Why 90% of Teams Use AI for Coding — and Half Still Hate the Results

In 2025, something remarkable happened in software development: the experiment finally became the industry standard. According to DORA's annual State of AI-Assisted Software Development report, ninety percent of technology professionals now use AI as part of their daily work. That is a fourteen-point percentage jump from the previous year alone — one of the fastest adoption curves seen in any engineering tool category.

But beneath that headline number lies an increasingly uncomfortable truth. Eighty-three percent of developers report that AI has increased their productivity, yet seven point two percent more production bugs roll out for every twenty-five percent increase in AI adoption. Teams are moving faster and shipping more code — but something is breaking along the way.

The question dominating technology leadership meetings this year is not whether to adopt AI in software development, but how to capture its benefits without undermining the quality foundations that separate high-performing organizations from the rest.

The Adoption Paradox: Speed Without Safety Net

DORA's survey of nearly five thousand technology professionals worldwide revealed a clear pattern. The teams that see genuine improvement in throughput and stability are adopting AI alongside three specific practices:

• Automated testing pipelines that catch regressions before they reach production — the more code generated, the more critical automated validation becomes.
• Mature version control practices including small batch sizes, meaningful commit messages, and rapid pull request feedback loops that let teams detect problems within hours rather than weeks.
• Loosely coupled system architectures where microservice boundaries prevent one poorly generated module from cascading failures across interconnected systems.

Meanwhile, teams that simply turned on AI assistants without strengthening these underlying controls saw exactly what the data predicted: acceleration amplified their existing chaos.

The Trust Gap No Tool Can Fix

Perhaps the most revealing finding from DORA's research concerns developer confidence. While seventy percent express some degree of trust in AI-generated code quality, thirty percent report little to no trust whatsoever. That number has barely moved from last year despite widespread tooling improvements.

The skepticism is not baseless. Large language models generate plausible-looking code that may contain subtle logic errors, security vulnerabilities, or architectural mismatches with existing systems. The output looks correct at a glance — which makes it exactly the kind of bug that slips past peer review before reaching testing.

This trust deficit points to an important reality about AI-assisted development: it is not a replacement for engineering discipline; it is a multiplier of whatever discipline already exists in a team. Organizations with strong code review practices, comprehensive test coverage, and clear architectural standards see AI amplify those strengths. Organizations without those foundations watch AI accelerate their existing problems.

Platform Engineering: The Missing Foundation

If there is one practice that consistently separates teams benefiting from AI from those struggling with it, platform engineering deserves the spotlight. Ninety percent of organizations now have at least one internal platform, and the data shows a direct correlation between platform quality and AI success rates.

Internal developer platforms solve several problems simultaneously for AI-assisted teams:

• Standardized toolchains. When every team member works with identical development environments, build configurations, and testing frameworks, AI-generated code follows consistent patterns that are easier to review and validate.
• Discovery and reuse. Well-maintained internal platforms make is obvious when existing libraries or APIs already solve a problem, preventing developers from accidentally generating duplicate functionality.
• Governance guardrails. Platform teams can embed coding standards, security scanning, and architecture review directly into the development workflow — catching issues before code leaves a developer's machine rather than discovering them in staging environments.

For organizations evaluating whether platform engineering is worth the investment, DORA's research suggests an unequivocal answer. The teams that treat their internal platform as a product designed for developer experience consistently outperform those that manage tooling reactively or leave it entirely to individual engineering discretion.

Gartner's Roadmap: Where AI Development Fits in 2026

Parallel to DORA findings, Gartner's 2025 Technology Adoption Roadmap for Software Engineering identified nine strategic trends shaping how organizations approach AI-enabled development. Three of these deserve particular attention from technology leaders planning their engineering strategy this year:

1. AI code assistants in deployment. After years of pilot programs, AI-assisted coding tools have moved beyond the testing phase into mainstream production use. The question has shifted from "should we use them?" to "how do we govern their use responsibly?"
2. Software supply chain security. With more code flowing through faster pipelines and coming partly from non-human sources, visibility into dependencies, license compliance, and security posture becomes critical. Software Bill of Materials adoption is accelerating exactly for this reason.
3. Cloud development environments. Remote and ephemeral development workspaces allow teams to standardize every developer's environment automatically, removing a major source of inconsistency from AI-assisted workflows before code ever enters version control.

