Two stories separated by 24 hours tell the same tale from opposite angles: AI is entering companies at a speed that even the companies themselves did not expect.
On June 2, 2026, OpenAI published “The Next Era of Knowledge Work,” announcing that Codex surpassed 5 million weekly active users — a 6x growth since the desktop app launch in February. The most revealing figure is not the total, but the composition: knowledge workers (non-developer knowledge workers) make up 20% of users and are adopting Codex more than three times faster than developers.
One day earlier, on June 1, Bloomberg reported that Walmart capped usage of its internal assistant Code Puppy, replacing unlimited access with a per-employee token allocation. The reason: excessive demand and rising costs.
Adoption is exploding. And with it, costs.
Codex: From Code Tool to Knowledge Platform
That Codex has 5 million weekly users is impressive. That knowledge workers are adopting it 3x faster than developers is a paradigm shift.
OpenAI describes these users as professionals who don’t write code as their primary function — analysts, product managers, operations, HR — who use Codex to automate tasks, analyze data, generate reports, and build workflows. The expansion includes new capabilities like “Sites” (AI-generated web applications) and a plugin system that extends Codex to enterprise tools.
GitHub Copilot, meanwhile, reached 20 million total users and 4.7 million paid subscribers (75% year-over-year growth). Microsoft 365 Copilot reached 20 million paid enterprise seats in May 2026, a 33% increase since January. But with 450+ million commercial M365 seats, penetration remains at only ~4.4%.
Walmart Code Puppy: The First Sign of a Limit
Walmart’s decision to cap Code Puppy tokens is a symptom of a problem many companies are beginning to face: AI demand is outstripping what budgets can sustain.
Code Puppy is Walmart’s internal assistant that helps employees with tasks ranging from programming to HR queries. It previously ran with unlimited tokens. Now each employee has a fixed allocation. Walmart explicitly cited “high demand” and “cost management” as the reasons.
This is not an isolated case. On the same day, June 1, 2026, GitHub Copilot migrated to usage-based billing for all users. Average enterprise AI spend rose from ~$7 million in 2025 to a projected $11.6 million in 2026 — a 65% increase. Consulting firm McKinsey projects global AI system spending will exceed $300 billion in 2026.
The Inevitable Tension
These data points reveal a structural tension. On one hand, the productivity these tools offer is real: Codex isn’t growing 6x because of marketing, but because users find value. On the other hand, the “all-you-can-consume” business model is not sustainable when demand scales faster than infrastructure.
Companies are in an uncomfortable position. AI promises efficiency, but requires investment. And the investment isn’t fixed — it scales with usage. A tool that saves 10 hours of work but costs more than those 10 hours is not a saving.
Walmart solved it with a cap. GitHub Copilot with usage-based billing. Microsoft 365 Copilot with a fixed $30/user/month price that many consider high for the 4.4% penetration it has.
No one has yet found the right pricing model for enterprise AI. What is clear is that the unlimited “free tier” was not sustainable. And the decisions Walmart, OpenAI, Microsoft, and GitHub make in the coming months will define how AI adoption is economically structured in enterprises for the rest of the decade.
Sources: OpenAI — The Next Era of Knowledge Work · Bloomberg — Walmart caps Code Puppy tokens · The Next Web — Codex expansion · BrainStorm — M365 Copilot 20M seats · Panto AI — Codex Statistics 2026