Big tech and enterprise adopters are discovering that token-based pricing for agentic coding tools can outpace traditional budgets. Forbes reports that Uber exhausted its full 2026 artificial intelligence budget by April after rapid roll-out of Anthropic's Claude Code to thousands of engineers, and Forbes also reports monthly per-engineer costs ranging from $150 to $250 on average and $500 to $2,000 for heavy users. Reporting by Fortune and TheStreet says Microsoft began canceling most internal Claude Code licenses in mid-May, ending access in its Experiences and Devices division by June 30. Industry coverage links the strain to the mismatch between predictable seat-based licensing and variable token consumption. Editorial analysis: Companies deploying agentic, token-priced models are encountering finance and governance gaps that make costs volatile and harder to forecast.
What happened
Uber exhausted its full 2026 AI budget by April, reporting rapid internal adoption of Anthropic's Claude Code across its engineering organization, according to Forbes and reporting summarized by PYMNTS. Forbes reported that adoption rose from roughly 32% of engineers in February to 84% classified as agentic coding users by March, and that about 11% of live backend updates were written by agents with no human in the loop. Forbes also reported monthly cost ranges per engineer: $150-$250 on average and $500-$2,000 for power users. The company spent $3.4 billion on research and development in 2025, up 9% year over year, per Forbes and Quartz reporting.
Microsoft canceled or is winding down most internal Claude Code licenses in its Experiences and Devices division, with internal access reportedly ending by June 30, according to Fortune and reporting echoed by TheStreet and Yahoo Finance. Coverage notes the pullback sits alongside Microsoft's broader commitments to Anthropic under a Foundry deal and does not affect that investment, per Fortune.
Editorial analysis - technical context
Token-based consumption billing charges enterprises for text processed and generated rather than fixed-seat licenses. Industry reporting frames the core friction as a structural mismatch between variable token consumption and traditional finance models that expect predictable license line items. Companies that rolled out Claude Code or similar agentic tools at scale saw rapid, non-linear increases in invoice amounts driven by: increased parallel agent execution, large-scale refactors, automated test generation, and high-frequency prompting. Observed vendor-side shifts, such as Anthropic's move to meter agent features separately, amplify the unpredictability of monthly bills, according to Forbes coverage.
Industry context
Editorial analysis: Observers across reporting outlets place these events in a broader pattern where enterprises that adopted powerful generative tools in late 2025 face new FinOps requirements. Reporting (including Yahoo Finance and Quartz) documents companies layering governance controls, quotas, cheaper model routing, internal metering, and usage-based leaderboards, to restrain consumption. Industry surveys cited in coverage (via Yahoo Finance) suggest many firms miss AI cost forecasts and that FinOps functions for AI are expanding rapidly.
What to watch
Editorial analysis: Practitioners and procurement teams will likely track a few leading indicators rather than company intentions. These include:
- •adoption-to-cost ratios per team (tokens consumed per feature shipped), as reported by engineering leads in public disclosures;
- •vendor billing changes that separate agentic or application-layer usage from base model access, such as the Anthropic meter described in Forbes; - •internal governance measures (quotas, routing to cheaper models) and their operational impact on developer workflows, as noted across reporting outlets.
Editorial analysis: For engineering managers and FinOps teams, the immediate operational implication is that consumption metrics must be instrumented and reconciled with product delivery metrics. Reported cases show high adoption and measurable code output gains, yet finance teams are encountering invoice volatility that seat-based budgeting did not produce.
Bottom line
Reporting across Forbes, Fortune, Quartz, PYMNTS, and TheStreet documents a growing pain point in enterprise AI procurement: token-priced, agentic tools can deliver substantial engineering productivity while simultaneously producing uneven and harder-to-predict cost profiles. That divergence is prompting operational and governance responses within large engineering organizations, per the cited coverage.
Scoring Rationale #
Multiple major technology employers reporting exhausted budgets and canceled internal licenses makes token-pricing a practical operational concern for practitioners. The story affects budgeting, procurement, and FinOps teams broadly but does not introduce a new model or paradigm shift.
Practice with real Ride-Hailing data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card