Written by: Mark Hull, Co-Founder and CEO, Exceeds AI | Last updated: April 23, 2026
Key Takeaways for AI-Era Engineering Leaders
- DORA metrics like Deployment Frequency and Lead Time for Changes miss AI’s code-level impact, so dashboards show inflated throughput without quality context.
- Traditional delivery metrics stay blind to AI-generated code, cannot separate it from human work, and fail to track long-term effects like technical debt.
- GetDX adds surveys and workflow analytics but still lacks objective ROI proof, as PR throughput gains trail AI adoption rates.
- AI creates longer reviews, subtle bugs, and multi-tool complexity that aggregate metrics cannot break down into actionable insights.
- Exceeds AI brings code-level observability that proves AI ROI, from specific tools to concrete outcomes.
DORA Metrics in 2026: Useful, Yet Out of Sync with AI
DORA metrics—Deployment Frequency, Lead Time for Changes, Change Failure Rate, and Mean Time to Restore Service—measure delivery capability across software organizations. GetDX (getdx.com) supplements these with developer experience surveys and workflow analytics. Both frameworks provide valuable baseline measurements but miss critical AI-era dynamics.
These four core metrics reveal why traditional approaches struggle once AI enters the stack:
Deployment Frequency measures how often code deploys to production. Elite teams deploy multiple times per day, but AI inflates this frequency without quality context, so teams may ship more often while quietly adding technical debt.
This speed pressure flows into Lead Time for Changes, which tracks time from commit to production. Elite benchmarks stay under one hour, yet AI accelerates coding while review queues grow longer, so cycle time improvements stall or reverse.
Those review delays attempt to protect Change Failure Rate, the percentage of deployments that cause failures (elite: 0–15%). AI introduces subtle logic bugs that slip through review and appear later in production, which makes this metric less predictive than it was before AI.
When failures occur, Mean Time to Restore measures recovery speed. Elite performers restore service quickly, but AI-generated logic errors are harder to trace and fix, so MTTR often rises even as teams adopt more AI tools.
GetDX metrics add developer satisfaction surveys and workflow friction measurements. These capture sentiment but cannot prove whether AI investments deliver measurable business outcomes or identify which AI tools drive the strongest results.
These foundational issues become clearer when you examine the specific ways DORA metrics break down in AI-driven development environments.
7 Critical DORA Metrics Limitations in the AI Era
DORA metrics fail engineering leaders because they measure outcomes without understanding AI’s code-level impact. Here are the seven critical limitations:
- Lagging Indicators Miss Real-Time AI Impact: DORA metrics reflect system-level outcomes but cannot decompose AI’s specific contribution. Teams may use AI tools heavily, see coding productivity rise 40%, and still show flat deployment frequency because downstream bottlenecks hide the gain.
- Gaming Through AI Volume: Faros AI’s telemetry from over 10,000 developers found teams with high AI adoption showed notable increases in merged PRs per engineer. This surge floods review processes and artificially inflates throughput metrics, so leaders see more merged PRs without understanding whether value actually increased.
- Blind to Cognitive Load and Context Switching: DORA metrics cannot capture the mental overhead of juggling multiple AI tools or handling rapid task generation. Excessive context switching reduces deep work time, so teams may deliver less meaningful output and introduce more defects even as surface-level activity rises.
- Team Silos Hide Adoption Patterns: DORA aggregates team performance but cannot identify which engineers use AI effectively versus those struggling with adoption. This aggregation means high-performing teams can mask individual productivity gaps, so three engineers may excel with AI while two fall behind, and the blended metrics hide that 40% of the team is not benefiting from the AI investment.
- No Quality Depth for AI-Generated Code: Change Failure Rate measures deployment failures but misses AI-specific quality issues. GitClear’s analysis found AI-generated code can show higher churn rates compared to human-written code, which signals rework and instability that DORA does not expose.
- AI-Blind Measurement: DORA cannot distinguish between AI-assisted and human-written code contributions. This blindness makes it impossible to attribute productivity gains or quality issues to specific AI tools, prompts, or adoption patterns, so leaders cannot double down on what works.
