AI impact on DORA metrics reveals a striking productivity paradox: individual developers merged 98% more pull requests while organizational software delivery performance remained essentially flat. The 2025 DORA Report—retitled “State of AI-assisted Software Development”—surveyed nearly 5,000 technology professionals and uncovered that AI tools amplify existing team capabilities rather than universally improving delivery metrics.
This article covers the 2025 DORA Report findings, the seven team archetypes that replaced traditional performance tiers, and practical measurement strategies for engineering leaders navigating AI adoption. The target audience includes VPs and Directors of Engineering responsible for measuring AI tool ROI, deployment frequency improvements, and overall engineering performance. Understanding why AI benefits vary so dramatically across teams has become essential for any organization investing in AI coding assistants.
Direct answer: AI acts as an amplifier that magnifies whatever work practices, cultural health, and platform maturity already exist in an organization. Strong teams see gains; teams with foundational challenges see their dysfunction worsen. This means engineering leaders must fix DORA metric baselines before expecting AI investment to deliver meaningful improvement.
By the end of this article, you will understand:
The 2025 DORA Report introduced a critical framing: AI acts as an “amplifier” or “multiplier” rather than a universal productivity booster. According to DevOps research conducted by Google Cloud, organizations with strong engineering systems, healthy data ecosystems, and mature internal platforms see positive gains from AI adoption. Organizations with weak foundations see those weaknesses magnified—higher change failure rate, more production failures, and increased rework.
AI adoption among software professionals surged to approximately 90% in 2025, up from roughly 75% the previous year. Most professionals now use AI tools daily, with median usage around two hours per day. Over 80% report improved individual productivity, and roughly 59% report improved code quality. Yet these perception-based gains don’t translate uniformly to organizational performance—the core insight that defines the AI era for engineering teams.
The DORA framework historically tracked four core metrics—Change Lead Time, Deployment Frequency, Change Failure Rate, and Mean Time to Recovery—as the foundation for measuring software delivery performance. These four metrics were used to categorize teams into different performance levels and benchmark improvement areas. In 2024, the DORA framework evolved to include five metrics, adding Deployment Rework Rate and removing the elite/high/medium/low performance tiers that defined earlier reports.
Throughput metrics now include:
Instability metrics include:
The addition of Rework Rate acknowledges that failures aren’t always outright rollbacks. Many disruptions are remediated via additional fixes, and tracking this provides a more complete picture of delivery stability. New metrics added to the DORA framework include Deployment Rework Rate and measures of AI Code Share, Code Durability, and Complexity-Adjusted Throughput.
Deployment Rework Rate measures the frequency of unplanned deployments required due to production issues.
AI Code Share tracks the proportion of code generated by AI tools.
Code Durability assesses how long code survives without major rework.
Complexity-Adjusted Throughput accounts for the complexity of changes when measuring delivery speed.
This evolution directly addresses AI-era challenges where AI-generated code may increase deployment volume while simultaneously creating quality assurance burdens downstream. Lead Time for Changes can drop initially as AI accelerates code writing, but bottlenecks may shift to code review, increasing the review time significantly. Tracking code that survives without major rework over time is also important for understanding long-term stability.
Research shows that platform engineering stands out as the primary enabler of successful AI adoption. Approximately 90% of organizations have adopted at least one internal developer platform, and 76% have dedicated platform teams. High-quality internal platforms correlate strongly with AI amplification benefits—teams can move faster because CI/CD pipelines, monitoring, version control practices, and developer experience infrastructure absorb the increased code velocity AI enables, especially when they already understand the importance of DORA metrics for boosting tech team performance.
Without strong platforms, AI tools’ output creates chaos. More committed code flowing through immature pipelines leads to bottlenecks in code review, longer queues, and ultimately more deployments fail. The DORA AI capabilities model emphasizes that platform prerequisites must exist before AI adoption can translate individual developer productivity into organizational outcomes.
This connection between foundational capabilities and the productivity paradox explains why some high performing teams thrive with AI while others struggle.
The productivity paradox represents the most significant finding from 2025: individual developers produce dramatically more output, but engineering teams don’t see proportional improvements in delivery speed or business outcomes. Faros AI, analyzing telemetry from over 10,000 developers, quantified this gap with precision that survey data alone cannot provide, which underscores both the strengths and pros and cons of DORA metrics for continuous delivery.
