On May 6, 2026, OpenAI published the first edition of B2B Signals, a recurring data report on how enterprises actually use AI. The headline finding: frontier firms in the top 5% of usage now use 3.5x more intelligence per worker than typical firms, and most of that gap is not about sending more messages. It is about asking AI to do harder, more complex work.
Why This Report Matters
Most enterprise AI research is built on surveys, where executives report what they think is happening. B2B Signals is different. It is based on privacy-preserving analysis of actual token generation and product usage across OpenAI's enterprise customer base. That gives it a level of empirical grounding most adoption studies do not have.
According to OpenAI's B2B Signals dashboard, the frontier firm (defined as the 95th percentile of intelligence usage per worker) is now using 3.5x as much AI as the typical firm (the 50th percentile), up from 2x just one year ago. That single number tells a clear story. The leaders are not standing still. They are pulling further ahead, and they are doing so faster than the rest of the field is closing the gap.
Our take: This is the empirical complement to PwC's recent finding that 20% of companies are capturing 74% of AI value. PwC measured outcomes. OpenAI is measuring inputs. Both data sets point to the same conclusion: a small group of companies is using AI in qualitatively different ways, and the rest are falling behind.
The Depth Gap, Not the Volume Gap
The most important data point in the B2B Signals report is not the 3.5x ratio. It is the breakdown of where that ratio comes from.
OpenAI calculates that only 36% of the frontier advantage is explained by message volume. The remaining 64% comes from depth: longer prompts, richer context, more complex outputs, and delegated work routed to agents rather than handled in chat. In practical terms, frontier firms are not just using AI more often. They are using it differently.
This matters because the typical enterprise AI strategy is built around volume metrics. License counts. Seat penetration. Messages per active user. None of those metrics capture what OpenAI's data shows is the real driver of value. A company can roll out ChatGPT to 100% of its workforce and still sit at the 50th percentile if those workers are only using AI to summarize emails or rewrite paragraphs.
Where the Gap Is Widest
The B2B Signals report breaks down usage by tool category, and the differences are striking.
Codex shows the largest gap. Frontier firms send 16x as many Codex messages per worker as typical firms. Codex is OpenAI's agentic coding tool, which means engineers describe a task and the AI completes it autonomously over multiple steps rather than answering one question at a time. The 16x ratio is a signal that engineering organizations adopting agentic development are operating in a different productivity regime from peers still using AI as autocomplete.
Education and learning workflows show a 7x gap. Frontier firms use AI 7x more per worker for explaining concepts, generating training material, and answering domain-specific questions. This suggests leaders are treating AI as a continuous learning layer for their workforce, not just a tool.
Coding more broadly shows a 4x gap. This includes ChatGPT-based coding assistance that is not Codex-specific. The gap is smaller than the Codex-specific gap, which underscores the point: the deeper and more delegated the workflow, the wider the lead.
ChatGPT Agent, Apps in ChatGPT, Deep Research, and custom GPTs all show similar directional patterns. The common thread is that frontier firms are better at adopting tools that move work from "assistant" mode to "delegated agent" mode. For background on why this shift matters, see our explainer on what AI agents are and whether your business needs one.
What This Means for Business Leaders
The B2B Signals data has three concrete implications for any company taking AI seriously.
Stop measuring AI adoption by license count. If your AI dashboard tracks seats sold and weekly active users, you are measuring the wrong things. A 100% adoption rate at shallow depth still puts you at the 50th percentile of value capture. Leaders are tracking complexity of tasks, percentage of work delegated rather than chatted, and the ratio of agentic tool usage to chat usage.
Treat enablement as a strategic line item, not a launch activity. Most enterprises run one training session at rollout and then assume usage will deepen on its own. The B2B Signals data suggests it does not. Frontier firms invest continuously in showing workers how to provide richer context, decompose complex tasks, and hand off multi-step work to agents. This is a workflow redesign problem, not a training problem.
Move budget toward agentic tools. If your AI spend is concentrated in chat seats with little in agentic or coding tools, you are likely under-indexed on the highest-leverage category. The 16x gap on Codex is not a curiosity. It is a leading indicator of where value is concentrating.
The urgency is reinforced by OpenAI's commercial commentary. In a May 11 CNBC interview, OpenAI Chief Revenue Officer Denise Dresser said enterprise AI adoption is "at a tipping point," with enterprise now representing more than 40% of OpenAI's revenue. The shift to deeper, agent-driven usage is not theoretical. It is showing up in the buyer behavior of OpenAI's own book of business.
How to Move From the 50th to the 95th Percentile
A practical sequence for closing the depth gap.
-
Audit your current AI usage by depth, not volume. Sample 50 typical interactions across your workforce. Tag each as "simple chat," "complex prompt with context," or "delegated multi-step work." If less than 20% are in the third bucket, that is where your improvement opportunity sits.
-
Identify three workflows ready for delegation. Look for repetitive, multi-step processes where workers already use AI for parts. Engineering tickets, research synthesis, customer ticket triage, and document review are common candidates. These are the workflows that benefit most from agentic tools.
-
Pilot at least one agentic tool inside a real workflow. Codex for engineering, Deep Research for analysts, or a custom agent for a specific business process. Measure cycle time and quality, not just adoption.
-
Build governance that allows autonomous operation. Agents that need human approval at every step do not produce frontier outcomes. Define the guardrails (data access, action scope, escalation rules) so that delegated work can actually run without micromanagement. The connection to safe scaling is covered in our notes on why most AI projects stall before production.
-
Track depth metrics quarterly. Tokens per worker, percentage of agentic interactions, and average task complexity should appear in your AI program review alongside license counts and ROI.
Common Pitfalls
Confusing tool variety with depth. Buying ChatGPT, Claude, and Gemini seats does not increase depth if all three are used the same shallow way.
Treating agents as risky and chat as safe. The B2B Signals data suggests the opposite is the strategic risk. Companies that stay in chat mode are falling behind.
Underestimating context engineering. A meaningful portion of the depth gap is workers providing richer context in their prompts. Templates, retrieval, and integrations that surface internal context automatically are high-leverage investments.
Assuming the gap will close on its own. It is widening, not narrowing. The 2x to 3.5x shift in twelve months is the rate at which leaders are pulling away.
Key Takeaways
- OpenAI's May 6, 2026 B2B Signals report shows frontier firms use 3.5x more intelligence per worker than typical firms, up from 2x a year ago.
- Only 36% of the gap is explained by message volume. The rest comes from depth: richer prompts, more complex tasks, and delegated agentic work.
- The gap is widest in agentic tools, with Codex showing a 16x ratio between frontier and typical firms.
- Adoption metrics based on license counts or seat penetration miss the real driver of value. Depth metrics are what separate leaders from the rest.
- Closing the gap requires investment in enablement, agentic tools, and governance that allows autonomous operation.
The businesses that move early on agentic AI workflows will have a meaningful advantage. If you want to be one of them, let's start with a conversation.