- Rising CFR and MTTR from AI Complexity: Baytech Consulting’s analysis shows DORA’s Change Failure Rate rising as teams trade reliability for speed, while Mean Time to Restore increases because AI introduces subtle logic-based bugs that are difficult to trace. Traditional dashboards show worsening reliability without explaining that AI complexity drives the change.
GetDX Pitfalls as a DORA Supplement
GetDX metrics attempt to supplement DORA with developer experience surveys and workflow analytics, yet they introduce their own AI-era limitations. Survey-based measurements capture sentiment rather than objective outcomes. GetDX’s longitudinal study found PR throughput increased only 9.97% despite 65% average increase in AI usage, which highlights the disconnect between reported AI adoption and measurable delivery impact.
GetDX surveys cannot prove whether AI tools deliver ROI or identify which specific AI adoption patterns drive results. They measure developer satisfaction with AI tools but miss code-level outcomes like quality degradation, technical debt accumulation, or long-term maintainability issues.
Given these limitations in both DORA and GetDX, engineering leaders need to know whether these frameworks still deserve a place in their measurement stack.
Are DORA Metrics Still Relevant in 2026?
DORA metrics remain foundational for measuring delivery capability, but they require an AI-specific intelligence layer to stay relevant. DORA metrics resist gaming more than activity metrics and have been validated across thousands of organizations, so they still provide valuable baseline measurements when you pair them with AI impact data.
AI-Era Gaps DORA and GetDX Miss
The most critical gaps come from AI’s code-level impact that metadata-only tools cannot see. AI coding tools generate 26.9% of production code, yet traditional metrics cannot distinguish this AI-generated code from human contributions or track its long-term outcomes.
Technical debt accumulation has become a hidden crisis. AI tools can generate code that passes initial review but introduces maintainability issues that surface weeks later. The Faros report analyzing 50,000+ pull requests found AI-generated code resulted in 91% longer pull request review times as reviewers apply higher scrutiny.
Multi-tool chaos compounds the problem. Teams use Cursor for feature development, Claude Code for refactoring, GitHub Copilot for autocomplete, and other specialized tools for niche workflows. DORA and GetDX metrics cannot provide aggregate visibility across this AI toolchain or compare tool-by-tool effectiveness.
ROI causation also remains invisible. METR’s controlled experiment found experienced developers were 19% slower on complex tasks when using AI due to higher verification costs, yet traditional metrics cannot capture this nuance or identify when AI adoption actually degrades performance.
These gaps demand a different approach that operates at the code level instead of relying on metadata and surveys.
The Code-Level Fix: AI-Impact Observability with Exceeds AI
Exceeds AI addresses DORA and GetDX limitations through repo-level observability that separates AI from human code contributions. Metadata-only tools like Jellyfish or LinearB track PR cycle times and commit volumes, while Exceeds AI analyzes actual code diffs to identify which specific lines are AI-generated and then tracks their outcomes over time.
AI Usage Diff Mapping provides granular visibility into AI adoption patterns. Instead of aggregate statistics, engineering leaders see exactly which 623 of 847 lines in PR #1523 were AI-generated, which tools created them, and how those lines performed compared to human-written code in the same pull request. This diff-level clarity sets the foundation for deeper outcome analysis.
AI vs. Non-AI Outcome Analytics then quantifies ROI at the commit and PR level. Teams compare cycle times, review iterations, test coverage, and long-term incident rates between AI-assisted and human-only code contributions. Leaders move from subjective surveys to concrete evidence that shows where AI helps, where it hurts, and which practices deserve scaling.
Longitudinal outcome tracking tackles the hidden technical debt crisis by monitoring AI-touched code over 30 days or more. This tracking reveals whether AI-generated code that passes initial review later causes production incidents, requires extra maintenance, or introduces security vulnerabilities that traditional metrics overlook.