At the individual level, AI assisted coding delivers measurable improvements:
Individual developers report that AI coding assistants help them code faster, produce better documentation, and move through routine tasks with less friction. These gains are real and substantial. The challenge is that individual productivity improvements don’t automatically flow through to organizational performance.
Despite the surge in individual output, Faros AI’s telemetry revealed that organizational delivery metrics—deployment frequency, lead time, and the ability to quickly restore service after incidents (recovery speed)—showed no noticeable improvement. The traditional DORA metrics remained essentially flat across their sample.
Worse, several quality and efficiency signals degraded:
This data reveals where AI benefits evaporate: somewhere between individual contribution and organizational delivery, bottlenecks absorb the productivity gains. The complete picture shows AI helps individual developers produce more, but without corresponding improvements in review processes, pipeline efficiency, and quality assurance, that output creates downstream burden rather than business outcomes and often surfaces as classic signs of declining DORA metrics.
The DORA AI capabilities model identifies seven foundational practices that determine whether AI adoption succeeds or fails at the organizational level:
Teams that score well on these seven capabilities convert AI adoption into real performance benefits. Teams lacking these foundations experience the amplifier effect negatively—AI magnifies their dysfunction rather than solving it.
The 2025 DORA Report replaced the traditional linear performance tiers (Elite, High, Medium, Low) with seven team archetypes. This shift reflects a more nuanced understanding that team performance is multidimensional—throughput matters, but so does instability, team health, valuable work time, friction, and burnout, which aligns with newer DORA metrics guides for engineering leaders that emphasize a broader view of performance.
The seven archetypes are built from multiple dimensions, which still rely on mastering core DORA metrics implementation:
Gene Kim and the DORA researchers developed this framework because teams with identical DORA metrics might have vastly different experiences and outcomes. A team deploying frequently with low failure rate but high burnout requires different interventions than one with the same metrics but healthy team dynamics.
Prioritize establishing basic CI/CD pipelines, test coverage, build quality, and simple rollback mechanisms. AI adoption before these foundations exist will amplify chaos.
Address technical debt, modularize monolithic systems, and create internal platforms to standardize processes. AI tools can help with code modernization, but platform investment must come first.
Identify process friction—reviews, decision bottlenecks, approval chains—and streamline or automate them. Adding AI-generated code to a team already drowning in review backlog makes things worse.
Guard against quality degradation by monitoring instability metrics closely. Success creates risk: as throughput increases, maintaining code quality and architecture discipline becomes harder.
Challenge: HR hierarchies define teams administratively, but actual collaboration patterns don’t match org charts. AI tool adoption may be high in one administrative group while the engineers actually working together span multiple groups.
Solution: Combine HR hierarchies with telemetry data to measure actual collaboration patterns. Track who reviews whose code, who co-authors changes, and where knowledge flows. This provides a more accurate picture of where AI adoption is actually impacting delivery.
Challenge: Developers move between teams, change roles, and contribute to multiple repositories. Attributing AI impact to specific teams or projects becomes unreliable.
Solution: Track AI-influenced code contributions across team boundaries with proper tooling. Engineering intelligence platforms like Typo can measure AI-influenced PR outcomes with verified data rather than relying on license adoption estimates or self-reported usage, which is critical when implementing DORA DevOps metrics in large organizations.
Challenge: Traditional DORA metrics don’t distinguish between AI generated code and human-written code. You can’t assess whether AI is helping or hurting without this visibility.
Solution: Layer AI adoption rate, acceptance rates, and quality impact on traditional DORA metrics. Track:
Challenge: AI productivity gains evaporate somewhere in the delivery pipeline, but without end-to-end visibility, you can’t identify where.
Solution: Implement Value Stream Management to track flow from ideation through commit, review, QA, deploy, and post-release monitoring. This stream management approach reveals where time or defects accumulate—often in review queues or integration testing phases that become bottlenecks when AI dramatically increases deployment frequency upstream, and it depends on accurately measuring DORA metrics across the pipeline.
