On May 6, 2026, OpenAI introduced B2B Signals, a new enterprise research stream designed to measure how AI is actually diffusing across firms. That matters because the conversation is shifting. The market is moving away from a simple adoption question and toward a workflow question: is AI becoming part of how work gets done?
OpenAI's core observation is direct. Frontier firms, defined as organizations at the top end of usage, now consume 3.5 times as much intelligence per worker as typical firms, up from 2 times a year earlier. Just as important, the gap is not mainly about sending more messages. Message volume explains only about 36% of the difference. Most of the advantage comes from deeper, more complex, higher-context use.
That shift is especially relevant for SMEs. Many businesses are still at the stage of asking whether staff use ChatGPT to help write faster. OpenAI is already measuring something more operational: agentic workflows. The article says the largest frontier advantage appears in advanced tools, with frontier firms sending 16 times as many Codex messages per worker as typical firms. In other words, the leaders are not just asking AI questions. They are delegating chunks of work, asking AI to analyze, draft, and produce outputs within defined boundaries.
OpenAI also makes a broader point: the next phase of enterprise AI is no longer about seat count alone. It is about whether a business has created repeatable, governable, scalable operating patterns. The article points to several behaviors common among leaders: measuring depth, building governance for production use, investing in enablement, scaling what works, and moving from chat-based assistance to delegated work with agents.
Another useful angle in B2B Signals is that business context is starting to matter more. Usage is becoming specialized by function. IT and security lean toward procedural guidance, while software development and data science lean heavily toward coding. That means enterprise maturity should not be measured by how many employees have access. It should be measured by whether AI is reaching the highest-value parts of each department's work.
From a VMTS perspective, this reinforces a practical point: AI Agent design, automation, websites, CRM, and internal data flow need to be planned together. If a business still keeps information scattered across WhatsApp, email, forms, spreadsheets, and manual reporting, stronger models alone will not create delegated workflows. Real value comes when enquiries, classification, follow-up, content, reporting, and internal knowledge all move through one operating chain.
The most important takeaway from OpenAI's update is not a single metric. It is that the benchmark has changed. In 2025, many companies were still asking whether staff should get AI access. In 2026, the sharper question is which parts of work can now be handed to AI safely first, while humans focus on direction, approval, and exceptions. That is the line between using AI as a tool and using AI as an operating model.
