How frontier enterprises are building an AI advantage
Key Points
- Frontier firms use 3.5x intelligence per worker
- Codex shows ~16x message gap for frontier firms
- Prioritize depth: governance, enablement, and agentic workflows
Summary
Frontier firms (95th percentile) now use ~3.5x as much AI intelligence per worker as typical firms (up from 2x in 2025). That gap is driven mainly by depth — richer prompts, more tokens, and more complex, delegated workflows — not merely message volume (which explains ~36% of the difference). Advanced, agentic tools (notably Codex) show the largest gaps, indicating the competitive edge comes from embedding AI into production workflows and delegating multi-step work to agents.
Key Points
- Metric: tokens generated used as a proxy for depth of AI work; frontier = 3.5x intelligence/worker.
- Volume vs depth: message volume explains ~36% of the frontier gap; most of the advantage is deeper, more complex usage.
- Agentic tools: Codex traffic is ~16x higher per worker at frontier firms; ChatGPT Agent, Apps in ChatGPT, Deep Research, and GPTs follow similar patterns.
- Production impact: Cisco reported ~20% faster builds, 1,500+ engineering hours saved/month, and 10–15x defect-resolution throughput using Codex in production.
- Product integrations: companies are embedding AI in in-app assistants, dev tools, and customer support (e.g., Travelers’ AI Claim Assistant expected to handle ~100k FNOL calls in year one).
Practical guidance for engineering teams
- Measure depth, not just seats or message counts: track tokens per worker, task complexity, and end-to-end outcomes.
- Enable safe production use: build governance (model/version controls, access controls, auditing) that enables, not blocks, adoption.
- Treat enablement as infrastructure: invest in training, templates, and shared components so teams can apply AI to domain-specific workflows.
- Start small, scale what works: identify frontier teams, validate impact, then productize or integrate successful agents into systems.
- Move from chat to delegation: prioritize agentic workflows that can take multi-step actions, operate on code/files, and integrate with internal systems.
Why engineers should care
Depth and delegation are where measurable ROI emerges. Focusing on production readiness, observability, and developer enablement will unlock compound advantages as AI becomes capable of executing more complex, long-horizon tasks.