Presented by Caja
Content access, governance and platform flexibility are emerging as the dividing lines between leaders and laggards in AI, according to new Report on the state of AI in the enterprise from Box, which surveyed 1,640 IT decision makers in the US, UK, France and Japan. One of the report’s key findings is the speed of change: The combined proportion of organizations that describe themselves as advanced or cutting-edge soared from 8% to 64% in just the past year, while the proportion that call themselves early stage or not yet started plummeted from 53% to just 9%. Eighty percent of organizations reported a notable return on their AI investment, defined in the survey as an improvement of at least 10%, and more than half saw measurable business impact within six months of approving a project.
The shift is largely due to how companies are now organizing their use of AI rather than a single technical advance, says Olivia Nottebohm, chief operating officer at Box.
"We have moved from independent experimentation that lived at the individual level to integrated and systematized agent operations, agents that are in production and can be used in a repeatable way." says Nottebohm. "That’s where the impact comes from."
Why AI leaders see higher ROI than early-stage companies
The division between levels is a matter of execution. Significantly, half of cutting-edge companies reported an AI-driven ROI of over 25%, compared to just 11% of early-stage companies, with advanced (33%) and development (16%) tiers consistently falling in between. But Nottebohm says the real differentiator was not whether companies adopted AI, but the rigor with which they integrated and managed it.
"What separates the leading edge is the operational muscle they’ve built: the right teams to deploy agents, formal governance to control them, and consistency in the content layer from which those agents work." she explains. "Early-stage companies are approaching it in a much more ad hoc and experimental way, letting people play with it without the same intention or structured design."
Access to content is the biggest barrier to enterprise AI ROI
Content, rather than model quality, will be the defining bottleneck in 2026. Ninety-six percent of organizations say agents need access to enterprise-specific content, yet only 36% have connected agents to trusted content across many use cases. It’s a question of confidence rather than raw ability.
"We started this journey assuming that enterprise AI was about access to the latest model," says Nottebohm. "But the question now is whether agents have access to the right content and whether that content is protected, because those agents are only as good as the content they can reference, and as secure as the security around them."
Getting that content layer right has a second benefit beyond security, as it’s also what finally allows agents to work in departments that previously operated in isolation from each other. And while about a quarter of organizations point to fragmented data across systems, 24% cite difficulties integrating AI into existing systems, 21% say they lack adequate permissions and access controls, and 18% describe their content as too disorganized to make accessible at all. Among more mature organizations, 63% now treat unstructured documents, contracts and reports as a competitive advantage rather than dead weight in a digital filing cabinet.
Reduce common AI data exposure incidents
Nearly half of all organizations say they have already experienced an AI-related data exposure incident. That figure rises to 60% among cutting-edge companies, which may face greater exposure from more agents and connected systems, but may also be better equipped to detect it.
The proportion of organizations reporting established or advanced governance frameworks increased from 24% in 2025 to 73% this year, but real gaps in instrumentation remain: only 39% have comprehensive visibility into authorized and unauthorized AI use, 34% have formal standards for how agents access company data, and 27% still describe their governance as ad hoc. But such incidents function as a force mechanism rather than a setback, Nottebohm says.
"Governance used to be seen as something that slows people down, but 93% of respondents told us that better governance is actually what allows them to move faster." she explains. "It makes scalable AI survivable. Once the content is protected and has many permissions, you can run multiple agents in multiple processes and get a real multiplier effect."
A practical consequence of that shift is that leave structures created for human employees are now being reviewed with agents in mind, a process most companies are only halfway through.
"It is necessary to review the permits that companies created two years ago," she explains. "Until recently, people didn’t set permissions on a document considering how an agent might use it, but now they’re much more deliberate about it. It leaves them with a whole corpus of unstructured data to review and clean or reauthorize."
This is part of a broader move away from governance designed for people and towards governance designed for agents from the start.
"Enterprises must transition from governance tailored for human workflows to governance designed specifically for agents," says Nottebohm. "That means tracking what an agent has touched, what permissions were applied, and what sources were used, and all of that is now shaping how governance is applied."
Companies should avoid relying on a single AI provider
"The days of token maximization are long gone," says Nottebohm. "Now it is the responsibility of delivering efficient AI. Organizations want to use the cheapest model that meets the quality standard they need, not necessarily the most expensive, because different model families continue to outperform each other and companies want to preserve that choice."
That means businesses are avoiding lockdown more than ever. Sixty-eight percent say they are concerned about being dependent on a single AI vendor, the average number of officially adopted AI tools has increased to 3.3, and 79 percent now consider it important or critical for agents to operate headless, connecting directly to systems and APIs without a human interface.
It’s a trend similar to the shift toward multi-cloud infrastructure, and driven by a similar reluctance to hand a single vendor outsized bargaining power.
"A flexible architecture is based on the interoperability of platforms," says Nottebohm. "It runs on multiple models, operates headlessly, and keeps each part of the AI stack interchangeable, so organizations don’t have to bet on which individual tool wins, and that’s part of a broader shift from the default model to the largest and most expensive model available."
The next steps towards AI success
Over the next three years, enterprises should prioritize organizing, sorting, and cleaning up unstructured content, actively hiring and building teams around emerging roles, and adopting a hybrid token compute budget model, where IT owns the core infrastructure and token budget, while business units own application-level spend. And right now, it’s easy to catch up quickly.
"It is not necessary to start early and move slowly," says Nottebohm. "By incorporating governance, the content layer, and the multi-model system from the beginning, you can enter as a leading company and capture that same enormous impact."
Sponsored articles are content produced by a company that pays to publish or has a business relationship with VentureBeat, and are always clearly marked. For more information, contact sales@venturebeat.com.
