DORA maps seven team profiles for AI-assisted software development

DORA’s 2025 research gives engineering leaders a way to assess AI adoption against team health, delivery performance and customer value. The model groups teams into seven profiles and identifies seven capabilities that help organizations gain more from AI-assisted development.

Nathen Harvey discussed the findings during his QCon London talk, “Scale High Performance, Not Dysfunction.” Harvey works with DORA, which studies software delivery and the team practices that support it. The research combines survey data with qualitative analysis.

The findings show a split in AI’s effects. Engineers reported higher individual effectiveness and better code quality as AI adoption increased. Teams also reported more delivery instability, including rollbacks and hot fixes after releases.

Harvey said the pattern can point to a gap between feature production and production outcomes. An engineer may focus on completing a change while another team handles the operational cost after deployment. AI can widen that gap by increasing the volume of code a team produces.

Seven team profiles

DORA identified seven distinct profiles in its 2025 data:

• Foundational challenges
• The legacy bottleneck
• Constrained by process
• High impact, low cadence
• Stable and methodical
• Pragmatic performers
• Harmonious high-achievers

Each profile combines measures such as team performance, product performance, individual effectiveness, friction and burnout. The profiles give leaders a starting point for deciding which constraint to address before expanding AI use.

The “constrained by process” profile shows the risk of improving one measure while leaving the wider system untouched. These teams report high burnout and friction, low individual effectiveness and limited time spent on valuable work. Their delivery instability remains low, yet their throughput also trails other groups.

A team in this position may need process and platform changes before it adds more AI tools. More generated code will not remove approval queues, unclear ownership or slow test environments.

Seven capabilities support better outcomes

DORA linked stronger AI outcomes with seven capabilities:

• A clear, communicated AI stance
• Healthy data ecosystems
• Internal data that AI tools can access
• Strong version control practices
• Small batches of work
• A user-centered focus
• Quality internal platforms

These capabilities connect AI adoption to the systems that govern software delivery. Teams need policies that explain acceptable AI use, data that models can access with proper controls and repositories that support review and traceability.

Small batches deserve special attention. AI assistants can produce large change sets and pull requests in a short time. Reviewers still need to understand each change, test its effect and confirm that the result solves the intended problem. A large pull request increases the cost of that work and makes defects harder to isolate.

Teams can ask an AI assistant to divide a large request into independent units, then review and deploy each unit through the existing delivery path. That approach preserves feedback from tests, reviewers and production while reducing the size of each decision.

AI increases the number of options

AI reduces the time needed to create a prototype. A team can explore several approaches during the time that one conventional prototype might have required. The extra capacity creates room for user research, design comparison and technical experiments.

Teams still need a method for choosing among those options. Harvey recommended building multiple candidates, testing them with users and discussing the results as a group. The process shifts attention from code volume to evidence about user needs.

A product team might use an assistant to produce two interface concepts, a service boundary proposal and a test harness. Engineers and product partners can then compare those artifacts against user behavior and operational constraints. The team gains value from the choices the work enables, not from the number of files the assistant creates.

Platforms can carry the policy burden

Internal platforms can make AI capabilities available across an organization while keeping complex controls in one place. A platform team might provide approved models, reusable workflows, security checks and deployment paths through a common interface.

That design can enforce organizational policy during development. For example, a platform can require security scanning, apply approved data-access rules or block a deployment that lacks required evidence. Developers use the capability without rebuilding the control for each project.

Platform teams also need to measure whether their services improve delivery. Adoption counts alone cannot show value. Teams should connect platform use with review time, change size, deployment outcomes, recovery work and user results.

A practical adoption path

DORA’s capabilities model gives teams a sequence for applying the research to their own setting. The first step involves measuring the current state. Teams can record delivery performance, change failure signals, review friction, time spent on valuable work and burnout indicators before they expand AI use.

The second step involves choosing a narrow priority. A team with long review queues may start with smaller pull requests and better repository guidance. A team with poor access to internal knowledge may improve data quality and retrieval controls. A team with repeated production incidents may strengthen tests, observability and release safeguards.

The third step involves testing one change over a defined period. Leaders can compare the results with the baseline and collect feedback from engineers, product partners and operations staff. The team should keep the change, revise it or remove it based on evidence.

DORA recommends treating its findings as hypotheses rather than universal prescriptions. Each organization has different customers, systems, compliance duties and staffing constraints. Teams can use the DORA research library and the AI capabilities model to frame experiments, then adapt the measures to their own delivery process.

The research points to a clear operating principle: AI can increase a team’s capacity, but the surrounding system determines where that capacity goes. Teams that pair AI adoption with small changes, usable internal data, sound version control, user feedback and platform guardrails have more ways to turn generated work into software that performs in production.