The 2025 DORA Report confirms that AI amplifies existing team patterns rather than uniformly improving software delivery performance. Teams with strong DORA baselines, mature platforms, and healthy engineering cultures see AI benefits compound. Teams with foundational challenges see AI worsen their dysfunction. The productivity paradox—individual gains that don’t translate to organizational outcomes—will persist until engineering leaders address the bottlenecks between developer output and business value delivery.
Immediate actions for engineering leaders:
The window for action is approximately 12 months. Organizations that successfully integrate AI with strong DORA foundations will achieve meaningful improvement in delivery speed and quality. Those that add AI to broken systems will see competitive disadvantages compound as their instability metrics worsen while competitors pull ahead.
Related topics worth exploring: Value Stream Management for end-to-end visibility, DevEx measurement for understanding developer friction, and AI ROI frameworks that connect tool investment to business outcomes.
The 2025 DORA Report found that approximately 90% of developers now use AI tools, with over 80% reporting productivity gains at the individual level. The central finding is that AI acts as an amplifier—magnifying organizational strengths and weaknesses rather than uniformly improving performance.
The report introduced seven critical capabilities that determine whether AI benefits scale to organizational performance: governance clarity, healthy data ecosystems, AI-accessible internal data, strong version control practices, small-batch workflows, user-centric focus, and quality internal platforms.
Notably, DORA researchers found no correlation between AI adoption and increased developer burnout, possibly because developers feel more productive even when downstream organizational stress increases.
AI improves individual developer metrics but creates organizational delivery challenges. Teams with strong DORA baselines see amplified benefits; weak teams see amplified dysfunction.
Quality and stability signals often worsen despite throughput improvements. Faros AI telemetry showed bug rates increased approximately 9% and code review time increased 91% as AI-generated code volume overwhelmed review capacity.
Platform engineering maturity determines AI success more than tool adoption rates. Organizations with strong CI/CD pipelines, monitoring, and internal platforms convert AI productivity into delivery improvements. Organizations lacking these foundations see AI create more chaos.
Deployment frequency increases due to AI-generated code volume, but this may not reflect meaningful output. More deployments don’t automatically translate to faster value delivery if those deployments require rework or cause production incidents.
Lead time for changes reduces for individual contributions, but review bottlenecks increase as reviewers struggle to keep pace with higher code volume. The 91% increase in review time documented by Faros AI shows where individual lead time gains get absorbed.
Engineering leaders need to measure complexity-adjusted throughput rather than raw deployment counts. Failed deployment recovery time becomes a more critical metric than traditional MTTR because it captures the full cost of instability.
The seven team archetypes are: Harmonious High-Achievers, Pragmatic Performers, Stable and Methodical, Constrained by Process, Legacy Bottleneck, High Impact Low Cadence, and Foundational Challenges.
Each archetype requires different AI adoption strategies and measurement approaches. Multidimensional classification considers throughput, stability, team well-being, friction, and time spent on valuable work—not just the four traditional DORA metrics.
One-size-fits-all AI strategies fail because a Legacy Bottleneck team needs platform investment before AI adoption, while Constrained by Process teams need to streamline workflows first. Harmonious High-Achievers can adopt AI aggressively but must monitor quality degradation.
Engineering leaders should combine traditional DORA metrics with AI adoption rates and code quality indicators. This means tracking not just deployment frequency and lead time, but also AI-influenced PR outcomes, PR size trends, review time changes, and rework rate.
Track AI-influenced PR outcomes with verified data rather than license adoption estimates. Engineering intelligence platforms like Typo provide visibility into actual AI usage patterns and their correlation with delivery and quality outcomes, complementing high-level resources that keep DORA metrics explained with practical insights.
Implement Value Stream Management to identify where AI gains evaporate in the delivery pipeline. Often, review queues, integration testing, or deployment approval processes become bottlenecks that absorb individual productivity improvements before they translate to business outcomes.
Use engineering intelligence platforms to correlate AI usage with delivery metrics, quality signals, and developer experience indicators. This comprehensive measurement approach provides actionable insights that surface problems before they compound.
Implement engineering metrics, gather SDLC insights & set continuous improvement goals in your dev teams.
Try Live Demo
We're on a mission to build engaged, productive tech teams. Try it out for free!
Follow us on:
