How to measure AI developer productivity in 2025 | Nicole Forsgren
Nicole Forsgren, creator of the DORA and SPACE frameworks, explains why traditional productivity metrics are failing in the age of AI. She details how AI impacts developer flow state, shifts the focus from writing to reviewing code, and provides a new framework for measuring and improving Developer Experience (DevEx) by aligning engineering efforts with strategic business goals.
AI is accelerating coding, but developers aren’t speeding up as much as one might think. The bottlenecks have simply shifted. While AI tools can generate code rapidly, developers still contend with broken builds, unreliable processes, and new challenges related to trusting and integrating AI-generated code. The focus must shift from measuring raw output to holistically improving the entire Developer Experience (DevEx).
Most Productivity Metrics Are a Lie
Traditional engineering metrics, especially lines of code, are now more misleading than ever. In the age of AI, “if the goal is more lines of code, I can prompt something to write the longest piece of code ever.” This approach is easily gamed and often introduces unnecessary complexity and technical debt.
Frameworks like DORA (Deployment Frequency, Lead Time, MTTR, Change Fail Rate) are still useful for assessing the overall health and speed of the delivery pipeline, but they are no longer sufficient on their own. AI introduces much faster, earlier feedback loops that these metrics don’t capture. The SPACE framework (Satisfaction, Performance, Activity, Communication, Efficiency) remains more adaptable because it’s a framework, not a prescriptive set of metrics. However, it should be augmented with new dimensions, particularly Trust. Given that LLMs are non-deterministic, developers cannot blindly accept generated code; they must constantly evaluate it for hallucinations, reliability, and style consistency.
The Three Pillars of Developer Experience
DevEx is about the day-to-day reality of building software. It can be understood through three core, interlocking components:
- Flow State: The immersive state of deep work where developers are most productive.
- Cognitive Load: The mental effort required to perform tasks. High cognitive load, often from wrestling with complex tools or processes, leaves less mental capacity for innovative problem-solving.
- Feedback Loops: The speed at which developers receive information about their work, from local builds to customer reactions.
AI has a complex, dual impact on these pillars.
How AI Reshapes Developer Workflows
The nature of software development is fundamentally changing from long, uninterrupted blocks of typing to a more interactive, interrupt-driven process of prompting, reviewing, and integrating.
- Impact on Flow State: While the constant interaction with an AI tool can feel disruptive to traditional flow, it also presents new opportunities. The machine can help manage context, allowing developers to re-enter a state of flow more quickly. This may make shorter work blocks (e.g., 45 minutes) more viable and productive than before, as the AI can assist in context-switching and offloading routine tasks to “agents.”
- The Shift from Writing to Reviewing: A significant portion of a developer’s time is shifting from writing code to reviewing AI-generated code. This requires a different skillset focused on critical evaluation, debugging, and ensuring the code’s quality and security.
- The Centrality of Trust: A primary challenge with AI is learning how much to trust its output. Developers must constantly be vigilant, which adds a new layer of cognitive load.
Signs Your Engineering Team Can Move Faster
Most teams have room for improvement. Key “smells” that indicate significant friction and opportunity for acceleration include:
- Constantly breaking builds.
- Flaky, unreliable tests.
- Overly long and cumbersome processes.
- High friction in provisioning new environments.
- Significant “switching costs” when developers move between projects or teams, often requiring them to relearn entirely new systems and workflows.
However, the goal isn’t just to “ship faster.” The crucial question is, “faster for what?” Without a clear strategy and smart decisions about what to build, you can “ship trash faster every single day.” The strategy is the core piece, and engineering velocity should serve that strategy.
A Framework for Improving DevEx
To systematically improve DevEx, start by treating it like a product. This involves a continuous cycle of discovery, prioritization, and iteration. The upcoming book Frictionless outlines a seven-step process:
- Start the Journey: Begin with a listening tour. Talk to developers to understand their daily workflows, pain points, and moments of delight.
- Get a Quick Win: Identify a low-lift, high-impact project to build momentum and demonstrate value.
- Use Data to Optimize: Establish a data foundation, combining system data with qualitative insights from surveys.
- Decide Strategy and Priority: Use the data to decide which larger problems to tackle next.
- Sell Your Strategy: Get buy-in from leadership and other teams by clearly communicating the “why” behind your plan.
- Drive Change at Scale: Implement changes, using different strategies for grassroots efforts versus top-down initiatives.
- Evaluate Progress and Show Value: Measure the impact of your work and loop back to the beginning.
Measuring the Real Impact of AI and DevEx
To measure gains effectively, you must align your metrics with what leadership cares about. Frame the impact in terms of business value.
- If leadership cares about market share, focus on speed and velocity. Measure metrics like time from idea to customer experiment.
- If leadership cares about profit margins, focus on cost savings. Quantify the impact of reduced cloud spend from more efficient build/test systems or recouped developer time from automating toil.
- If leadership cares about transformation, frame your efforts as enablers of disruption and innovation.
When starting with measurement, surveys are an incredibly effective tool. Ask developers about their satisfaction (which is more actionable than “happiness”) and have them identify and rank their top three biggest barriers to productivity. This provides a clear, prioritized list of where to focus your efforts.