OpenAI published its first B2B Signals report on May 6, revealing that companies at the 95th percentile of AI usage now consume 3.5 times more AI intelligence per worker than median firms — a gap that has nearly doubled from 2x just twelve months ago. The kicker: message volume accounts for only 36% of that divide. The rest comes from depth, complexity, and a dramatic shift toward agentic workflows where frontier firms outpace laggards by 16x.
The gap isn't about access — it's about depth
The conventional wisdom around enterprise AI has focused on deployment: how many seats, how many employees have access, how many pilots are running. OpenAI's B2B Signals report argues this framing is outdated. Access is table stakes. The question that now matters is whether organisations are using AI deeply enough to keep pace with the frontier.
OpenAI's COO Brad Lightcap has been blunt about this disconnect. "We have not yet really seen AI penetrate enterprise business processes," he stated, despite powerful AI systems being widely available. The B2B Signals data quantifies what that penetration gap actually looks like.
The report decomposes the gap and finds that even if a typical firm sent messages at the same rate as a frontier firm, it would close barely a third of the intelligence divide. Workers at frontier companies don't just use AI more often — they ask it to take on more complex work, provide richer context, and generate more substantive outputs.
Agentic workflows: the 16x dividing line
The most striking finding is the tool-by-tool breakdown. Codex — OpenAI's autonomous coding agent that can work across files, repositories, and multi-step tasks — shows a 16x usage gap between frontier and typical firms. ChatGPT Agent, Apps, Deep Research, and GPTs also show large gaps.
By contrast, simpler tools like file upload, web search, and data analysis show much smaller divides. The pattern is clear: the frontier advantage is most pronounced in tools that require delegating real work to AI rather than using it as a static assistant.
This isn't just an abstract metric. Cisco embedded Codex into production engineering workflows and achieved measurable results: 1,500+ engineering hours saved per month, approximately 20% reduction in build times, and a 10-15x increase in defect resolution throughput. Cisco's team described the key insight as treating Codex as "part of the team" — not a tool you query, but a colleague you delegate to.
Rakuten reported reducing mean time to recovery by approximately 50% after deploying Codex across engineering operations, while Balyasny Asset Management compressed research workflows from days to hours — a central-bank speech analysis that previously took two days now completes in about 30 minutes.
Why this matters more in Australia
The AI intelligence gap that OpenAI describes globally is arguably more acute in Australia. We've previously reported on how just 1% of Australian employers drive two-thirds of AI hiring — and the B2B Signals data explains the mechanism behind that concentration. Deloitte's State of AI in the Enterprise survey found that only 12% of Australian leaders report AI is already transforming their business or industry, compared to 25% globally. Just 65% of Australian respondents plan to increase AI investment in the next financial year, versus 84% globally.
The pilot-to-production pipeline is particularly clogged: only 28% of Australian respondents have moved at least 40% of their AI pilots into production. For most Australian organisations, AI remains in experimentation mode — precisely the shallow usage pattern that OpenAI's data identifies as the losing position.
There is a bright spot: 69% of Australian organisations report already using autonomous AI agents in some capacity. But only 22% have advanced governance models for those agents — and without governance, as OpenAI's report argues, agentic AI can't scale safely. This aligns with IBM CEO Arvind Krishna's observation at Think 2026: "The enterprises pulling ahead are not deploying more AI — they're redesigning how their business operates."
The compounding problem
What makes this data particularly urgent is the compounding dynamic. The gap grew from 2x to 3.5x in twelve months. Frontier firms aren't just ahead — they're accelerating away. Their deeper AI usage creates better internal knowledge about what works, which breeds more sophisticated use cases, which widens the gap further.
OpenAI's data shows this playing out in education and learning tasks, where frontier firms send 7x more messages per worker than typical firms. Leaders are using AI to train their people on AI — creating a flywheel where capability builds on capability. Organisations that haven't embedded AI into learning and development are missing this multiplier effect entirely.
The industry breakdown adds nuance. Professional, Scientific, and Technical Services ranks first in both Codex adoption and API intensity. Finance and Insurance leads in ChatGPT seat deployment. Educational Services shows the highest per-person message intensity. There's no single AI leaderboard — but organisations that aren't leading on any dimension should be asking why.
What to do about it
OpenAI's report offers five practices that distinguish frontier firms, and they're worth taking seriously given the data behind them:
- Measure depth, not just access. Track whether AI use is becoming more complex and tied to valuable workflows — not just how many seats are active.
- Build governance that enables agentic deployment. The goal isn't to restrict AI agents but to make them deployable at scale with clear rules about scope, information access, and human oversight.
- Treat enablement as infrastructure. Frontier firms invest in continuous learning — role-specific training, hackathons, internal champion networks, and shared workflow repositories.
- Find your frontier teams and scale their patterns. In most organisations, advanced usage is concentrated in a few teams. Identify them and replicate their conditions.
- Move beyond chat to delegated work. The biggest gap is in agentic tools. The shift from "ask AI a question" to "delegate AI a task" is where the compounding advantage lives.
For a 10-50 person company, the most actionable insight is the last one. OpenAI's own workspace agents — which automate lead qualification, reporting, and IT triage — are a concrete example of what this shift looks like in practice. You don't need enterprise-scale infrastructure to start delegating work to AI agents. But you do need to move past treating AI as a search engine or writing assistant and start treating it as a team member that can own multi-step workflows. The firms doing this are pulling ahead at an accelerating rate — and twelve months from now, the gap will be wider still.
Sources
- B2B Signals — OpenAI
- Cisco and OpenAI Redefine Enterprise Engineering with AI Agents — OpenAI
- Think 2026: IBM Delivers the Blueprint for the AI Operating Model as the AI Divide Widens — IBM Newsroom
- The State of AI in the Enterprise 2026 — Deloitte Australia
- OpenAI COO Says 'We Have Not Yet Really Seen AI Penetrate Enterprise Business Processes' — Yahoo Finance
