AI Is a Multiplier, Not a Magic Bullet: DORA 2025

90% of developers now use AI in their daily work. They spend a median of 2 hours per day interacting with AI tools. 80% report feeling more productive.

And yet organizational delivery metrics stayed flat. Then got worse.

The 2025 DORA Report surveyed nearly 5,000 developers and conducted over 100 hours of qualitative interviews to figure out what is actually happening when teams adopt AI. The answer should make every engineering leader pause.

AI does not fix your team. It amplifies whatever is already there.

#The Mirror Effect

DORA found that AI functions as both a mirror and a multiplier. It reflects your organization's true capabilities while amplifying existing strengths and weaknesses.

For teams with mature DevOps practices, well-defined workflows, and strong platform capabilities, AI becomes a powerful accelerator. They ship more, ship faster, and maintain quality.

For teams with fragmented tooling and broken processes, AI accelerates technical debt creation. It makes bad habits happen at machine speed.

This is not a subtle finding. Organizations lacking a user-centric approach actually experienced performance declines from AI adoption. Not flat. Declines.

#The Numbers Tell the Story

Individual developer output is up across the board. Faros AI tracked over 22,000 developers across 4,000 teams and found 33.7% more tasks completed per developer and 98% more pull requests merged.

Sounds great until you look at what happened downstream.

PR review time increased 441%. PR size grew 51%. Bugs per developer rose 54%. Incidents per pull request jumped 243%. PRs merging with zero review increased 31%.

Developers are also juggling 67% more PR contexts daily, handling 17.7% more task contexts, and seeing 26% more work items stall for 7 or more days without activity.

More output. Less throughput. More chaos. That is the AI productivity paradox.

#The 7 Capabilities That Actually Matter

DORA identified seven organizational capabilities that determine whether AI helps or hurts. This is the playbook.

#1. Clear and Communicated AI Stance

Your team needs explicit policies about what AI tools are acceptable, how they should be used, and where the boundaries are. Not a vague "we encourage innovation" statement. A real stance.

Teams without clear AI policies end up with developers making individual decisions about what to automate, which tools to use, and how much to trust AI output. That is how you get 31% more PRs merging with no review.

#2. Healthy Data Ecosystems

AI is only as good as the data it operates on. DORA found that organizations with high-quality, accessible, unified internal data get dramatically better results from AI tools.

If your data is scattered across 12 systems with no single source of truth, AI will confidently generate wrong answers faster than your team generated them manually.

#3. AI-Accessible Internal Knowledge

Connected documentation, searchable codebases, indexed decision logs. The teams that made their institutional knowledge machine-readable before plugging in AI saw compounding returns.

Most teams skip this step. They hand AI a codebase with zero documentation and wonder why the suggestions are generic.

#4. Strong Version Control Practices

With AI generating 98% more pull requests, your version control practices better be solid. DORA calls this critical for managing the increased code velocity that AI creates.

If your branching strategy was already messy, AI just turned it into a merge conflict factory.

#5. Working in Small Batches

Small batch delivery improves product performance and reduces friction. This is classic DORA, but it matters even more with AI because AI-generated PRs tend to be 51% larger.

Teams that enforce small batch discipline keep AI output manageable. Teams that do not end up with massive, unreviewable PRs that slip through with zero human oversight.

#6. User-Centric Focus

This one is critical. DORA found that a user-centric development focus amplifies AI's positive influence on team performance. Teams without it experience negative impacts.

AI can generate code incredibly fast. But if nobody is asking "does this actually solve a user problem," you are just shipping features nobody wants at unprecedented speed.

#7. Quality Internal Platforms

90% of organizations now have platform engineering capabilities. DORA found a direct correlation between platform quality and an organization's ability to unlock AI value.

The platform is the distribution layer. A great platform scales AI productivity gains across every team. A bad platform bottlenecks everything, regardless of how good the AI tools are.

#The Seven Team Archetypes

DORA also mapped teams into seven distinct profiles. Knowing where your team sits tells you exactly what to fix before expecting AI to help.

The Harmonious High-Achievers make up about 20% of teams. They excel across all dimensions - performance, stability, and well-being. For these teams, AI is pure fuel.

Pragmatic Performers, another 20%, deliver high speed and stability with moderate engagement. AI works well here too.

Stable and Methodical teams (15%) produce high-quality work at a sustainable pace. AI accelerates them without breaking their rhythm.

Constrained by Process teams (17%) have efficient processes but waste consumes their capacity. AI amplifies both the efficiency and the waste.

Legacy Bottleneck teams (11%) are constantly reacting to unstable systems. AI just creates more instability for them to react to.

High Impact, Low Cadence teams (7%) produce quality work but deliver slowly. AI can help with throughput but only if the cadence bottleneck is addressed.

Foundational Challenges teams (10%) are in survival mode. AI is the last thing they need. Fix the basics first.

#The Playbook

If you are an engineering leader about to roll out AI tools, or wondering why your existing rollout is not delivering the promised gains, here is what DORA says to do.

First, audit your foundations. Score yourself honestly against the seven capabilities. If you are weak on more than two, AI adoption will make things worse before it makes them better.

Second, identify your team archetype. If you are in the bottom three (Legacy Bottleneck, Foundational Challenges, Constrained by Process), fix your engineering fundamentals before investing more in AI tools.

Third, make your knowledge machine-readable. Document your architecture. Index your codebase. Connect your decision logs. This is the highest-leverage prep work you can do.

Fourth, enforce small batches ruthlessly. AI wants to generate big PRs. Your process needs to break them down. Automate the enforcement if you have to.

Fifth, establish your AI stance in writing. What tools are approved. What review standards apply to AI-generated code. Where human judgment is non-negotiable. Put it on paper.

The teams getting 33% more output with fewer incidents did not get there by installing Copilot. They got there by being good engineering organizations that happened to add AI.

#Read the Report

The full 2025 DORA State of AI-assisted Software Development report is available from Google. If you lead a team that writes software, it is required reading.

Related on imiel.dev: PwC found the same pattern - 74% of AI's economic gains go to 20% of companies. The winners are not spending more. They are executing better. And if you want to audit whether your own codebase is production-ready, The Last Prompt runs a 17-category review on any project.