2025年AI辅助软件开发状况报告

2025 Contents Executive summary Key takeaway:AI is an amplifier The greatest returns on AIinvestment come not from thetools themselves, but from astrategic focus on the underlyingorganizational system: the qualityof the internal platform, the clarityof workflows, and the alignmentof teams. Without this foundation,AI creates localized pockets ofproductivity that are often lost todownstream chaos. In 2025, the central question fortechnology leaders is no longer ifthey should adopt AI, but how torealize its value. DORA’s researchincludes more than 100 hoursof qualitative data and surveyresponses from nearly 5,000technology professionals fromaround the world.1The researchreveals a critical truth: AI’s primaryrole in software developmentis that of an amplifier. Itmagnifies the strengths of high-performing organizations and thedysfunctions of struggling ones. Key findings Drawing on qualitative data and a global survey conducted between June 13 and July 21, 2025, this reportuncovers several key findings on the state of AI-assisted software development, including: AI adoption has become nearlyuniversal. The majority ofsurvey respondents (90%) useAI as part of their work andbelieve (more than 80%) it hasincreased their productivity.Yet a notable portion (30%)currently report little to no trustin the code generated by AI,indicating a need for criticalvalidation skills. Successful AI adoption requiresmore than just tools. Our newDORA AI Capabilities Modelidentifies seven foundationalpractices—including a clearAI policy, a healthy dataecosystem, and a user-centricfocus—that are proven toamplify the positive impact of AIon organizational performance. AI adoption now improvessoftware delivery throughput,a key shift from last year.However, it still increasesdelivery instability. This suggeststhat while teams are adaptingfor speed, their underlyingsystems have not yet evolved tosafely manage AI-accelerateddevelopment. Read more in the Exploring AI’srelationship to key outcomeschapter. Read more in the DORA AICapabilities Model chapter. Read more in the AI adoptionand use chapter. This year’s research identifiesseven distinct team profiles,from “harmonious high-achievers” to teams caught ina “legacy bottleneck,” offeringa new framework for targetedimprovement. Value stream management(VSM), the practice ofvisualizing, analyzing, andimproving the flow of workfrom idea to customer, actsas a force multiplier for AI,ensuring that local productivitygains translate into measurableimprovements in team andproduct performance. 90% of organizations haveadopted platform engineering,making a high-quality internalplatform the essentialfoundation for AI success. Read more in the Understandingyour software deliveryperformancechapter. Read more in thePlatformengineering chapter. Read more in the Value streammanagement chapter. The landscape of AI’s impact Estimated effect of AI adoption on keyoutcomes, with 89% credible intervals For outcomes in orange, such as Burnout,a negative effect is desirable.Figure 1: The landscape of AI’s impact Analysis and advice fortechnology leaders Broad AI adoption withhealthy skepticism Successful AI adoptionis a systems problem,not a tools problem Quality platforms unlockAI’s value Platform engineering is now nearlyuniversal (90% adoption). Ourdata shows a direct correlationbetween a high-quality internalplatform and an organization’sability to unlock the value of AI.Organizations that treat theirplatform as an internal productdesigned to improve developerexperience see significantlygreater returns. While most developers use AIto increase productivity, thereis healthy skepticism about thequality of its output. This “trustbut verify” approach is a sign ofmature adoption. Our new DORA AI CapabilitiesModel reveals that the valueof AI is unlocked not by thetools themselves, but by thesurrounding technical and culturalenvironment. We’ve identifiedseven foundational capabilities—including a clear AI policy, ahealthy data ecosystem, a qualityinternal platform, and a user-centric focus—that are proven toamplify the positive impact of AIon performance. The conversation must shift fromadoption to effective use. Yourtraining programs should focus onteaching teams how to criticallyguide, evaluate, and validateAI-generated work, rather thansimply encouraging usage. Prioritize and fund your platformengineering initiatives. A poordeveloper experience andfragmented tooling may hamperthe impacts of your AI strategy. Treat your AI adoption as anorganizational transformation.The greatest returns will comefrom investing in the foundationalsystems that amplify AI’s benefits:your internal platform, yourdata ecosystem, and the coreengineering disciplines of yourteams. These elements are theessential prerequisites for turningAI’s potential into measurableorganizational performance. Seven profiles of teamperformance A systems view directs AI’spotential Simple metrics are not enough.We identified