When 2.1 million employees are told to use AI freely, the invoice eventually lands harder than the productivity gain.

Unlimited AI access was always a budget fiction

Enterprises handed workers open-ended AI tools measured by adoption, not spend. Now that LLM providers are billing per token rather than per seat, every spreadsheet summary and presentation draft carries a line-item cost that compounds at workforce scale.

Walmart’s internal AI tool now runs on a fixed allowance

Walmart’s AI workflows meet the realities of the balance sheet is Walmart’s internal AI assistant, built to handle tasks like spreadsheet analysis, slide creation, and general office automation. Employees open the tool, submit a request, and draw down from a fixed token allocation assigned by the company. When the allocation runs out, so does the access.

The workers most exposed to this policy shift

Token caps do not hit everyone equally. The pain concentrates in specific roles.

• Operations analysts who run repeated data queries against large files and will hit their cap mid-project
• Team leads responsible for weekly reporting who relied on Code Puppy to generate summaries and now must triage which tasks are worth the token spend
• HR and training staff who were encouraged to build AI-assisted onboarding materials and now face throttled output during high-hiring periods

Walmart has reportedly given employees guidance on choosing the right AI tool for each task, which suggests some access to alternative platforms remains, but rationing has arrived regardless.

Token maxxing became a KPI before it became a cost center

A Sequoia Capital partner publicly encouraged token maxxing as recently as April this year, and internal AI leaderboards at major companies rewarded the employees racking up the most complex AI interactions. That measurement logic now directly inflates the bills that CFOs are reviewing, and enterprises that tied performance reviews to AI usage volume have built a structural incentive to overspend.

What employees can still do within a token budget

• Prioritize high-output tasks like draft generation over low-yield lookups
• Batch related questions into a single prompt to reduce total token draw
• Reserve AI use for tasks where manual effort exceeds thirty minutes
• Audit weekly token usage before sprint deadlines hit the cap

The shift in policy at Walmart is not an isolated IT decision. It is a preview of the governance layer every large enterprise will need to build.

Pricing is now Walmart’s internal problem, not a vendor decision

Pricing for Code Puppy is an internal cost borne by Walmart, allocated to employees as token budgets rather than a per-user subscription rate.

The real limitation here is organizational: token limits without task guidance create arbitrary friction, and employees with legitimate high-volume needs have no obvious path to appeal their allocation.

Microsoft Copilot faces the same enterprise cost pressure as usage scales past pilot programs. Google Workspace AI is navigating an identical tension between encouraging adoption and controlling inference spend at headcount.

The subscription model for enterprise AI is already over

The move from flat-rate AI access to consumption billing is restructuring how companies measure the return on AI investment, and the tools that survive inside large organizations will be the ones that can prove value per token, not just per user. We cover tools like this every Friday — subscribe here and we’ll send the best ones straight to you.