What is the DevOps Research and Assessment (DORA) framework?

Generative AI adoption helps individual developers write code faster. However, localized acceleration paradoxically drives down team-wide software delivery throughput and stability. Engineering leaders relying on outdated measurement dashboards often fail to see the bottleneck until AI-generated technical debt breaks production environments. They assume more commits automatically mean faster overall value creation.

The DevOps Research and Assessment (DORA) framework evolved into a five-metric model specifically to expose modern instability. Managing the shift requires your team to move from passively tracking metrics to actively enforcing them through automated pipeline observability. The following guide covers the updated metrics, the AI productivity paradox, and how to use telemetry to protect deployment quality.

If you are searching for Digital Operational Resilience Act (DORA), the European financial cybersecurity mandate that applies to more than 22,000 financial entities and requires strict risk management standards by January 17, 2025, please refer to our guide to EU DORA compliance.

TL;DR

• The DORA framework is no longer just "four keys" and now functions as a five-metric model categorized into delivery throughput and instability.
• The official guidelines retired ambiguous acronyms like MTTR in favor of failed deployment recovery time and introduced deployment rework rate.
• High AI reliance led to a 7.2 percent reduction in delivery stability across teams, while a 25 percent increase in AI adoption generally correlates with declining throughput.
• Elite performance requires moving beyond static dashboards to capture software development lifecycle telemetry, deployment events, and testing data natively.
• Integrating automated quality gates directly into the continuous delivery pipeline prevents fast AI-assisted code from increasing your change failure rate.

Originating from a landmark research program that has collected insights from over 40,000 professionals over nearly a decade, the foundational DORA metrics changed how industry leaders viewed continuous software delivery. Many legacy vendors still treat the resulting framework as a static set of four basic indicators used primarily to categorize teams into High, Medium, and Low performers. Taking a narrow view masks modern delivery bottlenecks and turns valuable continuous improvement insights into a punitive management tool.

Focusing strictly on pipeline capabilities represents the true core intent of the framework. In 2025, the DORA organization officially de-acronymed the name and expanded its research scope beyond basic DevOps terminology to cover developer experience and platform engineering alongside AI-assisted development. For example, the 2024 DORA report highlighted that adopting platform engineering can cause a temporary dip in performance before benefits are realized. Because the original four metrics failed to capture such modern nuances, the official organization updated the measurement model.

Moving beyond legacy definitions, the official DORA guidelines expanded to five metrics as of January 2026. Evaluating software delivery now requires categorizing the measurements into two distinct buckets: throughput and instability. The updated structure provides a much clearer picture of pipeline health by retiring ambiguous terms and focusing on specific operational realities that actively affect daily engineering work.

Decades of research consistently demonstrate that delivery speed and stability represent linked outcomes beyond mutually exclusive tradeoffs. Throughput measures how quickly and frequently your team delivers code into production environments. It incorporates change lead time, deployment frequency, and failed deployment recovery time. The organization officially introduced the latter term to replace Mean Time to Recovery, eliminating widespread confusion about whether a recovery action meant rolling back a change or physically rebuilding the underlying infrastructure.

Instability measures the friction introduced by code changes. It evaluates change fail rate alongside deployment rework rate. Historical data repeatedly validates that engineering teams can improve both functional categories simultaneously when they apply continuous delivery practices diligently. Tracking the five areas isolates precisely where technical debt slows down momentum.

Adding the final indicator of software quality sparked ongoing industry debate. The official research site designates deployment rework rate as the essential fifth metric, measuring the percentage of deployments requiring immediate hotfixes or pipeline rollbacks. Conversely, many prominent vendors advocate for a broader reliability metric based on general availability, performance degradation, and data accuracy.

Choosing between rework rate and overall reliability depends heavily on your specific operational priorities. Rework rate targets specific software development lifecycle inefficiencies, whereas reliability tracks extended infrastructure health. Elite performers using accurate capability measurements are twice as likely to meet organizational goals compared to low performers. Tracking the chosen metric establishes a strong baseline and exposes the friction caused by an influx of AI-generated code.

Even though localized AI coding tools rapidly increase individual developer output, they actively erode team-wide delivery capabilities. Bottlenecks occur because rapid code creation safely outpaces available automated testing constraints at scale. Short-term bursts of local productivity ultimately overwhelm the delivery pipeline, resulting in slower overall shipping times and higher failure rates.

