DORA: AI Speeds Up Code Writing, but Code Review and Quality Demand New Team Practices
DORA finds that the time AI saves when writing code is often spent on verification. Here are three trade-offs and the team practices recommended to address them.
This article was prepared with AI assistance from the official sources listed below, and checked automatically before publication.
AI assistants for writing code have become part of the daily work of many development teams. According to the 2025 DORA State of AI-assisted Software Development report, 90% of technology professionals already use AI at work, and more than 80% feel their productivity has increased. However, in an article published on 10 March 2026, DORA cautions that the impact of AI on the software development life cycle (SDLC) is not a straight-line improvement. Written by Jessica Baolin and Nathen Harvey, the article outlines a number of trade-offs that teams and their leaders need to manage.
What DORA Studied
The researchers analysed 1,110 open-ended survey responses from Google software engineers in the third quarter of 2025. Respondents were asked to describe how AI-based tools had affected their workflow over the previous three months. Most of the feedback centred on four main activities:
- code generation,
- information retrieval,
- code review,
- testing.
DORA notes that AI managed to speed up work in all of the use cases analysed. AI helps automate boilerplate code, reduces the friction of starting new tasks, and bridges knowledge gaps, for example when engineers work with a programming language they have not mastered or with legacy code. DORA describes AI as an amplifier: it magnifies the strengths of organisations that already perform well, but it also magnifies the dysfunction of organisations that are struggling.
Three Trade-offs to Watch Out For
DORA summarises three main tensions in the use of AI in the SDLC:
- Time spent writing shifts to time spent verifying. The time saved when writing code is often spent again on checking the results. The 2025 DORA report notes that 30% of developers trust AI-generated code only a little or not at all. Hallucinations and verification burden appear in all of the use cases analysed.
- AI assistance can weaken skill building. Skill degradation was recorded as one of the negative themes, particularly in code generation and learning.
- Fast prototypes, slow production integration. Prototypes can be built very quickly, but bringing them to production-ready quality still takes time.
According to DORA, these tensions may explain their finding that higher AI adoption is associated with both a rise in software delivery throughput and a rise in delivery instability.
Common Mistakes in Adopting AI
From these findings, there are several pitfalls worth avoiding:
- Measuring output, not impact. The number of lines of code does not show whether software has become better or more stable.
- Cutting estimates too early. DORA asks leaders not to reduce estimates before investing in closing the gap between a quick prototype and production-equivalent quality.
- Relying on AI over a fragile foundation. According to DORA, if teams are wrestling with fragmented tooling, siloed data, or fragile infrastructure, AI will only help them "generate technical debt faster".
How Teams and Clients Can Apply This
DORA offers a number of practical recommendations for teams and leaders:
- Change how code review is done. Give AI feedback to code authors while they are still writing, use context-aware review agents to enforce standards, and work in small batches. Teams are also encouraged to reconsider whether asynchronous review is still the most optimal approach compared with investing in test automation.
- Measure impact. Google uses indicators of speed, ease and quality (speed, ease, quality/SEQ) for developer productivity. DORA also refers to the SPACE and H.E.A.R.T. frameworks and Value Stream Management.
- Adjust estimates realistically. Factor in verification and production integration time, not just the speed of building a prototype.
- Protect time for deep learning. Teams need to provide space so that engineers' skills, especially those of juniors, continue to grow.
For IT managers drafting AI usage policies, these DORA findings are a reminder that the speed of writing code is only one part of the bigger picture. For clients assessing vendor schedules, promises of AI-driven acceleration deserve further questions: how are the review process, testing and release stability safeguarded. The quality of software delivery still depends on strong team practices, with or without AI.
Sources & references
This article is a summary and interpretation by the Oemah Solution team of the sources below. Refer to the original sources for full details and context.


