Navigating AI Adoption in Software Development
All images and references from: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report
AI is rapidly transitioning from a nascent technology to a nearly universal and foundational component of the software development toolkit. In 2025, the central strategic challenge for technology leaders is no longer whether to adopt AI, but how to strategically realize its organizational value. Drawing on a global survey of nearly 5,000 technology professionals, the DORA research underscores a critical realization: AI functions primarily as an amplifier, magnifying the strengths of high-performing organizations while exposing the gaps of those that struggle.
This report details the adoption trends, the paradoxical impacts on organizational performance, and the required systemic capabilities necessary to successfully transform development practices for the AI era.
1. AI Adoption and Deep Reliance
AI adoption among software development professionals has reached 90%, reflecting a significant 14.1% increase from the previous year, suggesting AI use has become the standard. Developers are integrating AI into their daily routines, reporting a median of two hours spent interacting with AI tools on a recent workday, which is approximately one-quarter of an eight-hour workday.
This widespread usage is coupled with profound reliance across core development tasks:
Reliance Levels: A strong majority (65%) of AI users report relying on AI for a moderate amount, a lot, or a great deal of their work.
Core Applications: The single most common use for AI tools is writing new code, leveraged by 71% of respondents whose jobs involve coding. Other frequent applications include literature reviews (68%), modifying existing code (66%), and general proofreading (66%).
The individual benefits of this adoption are largely positive, creating self-assessed gains in efficiency and quality:
Productivity Gains: Over 80% of surveyed professionals perceive that AI has increased their individual effectiveness or productivity.
Code Quality: A majority (59%) report a positive impact on code quality. Developers observe that AI sometimes writes better code for specific tasks and often adheres to coding standards that they might otherwise forget.
2. The Paradox of Trust and Instability
Despite the clear gains in individual output, the path to organizational value remains complex, marked by a paradox of trust and persistent instability.
The Trust Paradox: Trust But Verify
High adoption rates coexist with a measured caution regarding AI output quality. While 70% of respondents express some degree of confidence, a notable portion (30%) report having little ("a little") or no trust at all in AI-generated output quality. This pattern indicates a "trust but verify" approach, suggesting developers apply healthy skepticism to AI output, similar to how they vet information from widely used resources like Stack Overflow. Absolute trust is not a prerequisite for AI-generated outputs to be useful.
Instability and Organizational Lag
The shift in AI's relationship with software delivery performance demonstrates organizational adaptation but highlights persistent systemic issues. A key shift from last year’s findings is that AI adoption is now linked to higher software delivery throughput (speed). However, AI adoption still increases delivery instability (unreliability).
This suggests that while developers are adapting and maximizing speed, the underlying organizational systems—the processes and technical capabilities—have not yet fully evolved to safely manage this AI-accelerated volume of changes. Outcomes like friction and burnout, which are consequences of the socio-technical system, show no measured relationship with AI adoption, underscoring that these issues reside beyond the individual developer's control.
3. Thinking Strategically: AI as a Systems Problem
The research reveals that the greatest returns on AI investment depend not on the tools alone, but on a strategic focus on the underlying organizational system, including platform quality, workflow clarity, and team alignment. Successful AI adoption must be treated as an organizational transformation effort.
The DORA AI Capabilities Model identifies seven foundational practices proven to amplify the positive impact of AI on organizational performance and individual effectiveness.
Clear and Communicated AI Stance - Amplifies positive influence on individual effectiveness and organizational performance, and decreases friction.
Healthy Data Ecosystems - Amplifies positive influence on organizational performance.
AI-accessible Internal Data - Amplifies positive influence on individual effectiveness and code quality.
Strong Version Control Practices - Frequent commits amplify individual effectiveness. Frequent rollbacks amplify team performance.
Working in Small Batches - Amplifies product performance and decreases friction, even if individual effectiveness gains are slightly reduced.
User-centric Focus - Amplifies positive influence on team performance. AI adoption negatively impacts team performance in its absence.
