Enterprise AI is finding real traction inside large companies, but mostly in a narrow band of work that is easier to measure, automate and verify, according to a recent report.
Andreessen Horowitz, a US-based venture capital firm, said 29% of the Fortune 500 and about 19% of the Global 2000 are now live, paying customers of a leading AI startup, based on deployments involving top-down contracts, converted pilots and live use inside the organization.
High momentum in bounded workflows
However, the progress is not spread evenly across the enterprise. Andreessen Horowitz said coding, customer support and search account for the largest share of enterprise AI use cases today, with coding the clear outlier, while technology, legal and healthcare are the sectors showing the strongest early demand.
The firm said it mapped revenue momentum by use case against OpenAI’s GDPval benchmark. OpenAI says GDPval measures model performance on 1,320 specialized tasks across 44 occupations, with tasks based on real work products.
Andreessen Horowitz said support has proved especially receptive because the tasks are time-bound, the intent is constrained and the outputs are easier to judge through metrics such as ticket volumes, customer satisfaction and resolution rates.
More broadly, it saw that the sectors moving fastest tend to be text-based, repetitive, reviewable and easier to verify, while adoption is slower in work shaped by physical-world constraints, heavier regulation, interpersonal judgment or multi-party coordination.
The shift toward operational discipline
A separate 2026 survey from WRITER and Workplace Intelligence suggests that once companies move beyond pilots, the harder problem becomes operational discipline. The study surveyed 2,400 knowledge workers across the U.S., U.K. and Ireland, Benelux, France and Germany, including 1,200 C-suite executives and 1,200 employees.
It focused on people already using generative AI at work, or executives at companies that permit it, giving it a view into organizations that are already past the awareness stage.
That survey found adoption is already embedded in daily work for many respondents. WRITER and Workplace Intelligence reported that 94% of C-suite respondents use AI tools at least 30 minutes a day, and 64% said they spend at least two hours a day with generative AI tools or AI agents.
The same report found that 52% of employees said they had used AI agents over the past year, while 97% of executives said their company had deployed AI agents in that period.
High usage meets uneven returns
The same organizations, however, are struggling to prove broader returns. WRITER and Workplace Intelligence found that only 29% of C-suite respondents said they had seen significant ROI from generative AI, while 48% said AI adoption at their company had been a “massive disappointment.”
It also found that 39% said their company still did not have a formal AI strategy to drive revenue, suggesting that usage is scaling faster than operating models are maturing.
Data-risk concerns and organizational silos
Control gaps ran through the survey as well. WRITER and Workplace Intelligence said 35% of employees admitted entering proprietary, confidential or sensitive company information into a public AI tool.
At the leadership level, 67% of executives said they believed their company had already suffered a data leak or security breach because of an employee using an unapproved AI tool. The survey also found that 79% of executives said AI applications were being deployed in silos, and 54% said adopting AI was “tearing their company apart.”
Moving beyond the pilot phase
Taken together, the two reports point to an enterprise market that is no longer stuck at the pilot stage, but is still far from settled. The clearest progress is showing up in a handful of workflows where AI can be inserted into defined tasks and judged against visible outcomes.
The broader challenge now is whether companies can build the governance, security controls and operating discipline needed to turn those early wins into something durable.