Two numbers from the same survey that should not coexist
Writer's 2026 survey of 2,400 executives and employees produced a statistic every vendor keynote will quote: 97 percent of companies deployed AI agents in the past year. It also produced a statistic almost no keynote will quote: only 23 percent of those same companies see significant ROI from the agents they deployed. Put those two numbers side by side and the conclusion is not that agentic AI failed to catch on. It caught on everywhere. The conclusion is that deployment and value are now two almost entirely separate questions, and most organizations have only answered the first one.
Writer CEO and co-founder May Habib frames the gap as a people problem rather than a model problem, arguing that AI transformation is ultimately about people and that the organizations that win will be the ones that give agent-building power to the people closest to the work rather than centralizing it in a single platform team. That framing is self-serving for a company selling an agent-building platform, and it is also consistent with what the rest of this year's enterprise surveys keep finding independently.
The struggle number is getting worse, not better
Seventy-nine percent of organizations in Writer's survey report significant adoption struggles, a double-digit increase from the prior year. That is the opposite of the trajectory a maturing technology is supposed to follow. Tools that are working through normal adoption friction usually show the struggle rate falling as best practices spread and tooling improves. A rising struggle rate alongside near-universal deployment suggests organizations are moving agents into more consequential, harder workflows faster than their governance and integration capacity can absorb, not that the easy use cases are simply taking longer than expected to settle.
A separate estimate from Digital Applied puts real numbers behind that struggle: 88 percent of enterprise agent pilots never reach production, with infrastructure gaps named by 41 percent of respondents as the primary blocker, governance and security by 38 percent, and ROI measurement difficulty by 33 percent. Only 11 percent of enterprises in that estimate run agents in production at all, a smaller figure than the headline adoption numbers in most 2026 surveys, because piloting and deploying are being counted as the same thing when they are not.
The budget keeps growing regardless
Fifty-nine percent of companies in Writer's survey already spend more than a million dollars annually on AI, and 88 percent of executives plan to increase their agentic AI budgets further, despite the ROI figures sitting at 23 percent. That is not necessarily irrational. A board that believes agentic AI is a multi-year structural shift can reasonably keep funding it through a rough measurement period the same way it would fund any other long-payback infrastructure investment. But it does mean most of that spending increase is currently happening without a reliable feedback loop telling finance whether the next dollar is better spent than the last one.
The sharpest number in the survey is internal rather than financial: 54 percent of C-suite leaders say AI adoption is tearing their company apart. That is a description of organizational strain, not technology performance, and it is consistent with a rollout that moved faster than the operating model around it. A company can afford a model that underperforms for a quarter. It has a much harder time affording a C-suite that is actively divided over how the rollout should work, because that division tends to slow every subsequent decision, not just the AI ones.
Governance maturity, not model choice, separates the outcomes
Deloitte's figures, cited alongside Writer's survey, put only one in five companies at a mature governance model for autonomous agents, and a finance-leader survey of 500 respondents found 79 percent had cost overruns in the past year with only 15 percent able to calculate AI ROI without significant bottlenecks. None of these figures point to a model-quality problem. They point to an operating-model problem: organizations bought capable agents and then ran them inside processes, budgets, and reporting structures that were never built to track autonomous decision-making. The failure mode in nearly every cited dataset is organizational, not technical.
The 56 percent figure worth remembering from this cluster of surveys is simpler than any of the others: enterprises that scale an agent successfully are far more likely to have named a single, accountable owner for it. Thirty-five percent of executives in the finance-leader survey said they could not immediately stop a rogue agent if one misbehaved, which is a direct consequence of nobody being clearly responsible for the kill switch. Ownership is a cheaper fix than almost anything else in this data, and it is the one most organizations have skipped.
What to take into the next budget cycle
The honest read of this survey cluster is that agentic AI adoption and agentic AI value are no longer the same conversation, and treating them as one is how a CIO ends up explaining a 97 percent deployment rate to a board asking why ROI sits at 23 percent. The fix is not slower adoption. Enterprises that scale successfully are not the ones that waited. They are the ones that named an owner, built a measurement framework before the rollout rather than after, and treated governance and security as the primary blocker to solve first, matching exactly the order Digital Applied's respondents ranked their own obstacles.
Before the next budget cycle locks in another increase to agentic AI spending, it is worth forcing three answers onto paper: who owns this agent if it fails in production, what KPI proves it is worth the spend, and which pilots are being allowed to quietly become production systems without the governance review that should gate that transition. Writer's survey and the datasets around it agree on little except this: the companies getting real ROI answered those three questions before they scaled, not after.
