Frontier AI Usage Gap Compounds Through Depth

Frontier firms—defined as the top 95th percentile of OpenAI enterprise users—consume 3.5x as much intelligence per worker as typical firms, measured by tokens generated as a proxy for work delegated to AI. This gap has widened from 2x in April 2025, with message volume accounting for only 36% of the difference. The rest stems from richer interactions: frontier workers provide more context, tackle complex tasks, and generate substantive outputs, shifting AI from simple Q&A to executing challenging work. Leaders track this depth to prioritize workflow redesign over mere access or frequency.

Agentic Workflows Mark Maturity and Largest Gains

Advanced tools reveal the biggest disparities, signaling a move to delegation. Frontier firms send 16x more Codex messages per worker for coding, with similar patterns in ChatGPT Agent, Apps in ChatGPT, Deep Research, and GPTs for multi-step tasks, company-context application, and complex analysis. Cisco exemplifies this: integrating Codex into engineering workflows cut build times 20%, saved 1,500+ hours monthly, and boosted defect resolution 10-15x by treating AI as 'part of the team.' As agents handle tools, files, codebases, and long-horizon tasks, early adopters build capabilities to offload meaningful work, redesigning operations from chat assistance to autonomous execution.

Specialization Across Functions and Paths to Frontier Status

AI adoption broadens into production via APIs for in-app assistants, dev tools, and support, with function-specific spikes: IT/Security on procedural guidance, dev/data science on coding, finance on analysis. No universal leaderboard exists—industries lead differently in ChatGPT, Codex, API, or intensity—offering entry points like scaling access or embedding AI in products. Travelers Insurance's API-built Claim Assistant handles policy queries, gathers data, and creates claims, targeting 100,000 first-notice calls yearly. Leaders close the gap by measuring depth, enabling production governance, prioritizing enablement (largest task-level edge in education/learning), scaling frontier teams, and advancing to agents. B2B Signals provides ongoing, privacy-preserved benchmarks from aggregated OpenAI data to guide this progression.