Enterprise artificial intelligence has made an unprecedented leap. According to the new State of AI in the enterprise report by Box, which surveyed 1,640 IT decision makers across the US, UK, France, and Japan, the share of organizations describing themselves as advanced or leading edge soared from 8% to 64% in just twelve months. Simultaneously, the share of companies in early stages or not yet started collapsed from 53% to 9%. Eighty percent of organizations reported a notable return on AI investment (at least 10%) and more than half saw measurable business impact within six months of project approval.
Internal organization beats technology as success factor
The shift is not due to a single technical breakthrough but to how enterprises organize their AI use. Olivia Nottebohm, COO of Box, explains that companies have moved from isolated individual experiments to systematized, integrated agentic operations. Agents are now in production and can be used in a repeatable manner. This structural change drives the positive impact. The gap between leaders and beginners is stark: half of leading-edge companies reported AI-driven ROI above 25%, compared with only 11% of early-stage firms. The real differentiator is not adoption itself but the rigor of integration and management. Leaders have built operating muscle: right teams, formal governance, and consistency in the content layer agents work from.
Sponsored Protocol
Content access is the main barrier to enterprise AI ROI
Content, not model quality, is the defining bottleneck of 2026. Ninety-six percent of organizations say agents need access to company-specific content, but only 36% have connected agents to trusted content across many use cases. It is an issue of trust rather than raw capability. Nottebohm notes that the question is now whether agents have access to the right protected content, because an agent is only as good as the content it can reference and as safe as the security around it. A proper content layer also lets agents work across previously siloed departments. Among obstacles, 24% cite data fragmentation, 21% lack adequate permissions, and 18% overly disorganized content. Mature organizations treat unstructured documents and contracts as a competitive advantage.
Sponsored Protocol
Data exposure incidents drive stronger governance
Nearly half of all organizations have already experienced an AI-related data exposure incident, rising to 60% among leading-edge companies. However, these incidents act as a forcing mechanism. The share of organizations with established or advanced governance frameworks rose from 24% to 73% in one year, but gaps remain: only 39% have comprehensive visibility across sanctioned and unsanctioned AI use, 34% have formal standards for agent data access, and 27% still describe governance as ad hoc. Nottebohm points out that 93% of respondents said better governance actually let them move faster. Governance does not slow down but makes AI scaling sustainable and creates a multiplier effect.
Sponsored Protocol
Enterprise permissions must be rethought for AI agents
Permission structures built for humans need to be revisited with agents in mind. Companies have started setting document permissions considering how an agent might use them, but an entire corpus of unstructured data remains to be reviewed. The shift is from retrofitted human governance to governance built specifically for agents, tracking every touch, applied permissions, and sources used.
Companies want to avoid lock-in with a single AI vendor
Sixty-eight percent of enterprises are concerned about depending on a single AI provider, the average number of officially adopted AI tools has climbed to 3.3, and 79% consider it important that agents operate headlessly. Flexible architecture is built on platform interoperability, with multiple models and swappable stack components. Organizations no longer default to the biggest, most expensive model but choose the cheapest that meets quality needs. As Nottebohm says, the days of token-maxing are gone.
Sponsored Protocol
Next steps for AI success
Over the next three years, businesses should prioritize organizing, classifying, and cleaning up unstructured content, hiring teams for emerging roles, and adopting a hybrid token-compute budget model where IT owns core infrastructure and business units own application spend. According to Nottebohm, you do not have to start from scratch: by building governance, content layer, and multi-model system from the start, you can enter as a leading company and capture the same outsized impact.
For more on enterprise AI, read how Google updates Android Bench with new AI models and explore SpaceX Grok 4.5 at half the price. The full Box report is available on VentureBeat.
