The budget blew up first

Uber CTO Praveen Neppalli Naga said the pace at which AI coding tools have spread was so fast that the initial AI budget has already collapsed.

The key point is the rapid spread of agentic software engineering. Inside Uber, AI has gone beyond simple assistance to take on writing code itself, with about 1,800 code changes generated weekly by AI agents without human intervention.

The numbers are also large.

• Share of engineers using AI tools monthly: 95%
• Share of committed code generated by AI: about 70%
• Contribution of internal AI agents to code changes: from under 1% to 8% within a few months

Anthropic's Claude enterprise pricing changes were pointed to as a background factor in the rising spend. As the structure shifted from a fixed subscription to one combining seat fees with usage commitments, companies now have to more precisely forecast token consumption rather than just the number of users.

The article also reports that this trend is spreading into a tokenmaxxing culture among companies. Some organizations track engineers' token usage on internal dashboards, pitting them against each other like a productivity metric.

In conclusion, AI coding has boosted productivity, but at the same time it is depleting corporate IT budgets much faster.