Chart of the Day
1. Key Themes
AI Adoption Is Rapidly Spreading Beyond Engineering Into All Business Functions
The chart tracks "Average user's share of monthly output tokens generated on OpenAI Codex (%)" across departments, showing that by June 2026, Engineering leads at 99%, but Finance (91%), Recruiting (89%), and Legal (88%) have nearly caught up — all converging toward full adoption within months of each other.
"Codex adoption has broadened from engineering into nearly every function across OpenAI."
The S-Curve of Enterprise AI Adoption Is Steepening and Accelerating
The chart shows a classic diffusion pattern: Engineering began adopting Codex in August 2025 and reached ~99% by June 2026 (~10 months). Non-engineering functions (Finance, Recruiting, Legal) started near zero as late as early 2026 but surged to 88–91% in just 3–4 months — a dramatically steeper adoption curve for later-adopting departments.
"Is this a predictor for the rest of the enterprise world?"
OpenAI Itself Is a Leading Indicator for Enterprise AI Diffusion
Coatue frames OpenAI's internal Codex usage data as a forward-looking signal for how adoption will play out across corporate America — suggesting that what happens inside AI-native companies today is a template for the broader enterprise tomorrow.
"Codex adoption has broadened from engineering into nearly every function across OpenAI. Is this a predictor for the rest of the enterprise world?"
2. Contrarian Perspectives
Codex Is Not Just a Developer Tool — It Is a Universal Enterprise Workflow Tool
The consensus view is that coding assistants like Codex are primarily engineering productivity tools. The data directly challenges this: Finance at 91%, Recruiting at 89%, and Legal at 88% adoption rates suggest that the output of an AI coding agent is being meaningfully used by non-technical functions. This implies the total addressable market for agentic AI tools is far larger than the developer-productivity framing suggests.
"Codex adoption has broadened from engineering into nearly every function across OpenAI."
Late Adopters Are Actually Adopting Faster, Not Slower
The conventional diffusion-of-innovation model assumes laggards adopt slowly. The chart inverts this: Finance, Recruiting, and Legal went from near-zero to ~90% in roughly 3–4 months (Feb–June 2026), while Engineering took ~10 months to reach ~99%. Later-moving functions may benefit from organizational learning, tooling maturity, and top-down mandates — compressing the adoption timeline dramatically.
(Source: OpenAI internal data, as charted by Coatue)
3. Companies Identified
| Company | Description | Why Mentioned | Quote |
|---|---|---|---|
| OpenAI | Developer of GPT models and Codex | Used as the primary case study; internal Codex adoption data sourced directly from OpenAI | "Codex adoption has broadened from engineering into nearly every function across OpenAI." Chart sourced: "Source: OpenAI" |
4. People Identified
No specific individuals are named in the article or chart.
5. Operating Insights
Don't Limit AI Tool Rollouts to Engineering — Deploy Cross-Functionally and Early
The data shows that once non-engineering teams at OpenAI began using Codex, adoption accelerated faster than it did for engineers. Operators should resist the instinct to gate AI tools behind technical teams and instead push broad, cross-functional deployment — Finance, Legal, and HR are high-ROI targets.
"Codex adoption has broadened from engineering into nearly every function across OpenAI."
Expect a Sudden Inflection, Not a Gradual Ramp, When Non-Technical Teams Adopt AI
The chart shows flat-to-zero usage for non-engineering functions through late 2025, followed by a near-vertical climb in early 2026. Operators should plan infrastructure, licensing, and change management for a sudden surge rather than a linear rollout.
(Derived from chart data showing near-zero adoption through Feb 2026, then rapid climb to 88–91% by June 2026)
6. Overlooked Insights
The "Gray Lines" Suggest Many More Functions Are Being Tracked
The chart shows numerous unlabeled gray lines between the highlighted departments, suggesting OpenAI tracks Codex adoption across a wide array of functions beyond the four called out. This implies the cross-functional saturation may be even broader than the headline figures suggest — and that Coatue has access to granular, department-level AI usage data that is not yet publicly discussed.
(Visible in chart image; not referenced in article text)
Token Output — Not Seat Count — Is the Adoption Metric
The Y-axis measures "average user's share of monthly output tokens generated," not simply the number of users or logins. This is a usage-intensity metric, meaning the chart reflects how deeply individuals are integrating Codex into their actual workflows — a far stronger signal of genuine adoption than license activation rates.
(Source: Chart axis label, Coatue/OpenAI)