Gartner predicts that by 2028, ninety percent of enterprise software engineers will use AI code assistants — up from less than fourteen percent in early twenty twenty-four. Organizations that invest in their foundational engineering practices now will be far better positioned to capture value when AI becomes the default tool rather than a experimental advantage.

What This Means for Businesses Evaluating Technology Partners

For organizations considering IT consulting engagements, custom software development partnerships, or ERP system modernization, these trends carry practical implications:

• Ask about AI governance strategies. Any technology partner should be able to articulate not just tools they use for code generation, but how they validate, review, and test AI-assisted output before it ships. If the answer only discusses productivity metrics without mentioning quality controls, proceed with caution.
• Prioritize platform-aware partners. Vendors who understand internal platform architecture can help you build development environments that work well alongside AI tools — rather than fighting against them or creating shadow infrastructure outside your governance model.
• Measure outcomes holistically. Productivity gains from AI are real, but they mean little if delivery stability degrades in parallel. Request metrics that track both throughput and quality indicators such as defect escape rates, deployment failure frequency, and mean time to recovery.

The Path Forward: Treat AI Adoption as Organizational Transformation

The consistent message across all major research this year is that AI in software development is not a tooling problem — it is an organizational challenge. The technology works. The question is whether your processes, architecture, and culture can handle the acceleration AI introduces.

Teams that succeed share several characteristics:

• Clear AI policies that specify what tasks are appropriate for AI assistance, what requires human-only review, and how generated code integrates into existing development standards.
• Healthy data ecosystems where models have access to enough organizational context to generate relevant suggestions without compromising sensitive information or architectural decisions.
• User-centric focus — the strongest finding from DORA's research is that AI becomes most useful when directed at clearly defined problems. Teams that identify specific pain points before deploying AI tools consistently outperform those that enable broad access and hope for the best.

Actionable Takeaways

If your organization is evaluating or expanding its use of AI in software development, here are the concrete steps that deliver measurable value without introducing instability:

1. Strengthen automated testing before expanding AI adoption. If test coverage is below seventy percent for core functionality areas, prioritize filling those gaps. AI-generated code will multiply your existing quality problems faster than it multiplies your productivity.
2. Define an AI policy document even if it starts simple. Specify when AI tools are appropriate, what requires human review, and how generated code enters your version control pipeline. A written policy is measurably more effective than informal guidelines.
3. Invest in platform engineering as infrastructure, not overhead. Consider your internal developer platform the foundation that enables all other engineering investments to deliver value. Teams with mature platforms see three times greater AI adoption rates and faster time-to-productivity gains.
4. Measure what matters. Track both delivery speed and stability metrics simultaneously. If only one goes up while the other declines, your investment in tools alone is insufficient — you need process improvements alongside technology.

Conclusion

The narrative around AI in software development has moved past the hype cycle into a more mature phase of organizational adaptation. Ninety percent adoption tells us that AI-assisted coding is not going away — but nearly half that population still has reservations, and those concerns are rooted in real operational challenges that no single tool can solve.

The organizations that will benefit most from AI this year are not necessarily the ones buying the best tools or hiring the most experienced practitioners. They are the ones building stronger foundations: better testing, clearer governance, more modular architectures, and internal platforms that make every developer's job easier regardless of which tools they use.

If you are a technology leader evaluating how to approach AI-assisted development responsibly within your organization, or if you are a business considering whether an IT consulting partner can help modernize your software delivery practices, the starting point is not choosing a tool. It is understanding where your organization sits on the foundation-building spectrum and investing accordingly before scaling up.