Agentic AI
,
Artificial Intelligence & Machine Learning
,
Next-Generation Technologies & Secure Development
Token Volume Climbs While Business Outcomes Remain Unclear
Jennifer Lawinski •
August 12, 2026
OpenAI reports that enterprise customers are shifting token spend from chatbot conversations to agents deployed across systems and workflows. (Image: Shutterstock)
Enterprise use of OpenAI’s models is transitioning from conversing with chatbots to deploying agents into systems, data and repeatable workflows, according to recent company data.
See Also: Why Traditional DLP Can’t Keep Up With AI Data Growth
OpenAI called the development a move from “assistance to delegation, giving agents the context and tools to complete complex tasks, and accelerating agentic artificial intelligence beyond software development.”
As of June 2026, Codex generated 64% of the combined Codex and ChatGPT output tokens among OpenAI enterprise customers, which the company cited as evidence of the shift to agentic uses.
OpenAI cites no data linking increased token consumption and agentic AI use to business value, but it classifies the top 10% of enterprise customers by output tokens per active user as “frontier firms.” It calls customers in the 45th to 55th percentiles “typical firms.”
In June, firms in OpenAI’s frontier tier generated 8.3 times as many output tokens per active user as typical firms, up from 2.6 times in January. This varied by industry, with the largest gap between frontier and typical users found in the information and technology industry and the smallest in manufacturing.
OpenAI acknowledged that token consumption is an “imperfect measure of business value,” adding that “a short response can be highly valuable, while a long one may add little. But token volume offers a useful proxy for the depth of AI use and how much work employees are asking AI to take on.”
OpenAI’s frontier firms made greater use of capabilities that give AI context and tools than typical firms. Among weekly active users, 21% used plugins, compared with 9% at typical firms. At frontier firms, 19% of weekly active users implemented reusable skills, compared with 3% at typical firms.
Agents Move Beyond Coding
While software engineers and tech teams were early adopters of agentic AI tools, OpenAI data shows that agents are proliferating beyond coding and IT.
Since February, OpenAI reported that weekly active enterprise Codex users increased 108 times in legal, 41 times in sales and recruiting, and 26 times in marketing, compared with five times in engineering. OpenAI did not disclose user counts.
OpenAI also released a companion working paper, “How Organizations Use AI: Evidence from ChatGPT,” that found that ChatGPT Enterprise output tokens grew about seven times between June 2025 and March 2026. That growth was driven by both new firms adopting the technology and a fourfold increase in usage by firms that were customers in June 2025.
Early adoption was concentrated among larger, more highly valued and more R&D-intensive U.S. public companies and was associated with previous investments in organizational capital, R&D and software, the report found. The analysis didn’t establish that those characteristics caused adoption.
The authors said larger, more organizationally intensive firms may be better positioned to identify valuable use cases and integrate the technology into existing workflows. But among adopters, larger firms recorded lower measured usage per employee, complicating the assumption that organizational scale necessarily produces deeper adoption.
Adoption Is Not Transformation
Other recent research illustrates the murky relationship between token consumption and business value.
Glean, which surveyed 6,000 digital workers in the United States, United Kingdom and Australia for its global Work AI Index report, found that 75% of respondents said AI had made them more productive and estimated it saved them about 11 hours per week. But only 13% of respondents said their organizations were performing significantly better as a result. Workers also reported spending 6.4 hours a week “botsitting” AI by supplying context, supervising output and correcting errors.
Glean found workers spent 2.3 hours weekly feeding AI context, and 53% said critical information was inaccessible through their AI systems.
AI strategist Sol Rashidi, chief strategy officer at Cyera and a senior fellow at Harvard Kennedy School, told Business Insider that she recently got rid of two of the four agents she had been using because they demanded constant supervision.
“I just fired half my agents because they were unreliable,” Rashidi said. “I was spending more time babysitting them.”
Atlassian found a similar trend emerging from its survey of more than 12,000 knowledge workers and 173 Fortune 1000 executives. The survey found that while 89% of executives said AI increased speed, only 6% could identify concrete organization-wide ROI.
“Too many companies are treating AI adoption like a vanity metric – more seats, more prompts, more usage,” said Rebecca Hinds, head of Glean’s Work AI Institute, in a statement.
“Adoption alone doesn’t equal transformation. If employees are spending the productivity dividend on botsitting and botshitting, companies haven’t eliminated work – they’ve created a new layer of overhead,” she said. “The organizations that win will be the ones that ground AI in real enterprise context, apply the right guardrails and measure success by business outcomes, not activity.”