The State of DevOps Modernization 2026 (Devops column)

Checked out the recently released report "The State of DevOps Modernization 2026" from Harness. The report is based on a survey ~700 Director and Chief Executive Officer, which was held in February 2026 of the year by Coleman Parkes. The main conclusion of the report is that AI has dramatically accelerated the local coding cycle, and the rest of the delivery loop in many teams within corporations does not have time to digest it.

The basic analytical model of the report is very simple: almost all conclusions are built through a cross-tab on the question of "how often do you use AI coding tools" - from a few times a day to more infrequent use. That is, this is not an analysis of telemetry and not a causal inference as in DORA, but a slice of perception and self-esteem of respondents.

The authors highlighted the following key results

1The speed has really increased. 45% of very frequent AI users report daily or more frequent deployments 32% frequent and 15% for occasional users. But next to that comes the Stability Tax: 69Very frequent users say that AI-generated code causes deployment problems in at least half of the cases. 22The percentage of deployments in this cohort end in a rollback, hotfix or customer-impacting incident; the MTTR for such incidents is higher. 7.6 hourly 6.0 and 6.3 hours in neighboring cohorts.

2️ But that tension extends far beyond incidents. Very frequent AI users 50Percentages of vulnerabilities/security incidents 50Percentage of non-compliance issues; 49Percentages see more productivity problems 51% - more code quality/efficiency issues. At the same time, the authors themselves specifically stipulate that there is a clear causal relationship between AI coding and these problems, the report does not prove. (Such an experiment design can only show correlations.)

3Coding automates faster than the rest of the SDLC. 84Percentage of respondents use AI daily for coding, but for QA testing this figure 68% for performance/cost optimization 63%, for refactoring 62%. 47% of very frequent users say that manual work further down the pipeline after the code has become more problematic (QA, code review, remediation); 69Losing time due to slow and unreliable CI/CD 70Percent believe their pipelines suffer from flaky tests and failed deployments 75% associate pressure on the shipping with burnout; 73Percent say they have almost no standard templates or golden paths

As a result, we can see that the problem is that AI amplifies throughput where the platform, CI/CD, gates of control and coordination have remained at the old speed. They hypothesize that heavy AI users are simply trying to deliver more code and therefore feel flaky tests and deployment failures more acutely; their pipelines are not necessarily worse, just the needs are higher. This is a strong insight, but as evidence of causality, the report does not provide:)

I wrote about the same thoughts in the article.From PDLC to AI-native development", but mainly based on other studies of the type DORA or AI4SDLCWe did the T-Tech at the end of last year and will do it again this year.

#Engineering #AI #Metrics #Software #DevEx #Productivity #DevOps #Architecture #Culture #Engineering #ML #SystemDesign