Quality Internal Platforms - Amplifies positive influence on organizational performance.
Value Stream Management (VSM), the practice of visualizing, analyzing, and improving the flow of work from idea to customer, acts as a force multiplier for AI investments. VSM ensures that localized AI productivity gains are applied to the system's biggest constraint, translating individual improvements into measurable organizational advantages rather than creating downstream chaos.
4. Relevance for Life Science and Regulated Industries
The DORA survey included respondents from the Healthcare & Pharmaceuticals industry. While the research does not provide specific industry-level analysis, the core finding that AI is an amplifier and especially with specific conditions has profound implications for the life science sector. Here are some of our thoughts on the top three impacts of AI adoption in software development in life sciences:
Given that AI adoption increases software delivery instability, organizations in life sciences must prioritize technical excellence as a safety net. The observed instability suggests that accelerating code generation without strengthening foundational practices like automated testing, frequent commits, and rollback capabilities creates an unacceptable risk to service reliability and long-term sustainability.
In an industry requiring strict security, compliance, and architectural standards, a quality internal platform is the essential foundation for AI success. The platform acts as the distribution and governance layer, ensuring AI benefits scale securely and effectively across the organization. It provides the necessary guardrails to prevent inappropriate use, which can justify a manageable increase in friction.
The finding that AI adoption can harm team performance if a user-centric focus is absent is especially important. Life science teams must ensure that AI-supported development is guided by user needs (e.g., clinicians, researchers, patients) to clarify goals and orient strategy, preventing teams from moving quickly in the wrong, potentially non-compliant or non-valuable direction.
Recommendations for Improving Organizational AI Adoption
For senior leadership teams aiming to unlock the organizational value of AI and mitigate the associated risks, investment must shift from tool procurement to systems transformation. These recommendations are based on cultivating the seven DORA AI Capabilities:
Clarify Your AI Stance to Build Trust and Reduce Friction
Establish and communicate a clear policy regarding which AI tools are permitted and the expectations for their use (the clarity of the stance matters more than the content itself).
This clarity reduces the risk of developers acting too conservatively (fear of overstepping) or too permissively (using tools inappropriately), thereby building developer trust, reducing friction, and increasing AI’s positive influence on performance.
Invest in Data Quality and AI Accessibility
Treat internal data as a strategic asset by investing in the quality, accessibility, and unification of internal data sources. Simultaneously, invest the engineering effort to securely connect AI tools to internal company information, such as documentation and codebases.
A healthy data ecosystem significantly increases AI’s positive influence on organizational performance. Connecting AI tools to internal context provides the company-specific relevance needed to maximize gains in individual effectiveness and code quality, moving beyond generic foundational models.
Improve Technical Safety Nets
Ensure teams are proficient in foundational engineering practices, especially strong version control. Encourage frequent code commits and ensure teams are skilled in using rollback and revert features.
These practices are critical safety nets for mitigating the instability caused by AI-augmented code generation. Frequent commits amplify individual effectiveness, and proficiency in rollback features is associated with better team performance, allowing teams to experiment with confidence.
Enforce Smaller Batch Sizes and Prioritize Product Performance
Implement the discipline of working in small batches. Break changes into manageable units that require fewer lines of code per commit and fewer changes per release.
While working in small batches may slightly reduce perceived individual effectiveness gains from generating large code dumps, it is essential for amplifying product performance and reducing systemic friction.
Build a Quality Internal Platform
Prioritize and fund platform engineering initiatives, treating the platform as a high-quality internal product designed to improve the developer and user experience. The platform must offer shared capabilities and necessary guardrails (like automated testing and deployment).
Center Your AI Strategy Around the End User:
Cultivate a deep user-centric focus where users' experience is the top priority and is understood as key to business success. Without it, AI adoption is likely to harm team performance. When present, it amplifies AI’s positive influence on team performance, ensuring accelerated development is aligned with meeting customer and organizational goals.
Join our membership to learn more about analysis like this and to join us for live conversations on these topics.
