Google research: What do developers want from AI?

Google’s deep research has raised real questions about AI involvement in developer workflows and other engineers’ processes.

AI is not new, but developers still have skepticism about new technology with deep integration in their tools for everyday jobs. Do AI tools improve the quality of the code, and do they help developers find more creative solutions? These are the questions frequently asked today.

Developers want to maintain control over AI

In many interviews and user studies with developers in Google who had varied experiences with AI-enhanced developer tooling, they expressed a desire for AI-enhanced developer tooling to save them time and energy by helping them do their jobs more efficiently.

Research has shown developers do not want AI to change their workflows (right now) fundamentally, they want AI to help them do what they are already doing, getting from point A to point B.

Notably, developers are not only focused on speed: efficiency is more than just speed, it’s also about making things easier and at least maintaining, if not improving quality. Google said they also heard from developers that they want AI to support simpler tasks and reduce toil, allowing the developers to focus more of their energy on the complex problem-solving and creative aspects of their jobs.

In the mentioned research, one of the developers said:

Automating menial and repetitive tasks could help me focus more on the use case and the problem at hand.

Similarly, another developer expressed how AI could help them feel more creative by saving time on simpler tasks:

It feels much more impactful if I can spend less time on the finer details. I think AI-powered tools will help me be more creative and think about the bigger picture and long-term goals.

This is because, right now, engineers want to stay in control, and that is okay. It is also consistent with perspectives shared by engineers outside of Google.

When it comes to AI in developer tooling, developers want to stay in the “driver’s seat”. They want to maintain control and still decide how to get from point A to point B, even if AI can help them on the way.

Engineers want AI to transform the way they work

Another developer in research reflected, “I don’t think I would ever trust it 100% like we don’t trust human-written code 100%. We’re like, it needs a code review; it needs to be looked over by someone else.” This is good because we are still in the early days of AI, so we still want someone “behind the wheel.”

Google highlighted that developers want to keep doing the most rewarding aspects of their workflows and want AI tooling to help expedite the less rewarding elements. And what falls into each category will likely change over time, also in response to developments in AI.

Majority of developers express positive impacts of AI

“When the suggestions and the assistance can get ahead of where I’m thinking and allow me to think faster, that to me is interesting. It’s not the typing speed that is the bottleneck for most software developers; it’s the ability to think things through.”, one of the developers in the research explained.

We can think of AI as a tool for enhancing and extending human capabilities, particularly in areas where developers have already expressed pain points.

The top hindrance is consistently technical debt, and the following two most common hindrances are interesting opportunities for AI:

1. Poor or missing documentation;

2. Learning a new platform, infrastructure, framework, or technology.

When asked how they feel about AI in their workflows, the majority of developers expressed positive impacts of AI and are in various stages of trusting it to assist their workflows; others said they do not want AI in their workflows.

Research has shown that engineers using AI-powered tools were more likely to write insecure code during security-related tasks than those without. Adding AI to developer tooling isn’t without risk, and not all developers are enthusiastic about it.

With the rapid innovation and development of AI specifically in developer tooling, many questions are increasingly important and urgent.

As we consider the future of software with AI, it’s important to consider developers’ fundamental goals (regardless of what they imagine them to look like) and how they want AI to support them.