AI pays off only on a good platform
In 2026, building a software platform means building the product and the system that ships it, and the second part now decides how much value teams get from AI. Internal platforms are close to universal: 90% of organizations have adopted at least one, according to DORA 2025, and the same research finds that AI improves organizational performance only when platform quality is high. This article covers where platform development stands, the challenges teams report, and what good practice looks like.
What does platform development mean in 2026?
The word covers two things. A product platform is everything your product runs on: the back end and APIs, the web and mobile apps, perhaps AI agents, and the pipelines that release them. An internal developer platform is the self-service layer engineers use to build and run those parts: environments, golden paths, CI/CD, observability. Most products need both, and the second decides how fast the first can change.
Internal platforms are now standard. In DORA 2025, a survey of nearly 5,000 technology professionals, 76% of organizations have at least one dedicated platform team and 29% already run more than one platform. Analysts saw this coming: Gartner forecast that 80% of large software engineering organizations would have platform engineering teams by 2026, up from 45% in 2022.
Platform work is not limited to platform teams. In the Stack Overflow Developer Survey 2026, 46.0% of 13,338 respondents built or planned cross-team or platform tools in the past or coming year, and 63.0% built team-level internal tools.
What developers built or plan to build in-house
- Small scripts and automations
- 77.3%
- Team-level internal tools
- 63%
- Customer-facing products
- 54.4%
- Cross-team or platform tools
- 46%
- Tools replacing paid vendors
- 18.6%
Source: Stack Overflow Developer Survey 2026, BuildTool (n=13,338, 2026)
Why does the platform decide whether AI pays off?
This is the most useful finding for anyone investing in AI. DORA 2025 reports that AI adoption has a negligible effect on organizational performance when platform quality is low, and a strong, positive effect when it is high. Without a shared platform, AI gains stay with individual developers. With one, they reach the whole delivery system.
The reason is practical. The platform is where AI-written code meets the same tests, security controls and deployment pipeline as code written by hand. DORA puts it directly: the same automated testing and deployment processes are applied whether a change came from a person or a model.
There is a trade-off. DORA links a better platform to a small but credible increase in delivery instability, meaning a higher change failure rate and more rework, which DORA reads as teams shipping and experimenting more. DORA calls this manageable. We read it as a case for strong automated tests: they make failures cheap to catch and cheap to reverse.
What are the main challenges in platform development?
1. People and skills, more than tools. In the CNCF Annual Cloud Native Survey (628 respondents, September 2025), the top challenge in using and deploying containers was cultural change with the development team, named by 47% of end-user organizations, ahead of lack of training and security at 36% each. In 2023 the top three were technical: security, complexity and monitoring, according to the same report.
Top challenges in using and deploying containers, 2025
- Cultural changes with the development team
- 47%
- Lack of training
- 36%
- Security
- 36%
- CI/CD
- 35%
- Monitoring
- 35%
- Complexity
- 34%
Source: CNCF Annual Cloud Native Survey (n=351 end-user organizations, September 2025)
2. Platforms that developers avoid. DORA names three failure patterns. "Build it and they will come" produces a platform nobody asked for. The "ticket-ops trap" turns the platform team into a queue of one-off requests. The "ivory tower platform" dictates tools without feedback and pushes teams to workarounds. DORA also found an experience gap: reliability and security are perceived as well provided, while acting on feedback and task automation lag behind.
3. Delivery plumbing that is still hard. CI/CD and monitoring were each named by 35% of organizations, and complexity by 34%, in the CNCF survey. These are the parts a product platform shares across web, mobile and back end, so a weak pipeline slows every client at once.
4. AI adds a new security surface. In the Stack Overflow Developer Survey 2026, 23.5% of respondents said security or privacy issues caused by AI mistakes cost their team the most time in the past year. A platform that governs model access, secrets and data flows is where that gets fixed once instead of in every app.
Which architecture choices matter most?
The runtime question is largely settled for teams that use containers: 82% of container users run Kubernetes in production, up from 66% in 2023. In the same CNCF survey, 98% of organizations use cloud native techniques in at least some capacity. The decisions that separate good platforms happen higher up:
- One data model and one set of APIs for web, mobile, agents and admin tools, so every client sees the same truth.
- Service boundaries that follow teams and load. A practical default is a modular back end that splits into services when scaling or ownership demands it, rather than many small services on day one.
- Loosely coupled teams with clear interfaces. DORA recommends this for platforms as well as applications, to avoid a platform of platforms that creates new silos.
- Paved roads, not mandates. DORA describes good platforms as easy and desirable to use, but not necessarily mandatory.
Where does AI fit inside the platform?
In two places: as a workload the platform runs, and as a tool the platform governs. Among organizations that host generative AI workloads, 23% run them fully on Kubernetes and 43% partly, according to the CNCF survey. Most teams consume models rather than build them: 52% of surveyed organizations do not build or train AI models.
For a product team, that makes inference the platform work: routing calls to model APIs, fallbacks across providers, caching, cost and latency monitoring, access control, and evals that run on every change. The same platform governs AI as a tool, deciding which agents may touch which repositories, environments and data.
What does good platform development look like, and how do you start?
- Treat the platform as a product. Its users are your engineers. Interview them, publish a roadmap and measure adoption.
- Give clear feedback on every task. DORA found that clear feedback when a task succeeds or fails is the capability most correlated with a positive platform experience.
- Use one release pipeline and one test strategy across back end, web, mobile and agents.
- Hold AI-written code to the same gates as human-written code: review, automated tests, security checks.
- Start with the slowest step. Map the path from commit to production, fix the step that costs the most waiting, and measure before and after.
Want one team to build and run your platform?
SDCLabs builds product platforms end to end: back end and APIs, web and mobile apps, AI agents, and the pipelines that release them. Senior engineers own the architecture, AI agents draft code, tests and reviews inside it, and automated tests from the founders of Codeception and CodeceptJS verify every change. In our published case study, a five-person team built the web, desktop and mobile apps of a productivity management platform.
Engagements start in 1 to 2 weeks. See platform development or tell us what you are building. Related: DevOps and TestOps and our report on the state of AI engineering in 2026.
Sources checked on 2026-11-17.
Frequently asked questions
How common are internal developer platforms in 2026?
In DORA 2025, 90% of organizations had adopted at least one internal platform and 76% had at least one dedicated platform team.
Does platform quality change the value of AI?
Yes. DORA 2025 found that AI adoption has a negligible effect on organizational performance when platform quality is low, and a strong, positive effect when it is high.
What is the biggest challenge in running a cloud native platform?
People more than tools: in the CNCF Annual Cloud Native Survey, cultural change with the development team was the top challenge at 47%, ahead of lack of training and security at 36% each.
How many teams run Kubernetes in production?
Among container users, 82% ran Kubernetes in production in 2025, up from 66% in 2023, according to the CNCF Annual Cloud Native Survey.
