Full Breakdown
Companies Clamp Down on Employee AI Use Amid Soaring Costs
7/2/2026, 10:58:45 PM
Corporate AI Cost Controls
Mid-2024 internal Slack chats and emails from Atlassian, Adobe, Amazon, Citi, GitHub and Accenture show firms limiting employee use of LLMs—Citi disabled Claude Opus 4.6/4.7 and GPT-5.5, Adobe ended unlimited Claude access, and Atlassian removed unlimited AI dashboards—citing usage-based billing as the cause of spiraling token costs.
Background & Data
Early 2024 pricing changes moved AI providers to per-token billing. Atlassian’s dashboard shows spend rising from $5 million (Aug 2025) to >$15 million (May 2026), on track for >$120 million FY. Citi advises using lower-tier models—GPT-5.3-Codex for quick queries, Claude Sonnet 4.6 for code review—over high-credit Opus 4.7 and GPT-5.5.
Official Statements & Responses
Citi’s memo urged staff to “choose the right model for the task,” added monitoring, and said no models were disabled. Atlassian ended unlimited AI use, launched a spend-tracking dashboard, and disputed its $15 million spend figure. Adobe warned Claude access ends June 30. Amazon removed a leaderboard, imposed a token-limit notice, and a spokesperson said guidance “hasn’t changed.” GitHub is testing user-based billing and open-source alternatives. Accenture highlighted “soaring token spend” among clients and positioned itself as a consultant on “token economics.”
Criticism & On-the-Ground Feedback
An Atlassian staffer asked, “Now how do I do my job?” and called spending “insane.” Amazon workers reported a shift from a leaderboard to “actual usage limits in two weeks.” A developer noted a token drain with “no obvious ROI,” reflecting frustration.
Conflicting Reports & Gaps
Citi says no models were disabled, yet internal emails and screenshots show Claude Opus 4.6/4.7 and GPT-5.5 blocked. Atlassian disputes its $15 million spend figure without revised data. Amazon claims guidance unchanged while evidence shows token caps. Adobe, GitHub and Accenture declined comment, leaving enforcement details unclear.
Verbatim Quotes
- “A lot of people had ideas about how to adjust workflows with lower-reasoning models for certain tasks in order to mitigate token consumption,” — Adobe employee
- “We need everyone to be intentional about model selection to ensure fair access for all users across the enterprise.” — Citi internal email
- “Lots of angsty messages in Slack like ‘now how do I do my job,’” an Atlassian employee told us.” — Atlassian employee
- “Crazy, we go from no more leaderboard to actual usage limits in two weeks,” — Amazon employee
Why It Matters
Per-token pricing forces enterprises to treat AI as a consumable resource, prompting model-selection guidelines, spend dashboards and usage caps. The focus on “token economics” may reshape procurement, vendor negotiations and the pace of AI adoption across sectors.
What’s Next
Citi plans to re-enable disabled models on July 1 while monitoring token use. Atlassian, Adobe and GitHub are piloting tiered access and user-based billing. Accenture is marketing token-management consulting. Audits and scrutiny are expected to tighten corporate AI usage policies.
