Applying AI where it matters
Microsoft’s recent study offers a path to target AI investments based on what developers actually need.
Welcome to the latest issue of Engineering Enablement, a weekly newsletter sharing research and perspectives on developer productivity.
On December 11th, Laura is hosting a live year-in-review roundtable with developer productivity researchers from Microsoft, Google, and GitHub. Register to join here.
I also just announced my book, Frictionless, with Nicole Forsgren, on developer experience. Get a copy here.
This week, I’m summarizing a recent paper by Microsoft researchers: AI Where It Matters: Where, Why, and How Developers Want AI Support in Daily Work. This study examines where developers actually want AI support in their daily workflows. For leaders evaluating or rolling out AI tooling, it offers insight into where AI can deliver real value and where gaps still exist.
My summary of the paper
Leaders want to integrate AI where developers need it, and this has been an ongoing area of research. However, while recent studies have explored which tasks developers want automated, the authors of this paper saw an opportunity to dig deeper into questions of where in their workflow developers actually turn to AI, why they want help there, and how they decide when to use or avoid it.
To answer these questions, the researchers conducted a mixed-methods study with 860 Microsoft developers. They started by developing a list of day-to-day tasks to ask developers about, as well as a list of AI safety and control features. They then surveyed developers and asked several questions about each task they selected, including:
How they view each task (how important it is, whether it feels tied to their identity or job pride, whether they would be held accountable if it went wrong, and how difficult or tiring it is)
How open are they to AI help for that task
How often do they use AI for that task
Where they want AI support and where they don’t want it
Which five Responsible AI principles are most important for AI tools in that task area
Here are the key findings from the study.
Developers’ perceptions of a task strongly shape their willingness to use AI
The researchers wanted to understand whether the way developers feel about different tasks affects how open they are to using AI, and how often they actually use it. Here’s what they found:
For tasks developers see as important, high-stakes, or difficult, they are more open to using AI and tend to use it more often. They use AI as a way to reduce effort, double-check their thinking, or avoid mistakes, but they still want to stay in control.
For tasks that feel central to their identity (like core coding or design decisions), developers are less open to letting AI take over, but they do use AI to help them improve or speed things up.
For “people work” like mentoring teammates or building AI features themselves, developers were much more likely to limit AI involvement. They see these tasks as deeply human, involving skills like judgment, experience, relationship-building, and personal growth. Developers said mentoring needs trust and empathy, and building AI tools requires craftsmanship. In these cases, they prefer to do the work themselves, using AI only at the margins.
Personal traits matter. Junior developers and those with more AI experience use AI more. People who are more risk-tolerant or enthusiastic about AI tend to rely on AI more often, especially for high-stakes or demanding work. More cautious developers remain careful with AI when the work feels risky.
Finally, for operational and coordination work, like maintaining systems, setting up environments, updating documentation, and managing logistics, developers strongly want AI help to cut down on repetitive or boring tasks. But there’s a catch: they only want AI to help if it is safe, reliable, and easy to supervise. They don’t want AI making risky changes on its own or replacing the human judgment needed for strategic decisions or stakeholder communication.
In short, AI is welcome for tedious tasks, cautiously accepted for high-stakes technical work, and firmly limited for interpersonal or identity-defining work.
To make these patterns easier to act on, the researchers mapped every task onto a simple chart showing how much developers want AI support (need) versus how much they actually use it today (use). This creates four quadrants that leaders can use as a decision tool:
Tasks with high need but low use (“Build”) are prime opportunities for new AI support.
Tasks with high need and high use (“Improve”) benefit from hardening and reliability work; tasks with low need and high use (“Sustain”) should be maintained but not over-invested in.
Tasks with low need and low use (“De-prioritize”) should remain mostly human-led.
This map offers a practical way to prioritize where AI can help and where automation is less likely to add value.
Developers prioritize AI that is safe, private, transparent, and easy to control, especially when the work is high-stakes
The researchers asked developers to pick the five qualities they think are most important for AI tools to have when helping with their work. Across the board, developers overwhelmingly prioritized the basics: the AI must be reliable and safe, protect sensitive information, explain what it’s doing, stay aligned with the developer’s goals, and be easy to steer or override. These qualities mattered most for technical, high-stakes tasks like coding, testing, and operations, where AI mistakes can create real risk or wasted effort; for more human-facing or creative tasks, developers placed greater emphasis on fairness and inclusiveness.
Individual differences shaped priorities as well: more experienced and AI-savvy developers placed even higher value on transparency and control, wanting AI they can inspect and correct.
Final thoughts
This study offers a grounded way to align AI investments with what developers actually want and need. The quadrant map should be an especially helpful starting point for distinguishing high-leverage opportunities for applying AI from areas where automation is unlikely to pay off.
Who’s hiring right now
This week’s featured DevProd job openings. See more open roles here.
American Express is hiring a Sr. Manager, Digital Product Management - DevProd | Hybrid - London UK
Capital One is hiring a Product Manager - Developer Experience | Plano TX; McLean VA; Richmond VA
Plaid is hiring a Software Engineer - Platform | New York, NY
Reddit: Staff Software Engineer - Developer Experience | Remote - United States
Tesco: Senior Product Manager - Infrastructure | Hybrid - Tesco UK, Welwyn Garden City
Whatnot is hiring a Software Engineer - Platform | San Francisco, Los Angeles, Seattle, NYC
That’s it for this week. Thanks for reading.
-Abi