AI-Assisted Software Development: Team Profiles and Capabilities for Putting Research into Action
13:30 · July 30, 2026 · RSS APP - AI Primary Research

AI is an amplifier; strategic focus on the organizational system brings the greatest returns. DORA's 2025 research on AI in software development provides team profiles and success capabilities that can be used to put the research into practice.
Summary
DORA’s 2025 research on AI-assisted software development frames AI primarily as an amplifier that magnifies both strengths and weaknesses in existing team and organizational practices. According to Nathen Harvey, who presented the findings at QCon London, the largest gains come from deliberate attention to the underlying system rather than isolated tool adoption. Without that foundation, localized productivity improvements often dissipate into downstream instability and rework.
The study distinguishes seven team profiles based on metrics that include throughput, stability, individual effectiveness, friction, and burnout. One profile, labeled “constrained by process,” shows high burnout and friction alongside low individual effectiveness, yet relatively stable delivery at the cost of limited throughput. The other profiles range from legacy-constrained teams to high-achieving groups that combine strong performance with sustainable practices.
Alongside the profiles, researchers identified seven capabilities that, when paired with AI use, correlate with better outcomes. These include a clear organizational stance on AI, healthy and accessible data ecosystems, strong version-control hygiene, working in small batches, sustained user-centric focus, and quality internal platforms. The last two items prove especially relevant: platforms can enforce policy and hide complexity, while small-batch workflows counteract the tendency of AI models to generate large, hard-to-review change sets.
The data also reveal a notable tension. Higher AI adoption tracks with improved self-reported individual effectiveness and perceived code quality, yet it simultaneously raises software-delivery instability, measured by increased rollbacks and hotfixes. Harvey suggests this pattern may reflect misaligned incentives when engineers optimize for feature output without visibility into production impact.
For teams seeking to apply the findings, Harvey recommends assessing their current profile, selecting a small number of capabilities to strengthen, and treating the report results as hypotheses to test within their own context rather than universal prescriptions.
Why it matters
This article provides actionable, research-backed insights into the organizational and team-level impacts of AI coding assistants. It is highly relevant for Dutch AI practitioners and engineering leaders looking to optimize AI adoption and mitigate risks like delivery instability in enterprise environments.