Applying generative tools offers immediate individual benefits. Generative AI boosts individual documentation quality by 7.5 percent and code review speed by 3.1 percent. Yet the 2024 DORA report reveals that a 25 percent increase in team AI adoption correlates with estimated declines in delivery throughput and stability. In reality, high AI reliance has led to a 7.2 percent reduction in delivery stability and a 1.5 percent reduction in delivery throughput across many organizations. Writing basic lines faster means nothing if you cannot successfully deploy them.

Consider a highly relatable scenario inside modern engineering environments. A mid-market software team rolls out an enterprise AI coding assistant in January. By March, individual code commits double across the organization. By June, the massive influx of unvalidated code causes the testing environment to constantly crash, integration tests stretch to four hours long, and the change failure rate spikes. Developers blame the deployment infrastructure, operations blames the unreviewed AI suggestions, and collaboration grinds to a halt. Since AI-generated code introduces hidden friction at a vast scale, engineering leaders need to move beyond merely measuring outcomes and start actively enforcing strong pipeline safeguards.

Tracking deployment frequencies manually creates an operational dead end. You cannot permanently fix throughput issues if engineering managers spend three hours pulling chronological ticket timestamps to build a weekly reporting spreadsheet. Elite performance builds on continuous delivery, deployment automation, secure code analysis, and dedicated observability. Isolated measurement dashboards are no longer enough.

To evade manual tracking pitfalls, your organization needs a telemetry layer that ingests data directly from the tools developers already use daily. Dynatrace natively captures Software Development Lifecycle events to provide necessary pipeline observability. The platform pulls traces and commit signals directly from your primary integration systems, capturing test execution times simultaneously. Gathering the underlying data automatically prevents internal teams from arguing over reporting accuracy and establishes a single source of truth.

Establishing deep visibility represents a fundamental shift in software delivery management. Centralizing telemetry data directly transitions teams from passively watching performance graphs decline to proactively protecting production environments from hazardous code changes.

Applying deep pipeline telemetry allows you to power release validation and reject flawed code before it reaches the end user. Validating proposed changes through continuous delivery tools and observable guardrails naturally drives down overall instability metrics. When a specific artifact fails predefined service-level objectives during the testing phase, the pipeline halts progression instantly.

Organizations implement Dynatrace Site Reliability Guardian to enforce automated release checkpoints actively. Measuring initial code speed holds very little value unless the underlying application functions flawlessly upon release. For instance, Clario uses telemetry-driven safeguards to stop unstable software builds from reaching production in highly regulated clinical trial environments. Their deployment frequency remains consistent despite heavy compliance requirements because the automated validation layer supports the risk mitigation natively. Evaluating releases programmatically ensures you maintain development velocity while neutralizing operational risks.

The modern measurement framework functions as an operational standard and proves that fast code generation yields poor results if the pipeline cannot stably deliver it. Realizing the full promise of continuous delivery requires an underlying shift from passive performance measurement to full automated enforcement. By orchestrating reliable software delivery automation, Dynatrace provides the necessary observability to catch deep instability right where it begins. The engineering teams that win tomorrow will prioritize deployment systems that effectively handle velocity safely over processes that prioritize pushing unverified code out faster.

The DevOps Research and Assessment program measures software delivery performance, while the Digital Operational Resilience Act is an EU cybersecurity compliance mandate for financial institutions. The European regulation applies to more than 22,000 financial entities and their third-party service providers, requiring them to establish strong resilience standards by January 2025.

The updated framework categorizes measurements into throughput and instability groups. Throughput metrics track change lead time, deployment frequency, and failed deployment recovery time. Instability tracks both change fail rate and deployment rework rate.

The research organization updated its terminology to be more precise and actionable for engineering teams. As of January 2026, the official guidelines discarded ambiguous legacy acronyms like MTTR in favor of failed deployment recovery time to clearly reflect throughput recovery events.

Artificial intelligence adoption creates a productivity paradox where localized individual gains introduce pipeline bottlenecks that lower overall team performance. The 2024 DORA report notes a 25 percent increase in AI adoption correlates with estimated declines in delivery throughput and stability. Specific measurements show high AI reliance has led to a 7.2 percent reduction in delivery stability and a 1.5 percent reduction in delivery throughput.

Relying on a distinct dashboard is less effective than extracting telemetry directly from the software development lifecycle to calculate pipeline health natively. Elite performance demands validated capabilities spanning continuous delivery and observability. These integrated systems allow teams to pull data automatically, eliminating the need for manual timestamp tracking.