digiday.com/?p=553191