Unlike Jellyfish, which has a 2-month setup time and commonly takes 9 months to show ROI, or DX’s survey-heavy approach, Exceeds AI delivers insights within hours through simple GitHub authorization. The platform works across all AI tools, including Cursor, Claude Code, Copilot, and Windsurf, and provides tool-agnostic detection with outcome comparison.
Exceeds AI Implementation Blueprint for Fast ROI
Implementation requires minimal overhead compared to traditional developer analytics platforms. The process starts with GitHub OAuth authorization, which takes about 5 minutes to grant the necessary permissions. After authorization, you spend roughly 15 minutes selecting repositories and defining the pilot scope, which determines what data Exceeds AI will analyze first.
This scoped setup enables the platform to deliver initial insights within one hour, including early AI usage patterns and outcome comparisons. Complete historical analysis typically finishes within 4 hours, while metadata-only competitors often need months of onboarding before they show comparable value.
Security follows enterprise standards with no permanent source code storage, real-time analysis that fetches code only when needed, and SOC 2 compliance in progress. Data encryption covers rest and transit, and enterprise customers can access audit logs and penetration testing results.
Start your free pilot to get code-level AI impact insights within hours.
FAQ: Answering Key Objections
Why Choose Repo Access Over Metadata Tools?
Metadata-only tools like Jellyfish, LinearB, and GetDX cannot distinguish AI-generated from human-written code, which makes AI ROI measurement guesswork. Without repo access, you might see PR cycle times drop 20%, yet you cannot prove AI caused the improvement or identify which adoption patterns worked. Code-level fidelity delivers authentic ROI measurement and actionable insights that metadata alone cannot provide.
How Does Exceeds AI Compare to Jellyfish and GetDX?
Exceeds AI provides AI-native intelligence, while Jellyfish focuses on financial reporting and GetDX measures developer sentiment. Jellyfish’s lengthy onboarding and ROI timeline (mentioned earlier) stem from its metadata-only approach, which cannot prove AI impact at the code level. GetDX relies on subjective surveys rather than objective code analysis. Exceeds AI delivers code-level proof of AI ROI in hours, with guidance for scaling adoption instead of static dashboards.
Does Exceeds AI Support Multiple AI Tools?
Exceeds AI uses tool-agnostic detection to identify AI-generated code regardless of which tool created it. The platform works across Cursor, Claude Code, GitHub Copilot, Windsurf, and other AI coding tools, providing aggregate visibility and tool-by-tool outcome comparison. This breadth matters because teams rely on multiple AI tools for different workflows, and leaders need comprehensive ROI measurement across the entire AI toolchain.
What About Setup Time and AI Detection Accuracy?
Setup takes hours, not weeks or months like traditional developer analytics platforms. GitHub authorization and initial data collection complete within about one hour, and full historical analysis becomes available within roughly 4 hours. AI detection uses multi-signal analysis, including code patterns, commit message analysis, and optional telemetry integration, to reach high accuracy across languages and frameworks.
Are DORA Metrics Still Relevant with Exceeds AI?
DORA metrics remain valuable as foundational delivery measurements, yet they need an AI-specific intelligence layer to stay useful in 2026. Exceeds AI complements DORA by adding the code-level context required to understand AI’s impact on delivery outcomes. Teams use both together, with DORA for baseline capability and Exceeds AI for AI-specific ROI proof and adoption guidance.
Conclusion: Move Beyond Blind Spots and Prove AI ROI
DORA and GetDX metrics served engineering leaders well in the pre-AI era, but they cannot capture the code-level reality of AI’s impact on software development. With more than a quarter of production code now AI-generated, leaders need observability that separates AI from human contributions and proves ROI at the commit and PR level.
The path forward does not require abandoning DORA metrics. Instead, leaders should supplement them with AI-native intelligence. Exceeds AI provides that missing layer, with code-level observability that shows which AI tools work, surfaces adoption patterns that drive results, and helps leaders scale effective practices across teams.
Engineering leaders can finally answer executives with confidence: “Yes, our AI investment is paying off, and here is the evidence.” See how Exceeds AI analyzes your codebase to prove AI ROI.