How AI Spreads in Software Teams

AI adoption spreads through broad, repeated success and policy legitimizes rather imposes

AI adoption in software development rarely happens all at once. It spreads unevenly, usually beginning with a handful of developers who start using the tools.
Then the tools stop feeling experimental and become routine. Developers integrate them into coding, debugging, documentation, review, and testing.

That transition matters because AI adoption in software engineering follows a pattern: more use leads to familiarity, familiarity improves integration, and successful integration encourages broader use.

A recent study of 147 professional developers helps explain why some teams are accelerating rapidly while others remain hesitant.

It Starts With Repetition

Developers who use AI tools often — and across many different tasks — report the largest productivity gains. Habitual use matters more than occasional experimentation. This should not surprise anyone who has seen major technology shifts before. Productivity gains rarely arrive on day one. They emerge after workflows adapt around the tools.

That is especially true with AI.

The developers seeing the greatest value are using it to think through problems. Prompting becomes iterative. Debugging becomes conversational. Architecture exploration becomes faster and more interactive.

At that point, AI stops behaving like a feature and starts behaving like part of the development environment itself.

**Productivity and Quality**

One of the biggest fears surrounding AI-assisted development is that faster output inevitably means lower quality. The data suggests something more nuanced.

Developers are not broadly experiencing AI as a trade-off between speed and reliability. They report saving time in coding and testing. That does not mean hallucinations or bad code disappear. Experienced developers are aware of those problems. But AI changes where effort is spent.

Less time goes toward repetitive implementation work. Time shifts toward review, design decisions, verification, and system-level thinking. Developers exercise more strategic judgment.

Innovations stall when speed undermines trust. In our research, AI earns trust as it scales.

Testing Lags

The most revealing gap in the research may be testing. AI usage in coding is already widespread but testing adoption is narrower and productivity gains are lower.
Testing sits closer to reliability and production risk so teams are naturally more cautious about automating decisions that affect verification and quality assurance. But this also points to the next major opportunity in AI-assisted development: testing, validation, and software quality. The organizations that figure this out early can gain much larger advantages than those focused solely on generating code faster.

Ease Enables Intent

Developers adopt AI when the tools fit naturally into existing workflows. Ease of integration matters. Developers continue using AI when context switching is minimal, workflows remain uninterrupted, and productivity gains become tangible quickly.

On the other hand, adoption becomes difficult when premature policy and governance introduce excessive friction. Heavy approval chains, unclear rules, or disconnected tooling can inhibit the repetition needed for habits to form.

Without repeated use, adoption can’t compound. Once established, the pattern reinforces itself:
Use → Productivity → Quality → Confidence → More Use

The Early Majority Steps In

Developers who are watching—evaluating quality and waiting for organizational clarity—begin to engage more seriously. This is where adoption scales.

Overall, successful AI adoption looks like this:

Initial Use → Productivity Gains → Quality Confidence → Broader Use → Organizational Policy → Accelerated Diffusion

Once established, each step makes the next easier.

The Role of Policy

Policy does not start this process. It arrives in response to it. Organizations observe early success and then formalize expectations—defining acceptable use, processes, and responsibility boundaries. Policy legitimizes adoption.

For developers already using AI, policy reduces friction. For more cautious developers, policy opens the door. Clear AI policies reduce uncertainty about acceptable use and responsibility, making frequent use easier. Without policy, the adoption still happens, but it is weaker and more individual. With policy, the process becomes organizational.

So there is a right time and a right amount of governance for adoption to scale.

The Bottom Line

AI adoption spreads inside software organizations through visible success, repeated use, organizational legitimacy, and workflow integration. Companies that understand this dynamic are far more likely to turn AI into durable operational capability rather than isolated experimentation.

For managers, the lesson is straightforward:

Reduce friction
Expand usage beyond coding
Focus on integration
Add policy and governance at the right time

For developers, it’s even simpler:

Use AI often
Use it broadly
Stay skeptical, but engaged

The teams that do these things are quietly rewiring how software gets built.

Diffusion of Innovations in AI Adoption

In software development, adoption of AI technologies follows a familiar diffusion process. It starts with a virtuous cycle where initial use leads to productivity gains, which in turn enhance quality and build confidence, ultimately prompting further usage. At Looi Consulting, we understand that this cycle is central for teams aiming to integrate AI effectively. By leveraging this cycle, organizations can accelerate their AI adoption journey, ensuring that AI tools become an integral part of their development processes. Our study highlights that the key to successful AI adoption lies in frequent and broad usage, which not only boosts productivity but also reinforces the quality of work. As teams become more confident in the capabilities of AI, they draw in skeptics and holdouts, diffusing the innovations through the organization.