The share of developers reporting ten or more hours saved each week through AI fell from 68% to 26% in a year, according to Atlassian's State of Developer Experience 2026. Adoption over the same period became close to universal, with 95% of developers and 96% of engineering leaders saying their organization actively uses AI in engineering.
The report draws on 1,155 individual contributors and 405 engineering leaders surveyed during 2026. Atlassian is unusually direct about what that means, stating that its findings "reflect self-reported experience and perception, not system-level telemetry such as deployment frequency, cycle time, or revenue impact," and that associations in the data "should not be interpreted as causal proof." Taroon Mandhana, Atlassian's CTO for AI and Teamwork, summarised the shift as one where "the bottleneck has moved to managing AI's work."
Eighty-seven percent of developers say AI helps them ship faster and 80% say it helps them deliver more without longer hours, but only 28% report noticeably less friction than before AI. Half report about the same amount, which the report describes as friction "of a different type than before AI."
Part of that difference is the AI work itself: 49% of developers now spend four or more hours a week managing AI tools, 65% say they spend time correcting AI output that missed the original intent, and both figures run higher among engineering leaders, at 57% and 72%. Twenty-nine percent name reviewing or managing AI-generated code as a source of friction outright.
The friction list below AI has barely moved in three years. Finding information leads at 51%, followed by meetings at 50%, context switching at 45% and technical debt at 44%; finding information also topped the 2025 list and placed second in 2024. Coding has never appeared on it.
Sixty percent of developers say AI use is treated as a productivity signal in their organization, 41% say they sometimes use AI more than necessary because they feel expected to, and only 42% of leaders say they have the right metrics to separate real gains from activity. Those three figures are where the report is most pointed, and where the discussion under Mandhana's post concentrated. Shan Huang, who builds production AI agents for enterprise workflows, connected two of them:
If people are using AI more than necessary because they feel expected to, part of that 49% spending 4+ hours a week managing AI output is not a tooling problem, it is a metric problem. Call it performative throughput: the usage gets counted as productivity, and the cleanup lands on somebody else’s calendar.
Taylor Johnson of Athena Intelligence proposed a replacement in one line:
AI activity is not impact; measure accepted work, not usage.
Organizations with a stated position are rare and appear to fare better. Seventeen percent of developers and 23% of leaders report a clear, actively implemented AI strategy across the software delivery lifecycle, and among developers who do, 41% report noticeably less friction than before AI, against 22% of those describing an inconsistent or poorly communicated strategy. Andrew Boyagi, Atlassian's head of DevOps evangelism, calls that gap the differentiator, though the report itself cautions that the association is not causal. One trend, meanwhile, has run the same way for three years: the share of developers saying leadership does not fully understand the challenges they face was 44% in 2024, 63% in 2025, and 66% in 2026.
The headline figure carries caveats of its own. This is vendor research from a company selling developer tooling, arriving at a conclusion that favors formalizing AI delivery; the comparison base shrank from 3,500 developers and managers across six countries in 2025 to 1,560 respondents this year; and neither edition publishes fieldwork dates. Self-reported hours are also soft in both directions, which commenter Harshit Jain raised directly: is the drop about review load, or about higher expectations? The report does not separate them.
Cam Turner, a designer and former AWS staffer, put the practical question that neither the report nor the thread answered: "68 to 26 in a year says the work moved from making to reviewing. Is anyone designing the review step yet, or are teams just absorbing it?"