OpenAI: leading companies make 3.5x deeper use of AI and deploy Codex more actively

OpenAI launched B2B Signals — a new research format exploring how businesses actually deploy AI within organizations. The first report, published on May 6, 2026, reveals a key finding: the gap between ordinary and leading organizations is growing rapidly, and the primary factor isn't access to tools, but the depth of their usage.

The Gap Is Growing

In its report, OpenAI compares "frontier" companies — organizations at the 95th percentile of AI adoption — with typical companies at the median level. The difference is already striking: leaders deploy 3.5 times more "intelligence" per employee than average businesses. A year ago, this gap was half as wide — around 2x.

The report's logic is straightforward: access to ChatGPT or enterprise licenses alone no longer creates competitive advantage. What matters more is how deeply AI is embedded in teams' real work.

OpenAI specifically emphasizes that message volume is not the primary indicator of maturity. It explains only 36% of the gap between leaders and "mid-market" companies. The rest comes from more substantive use: employees provide models with more context, tackle more complex tasks, and get more valuable results. In other words, typical companies still often use AI as smart search or a helper for quick answers, while leading companies already delegate chunks of real, complex work to it.

The Agent Stage

OpenAI observed the most striking gap in agentic and advanced tools. According to the report, frontier companies send 16 times more messages per employee to Codex than typical companies. Similar dynamics appear with ChatGPT Agent, Apps in ChatGPT, Deep Research, and GPTs.

This represents an important shift: the market is moving away from the "ask-and-get-an-answer" scenario toward a model where AI helps write code, execute multi-step tasks, leverage internal company context, and deliver more complete results.

For businesses, this means more than just buying new licenses — it requires process redesign. The more capable AI becomes at working with files, codebases, and long chains of actions, the more critical governance, access controls, and the habit of delegating full tasks rather than snippets of text become.

OpenAI explicitly states that leaders are building "operational muscle": they use AI not as a faster interface, but as a way to reorganize how teams work.

Company Practices

The report shows that AI adoption is becoming simultaneously widespread and specialized. The broadest scenarios currently involve writing and communication, but differences emerge by function: IT and security teams more often use models for procedural instructions, developers and data science teams for code, and finance teams for analysis and calculations.

There's no single leaderboard here: one company might be stronger in ChatGPT, another in APIs, a third in Codex or internal AI services.

• Cisco reports approximately 20% reduction in build time using Codex in production scenarios.
• The company also estimates savings of over 1,500 engineering hours per month.
• Cisco's bug-fix throughput increased 10–15 times.
• Travelers expects its AI Claim Assistant to process approximately 100,000 initial insurance claims in its first year.

OpenAI also highlights another marker of maturity: leaders use AI not just to complete tasks, but to train employees. In education and learning scenarios, the gap between leading and typical companies turned out to be among the most dramatic. This is a positive signal for large organizations: advantage emerges not when a tool is simply distributed to everyone, but when people learn to formulate complex queries, trust agents with parts of processes, and regularly return to these workflows in their daily work.

What This Means

The key takeaway from B2B Signals is that next-generation corporate AI competition won't be about user count, but adoption quality. If OpenAI's data is right, advantage goes to companies that move AI from helper mode to executor mode first — especially in development, analytics, support, and other functions where results can be measured in hours saved, speed, and throughput.