Why an internal platform affects efficiency, delivery stability, and scaling AI in software development.
5.12.2025 This article is an adapted translation of the report "2025 State of AI-Assisted Software Development" by Google Cloud and DORA. The original is available at the link. Reading time: 9 min. This chapter was prepared with contributions from: Eric Maxwell - Head of 10x Technology, Google Cloud Benjamin Good - Head of Platform Engineering, Google Cloud
In the 2024 report, we began examining how internal platforms affect software delivery performance.
Back then, the data showed that platforms improve organizational performance and team productivity, but at the same time increase delivery instability and reduce throughput. In this year's study, we went beyond stating the value of platform engineering and focused on how successful platforms actually work.
The results show that a platform is not a set of tools, but a coherent user experience that directly affects efficiency, resilience, and the organization's ability to use new technologies.
Users perceive the platform as a single whole, and the overall coherence of that experience matters more than the quality of individual features. This chapter examines the key patterns in the evolution of platform engineering: widespread adoption, changes in team structures, and the role of platforms as a strategic foundation for innovation and risk management.
Platforms have become standard: adoption is around 90%. 76% of companies use dedicated platform teams.
The task changes: what matters is not adaptation itself, but managing a multi-team platform environment.
Platform quality amplifies AI's impact on organizational outcomes.
The higher the quality, the more visible the effect.
Strong platforms improve team efficiency, productivity, and well-being.
A platform is a coherent product that should deliver a high-quality developer experience.
The platform helps manage risk and speeds up experimentation.
This is accompanied by a small but justified increase in delivery instability - an acceptable tradeoff for higher efficiency.
Most companies already use internal platforms.
90% of organizations have adopted at least one platform, and nearly 30% operate in a multi-platform environment. 76% of companies have a dedicated platform team.
More than a quarter of respondents work in organizations with multiple such teams.
This points not to duplication, but to a shift from a universal platform to a set of specialized and federated platforms for different domains and stacks.
The challenge for leaders has changed too.
Now it is important not just to "have a platform," but to manage an ecosystem of multiple platforms.
This requires principles of loose coupling between teams, clear ownership boundaries, and aligned interfaces.
This structure helps platforms work as a unified system, improving the developer experience instead of creating new organizational barriers.
Respondents were asked to rate how well their platform performs different tasks. For example: "The platform helps me build and run secure services."
Core technical capabilities - reliability and security - are seen as strengths. But user experience elements - feedback handling, usability, and the level of automation - are notably weaker.
A developer's overall impression of the platform - as a helpful partner or as a source of friction - strongly shapes how each individual feature is rated. The fact that platform capabilities cluster fairly tightly shows that respondents do not see the platform as a set of separate components; they experience it as a single whole.
The experience gap likely reflects a situation where the platform's technical foundations are built first.
These data show that until the user experience improves, the platform does not realize its full value.
Adopting a product mindset for the platform - treating internal development tools as products and developers as users - helps keep the focus on the user.
This is a key takeaway from the 2024 report.
User focus helps avoid common mistakes when building internal developer platforms. "Build it, and they will come"
The team builds the platform based on its own assumptions about what developers need, without doing research, interviews, or hypothesis testing.
They focus only on technology, assuming the value will be obvious on its own. Why this does not work:
The platform turns into a ghost town: it does not solve real developer problems or fit into their workflows. What it should look like:
The product approach starts with empathy and an understanding of developer needs.
The platform team stays in constant contact with users to understand their main pain points and build what is actually needed. The "ticket-ops" trap
They have no vision or roadmap; the work is fully reactive and consists of an endless stream of developer requests ("Create a database", "Set up CI/CD"). Why this does not work:
A bottleneck emerges, increasing the load on both the platform team and developers.
The team spends all its time on one-off tasks and never gets around to building coherent, self-service capabilities. What it should look like:
The product approach means building a self-service platform with a clear strategy.
The goal is to eliminate ticket queues by giving developers tools and automation that let them get the resources they need on their own. The platform as an ivory-tower elephant
A central team dictates the platform architecture and toolset from the top down, setting rigid standards without feedback or collaboration.
They act as technology gatekeepers rather than developer helpers. Why this does not work:
As a result, shadow IT or unofficial workarounds emerge, undermining the platform's purpose. What it should look like:
The platform product manager actively gathers feedback, and developers are treated as customers.
The platform should help, not constrain: it should offer convenient golden paths that can be used, but do not have to be.
When we look at platform capabilities and how they relate to one another, most of them correlate with each other about equally. But two capabilities stand out more than the rest. First, the one most strongly linked to a positive user experience is clear feedback on tasks in progress.
