AI is changing how software gets built. Engineering teams are generating more code, development is moving faster, and organizations continue investing in AI-powered development tools. Harness found that nearly 9 in 10 engineering leaders report higher productivity after adopting AI.
Developers are spending more time reviewing AI-generated code before it reaches production. That work has become a larger part of the development process, even though many organizations still don’t measure it.
Harness surveyed 700 developers and engineering leaders for its 2026 State of Engineering Excellence report and found many organizations are still measuring engineering performance with frameworks built before AI became part of the development process.
Why It Matters: AI is producing more code, but it has also created work that many engineering metrics don’t capture. Without a clear view of where engineering time is actually being spent, organizations may have an incomplete picture of productivity and the impact of their AI investments.
- Developers Are Becoming Validators: AI is generating more of the initial code, leaving developers responsible for deciding what reaches production. Harness says reviewing AI-generated output has become a routine part of software development, yet most engineering frameworks still measure delivery more effectively than the work required to get software ready for release.
- Code Review Keeps Growing: Nearly 89% of engineering leaders said productivity has improved since deploying AI. Another 81% reported spending more time on code review. According to Harness, organizations have a much clearer view of code output than the effort required to review and validate it before deployment.
- Validation Work Is Largely Invisible: Developers estimate AI-related activities now account for 31% of the average workday. Reviewing AI-generated code ranked as the biggest source of friction, but only 38% of organizations currently measure it. That leaves a meaningful share of engineering work outside the productivity metrics many teams rely on.
- Current Metrics Leave Important Gaps: Nearly all respondents said today’s engineering metrics leave out important parts of AI-assisted development. Validation time topped the list, and only 6% believe existing frameworks are sufficient. The survey also found developers were more concerned than managers about AI productivity data being used in performance evaluations.
- Existing Frameworks Are Being Expanded: Harness isn’t recommending that organizations replace DORA metrics or other established engineering measures. The report recommends building on those frameworks by adding measures for validation work and tracking what actually reaches production. It also encourages organizations to establish clear policies for how AI productivity data will be used.
Go Deeper -> AI Productivity Is Up. The Way We Measure It Is Falling Behind – Harness
