Claude Code blows up Uber AI budget

This is the enterprise AI coding bill finally becoming visible: the productivity story is real, but token-metered agents are much harder to budget than seat-based devtools.

• –Claude Code has moved beyond autocomplete into agentic engineering work that edits files, runs tests, handles git workflows, and can operate across full codebases.
• –Uber’s adoption numbers make AI coding look less like a pilot and more like a new engineering substrate, with human work shifting toward architecture, review, and orchestration.
• –The cost problem is structural: agentic coding burns tokens through planning, context loading, tool calls, retries, and test-fix loops, so usage can scale faster than headcount.
• –Enterprise buyers will start demanding cost controls, usage dashboards, model routing, and clearer ROI metrics, not just better demos.
• –For developers, this is a signal that AI coding skills now include budget awareness: good prompting, smaller contexts, and scoped tasks are becoming engineering hygiene.