DORA Metrics: Navigating AI's Impact on Software
Explore how DORA metrics and AI reshape software delivery, revealing challenges, opportunities, and evolving industry trends.
Written by AI. Yuki Okonkwo
Photo: GOTO Conferences / YouTube
When Nathen Harvey and Charles Humble sat down to discuss the state of software development, they dove into a world where DORA metrics and AI intersect—a landscape that's as thrilling as it is complex. With AI promising to speed up development, the conversation shines a light on a paradox: increased AI adoption initially correlates with decreased stability and throughput, the very metrics teams have been optimizing for decades. So, what's going on here?
The DORA Metrics
First things first, let's get our bearings. DORA—short for DevOps Research and Assessment—has been a staple in software development for over a decade. Harvey, leading the DORA team at Google Cloud, describes it as a research program that looks into metrics, capabilities, and conditions that teams need to assess and improve software delivery performance. In plain English: How fast and reliably can you get your code into the hands of users?
These metrics fall into two categories: throughput and stability. Throughput measures how quickly changes are made, while stability assesses how reliable those changes are. "Teams either have fast throughput and high stability or low throughput and low stability," notes Harvey.
Enter AI, Stage Left
Now, onto the twist in the tale. Harvey reveals a surprising finding from DORA's 2024 report: AI adoption initially leads to decreased stability and throughput. Picture this—teams are churning out code faster, but the systems can't handle the extra load, leading to chaos rather than efficiency. It's like giving a Ferrari to someone who hasn't mastered driving a stick shift yet.
But Harvey notes a shift in the 2025 report: "Throughput is actually improving as you use more AI, but we still see high levels of instability." This aligns with the idea that AI isn't a magic wand; it's an amplifier of existing systems. If your systems are shaky, AI might just make them wobble more. The real challenge is integrating AI in a way that complements and enhances existing workflows.
The Paradox of Code Generation
Harvey points out a critical insight: "It is rarely the case that writing code is the bottleneck in that process." And yet, much of the AI hype focuses on generating more code. But if the bottleneck lies elsewhere, like in testing or deployment, all that extra code can actually slow things down. It's a classic case of optimizing the wrong part of the process.
Charles Humble echoes this sentiment: "Measuring lines of code or GitHub commits are just terrible metrics... they don't really tell you anything." In other words, more code doesn’t equal better software.
The Road Ahead
So, where does this leave us? Harvey and Humble's conversation points to a crucial takeaway: AI's true potential in software development lies in its ability to enhance, not replace. Organizations must assess their capabilities alongside DORA metrics to formulate actionable hypotheses for improvement.
"As you improve software delivery performance, you're also improving those outcomes that matter," Harvey asserts. The key is using DORA metrics not just as a measure, but as a guide to identify where improvements are needed.
As we stand on the brink of an AI-driven era in software development, it's clear that the journey won't be without bumps. But with thoughtful integration of tools like DORA metrics, teams can navigate the complexities and seize the opportunities AI offers—if they know where to look.
Yuki Okonkwo
More Like This
The 5 Stages of DevOps: Console to AI-Assisted Infra
From AWS console clicks to AI-generated Terraform—here's how infrastructure management actually evolved, and why skipping stages costs you later.
AI's Wild Week: From Images to Audio Mastery
Explore the latest AI tools reshaping images, audio, and video editing. From OpenAI to Adobe, discover what these innovations mean for creators.
Why Docker Books Still Matter in 2025
Elton Stoneman's updated 'Learn Docker in a Month of Lunches' reveals the gap between Docker beginners and experts—and why fundamentals still matter.
Effect-Oriented Programming: Making Side Effects Safe
Three authors explain how effect-oriented programming brings type safety to the messy, unpredictable parts of code—without the intimidating math.
Your CI/CD Pipeline Is Now Your Product
CircleCI's Olaf Molenveld on why managing deployment infrastructure has become as complex as the software it produces—and what that means for teams.
Enterprise Architecture Still Runs on Waterfall Time
Simon Rohrer argues EA must shed its phase-gate past and align with DevOps, outcomes, and evolution. The diagnosis is right. The prognosis is harder to share.
Gen Z's Complicated Relationship With AI
Gen Z uses AI daily but resents it deeply. A Harvard poll and campus booing incidents reveal a generation caught between FOMO and genuine fear about their future.
AI Leaderboards Are Lying to You About State-of-the-Art
Bertrand Charpentier of Pruna AI makes the case that 'state-of-the-art' is a broken concept—and that efficiency belongs in the same sentence as quality.
RAG·vector embedding
2026-04-15This article is indexed as a 1536-dimensional vector for semantic retrieval. Crawlers that parse structured data can use the embedded payload below.