When the platform clearly shows that a task completed successfully or failed, users feel they can move on:
Second, a "simple and clean interface"
shows a weaker correlation.
Polished UI can improve perception, but it does not make a platform effective on its own.
Although clear feedback and UI are closely tied to other capabilities, they remain among the weakest in perception.
This means that improving one capability in isolation is not a viable strategy.
To improve the perceived quality of the platform, teams must treat it as a cohesive internal product and improve the entire developer journey.
If a platform is technically strong but not user-oriented (see the DORA 2024 report), it cannot be considered successful.
A high-quality platform - by our metrics - has a broad and statistically significant positive impact. It is associated with higher organizational performance, better product outcomes, and increased productivity. Consistent with previous research, we also found that the better the platform, the greater the small but reliable increase in delivery instability
That means software - in other words, a higher rate of failed changes and rework. "The general idea was to do continuous delivery as often as possible: if a feature made it into main, ship it, launch it, run the tests - as much as possible - to catch issues. But the main thing was just to ship. And in most cases that was fine, because even if something broke during deployment, we had a lot of mechanisms that kept the site from going down. So the deployment could fail, and we had to fix it quickly."
But the site itself kept working"
A good platform acts as an amplifier: it directly increases productivity and efficiency.
A small increase in instability can be a sign of a healthy high-speed system, where additional failures are acceptable as long as they do not affect the product.
The increase in efficiency despite greater instability points to a kind of risk compensation: the platform makes rollback fast and inexpensive, teams recover from errors more easily, and they can experiment more often by accepting a slightly higher level of minor failures in exchange for speed.
A small amount of instability should be viewed as a controlled tradeoff for a substantial gain in performance.
The product quality improvements a platform enables are likely more important than a moderate increase in delivery speed and a slight drop in stability.
Investing in a high-quality platform is a powerful strategic lever that delivers broad and durable impact.
The positive impact of adopting AI on organizational effectiveness directly depends on the quality of the internal platform. If platform quality is poor, AI's impact is barely noticeable. When the platform is strong, the effect becomes clear and positive.
This is a critically important takeaway for any leaders investing in AI. "At Wayfair, they realized that the biggest impact came not only from detecting errors faster, but also from reducing the effort required to fix them. Embedding AI into the CI/CD loop showed that developers are most likely to use precise, explainable recommendations and automatically generated fixes when they appear directly in the tools they already use."
- Wayfair A high-quality internal platform serves two key functions, amplifying AI's impact on organizational outcomes. First, it acts as a distribution and governance layer that makes it possible to scale AI benefits from individual productivity gains to system-wide improvements across the organization.
Without that foundation, AI adoption remains a set of disconnected local optimizations.
The platform provides a shared context and takes on the complex or routine aspects needed for AI to be useful and operate at scale.
At the same time, it is important to remember that AI is evolving very quickly.
Do not over-standardize AI practices, tools, and ways of working, because that can limit its benefits and the organization's ability to adapt to change. Second, the platform acts as a risk-management mechanism.
Since the platform reduces risk in other areas of engineering, the same effects are also useful for AI adoption.
The platform should create a safe space where employees can learn and experiment.
This space allows the platform and its teams to adapt faster and support new models, interaction methods, and approaches to application development.
In addition, whether code is written manually or with AI, it goes through the same automated testing and deployment processes, which helps ensure changes are safe.
Investments in AI without parallel investments in high-quality internal platforms are unlikely to produce meaningful organization-wide impact.
To gain real competitive advantage from AI, leaders need to treat platform engineering as a fundamental strategic tool.
Look at the platform holistically. You cannot fix a bad platform by improving one or two features. Treat the platform as a full product and work on the entire developer experience loop - from feedback to automation.
Make the platform the foundation for AI. The platform is a strategic prerequisite for unlocking AI value across the organization. It is the engine that turns AI investment into competitive advantage.
Use the platform to manage risk. A good platform changes how the company relates to risk: mistakes become cheap and easy to fix. It is important to understand and manage the balance between the speed the platform enables and the instability it creates, while remembering that the platform does not eliminate the local consequences of risk.
To understand how organizations turn AI from isolated improvements into scalable outcomes, we turn to value stream management.
This is where it becomes clear why some teams get real value from AI while others get lost in chaotic activity.
The seventh part of the study shows that the key factor is not delivery speed or a set of practices, but the ability to see the full value-creation journey and manage it as a coherent system.
This is the foundation that turns AI into a competitive advantage rather than a source of added complexity.
