DevLoops

The wall isn't technical anymore — it's financial. Most enterprise AI budgets were drawn up in late 2025, before agentic coding tools blew past every assumption behind them. Six months later, the bills are landing.

Uber exhausted its entire 2026 AI budget in four months. After putting Claude Code in front of roughly 5,000 engineers, per-developer token spend ran $500–$2,000 a month, and the company now caps usage at $1,500 per tool. Asked whether the spend was producing better products, its COO admitted it's “very hard to draw a line” to more useful features.

Microsoft cancelled most of its internal Claude Code licenses just six months after handing them to thousands of engineers — cost a clear factor, with the cutoff set to its June 30 fiscal year-end. When the industry's most aggressive AI adopter starts rationing the tool, the unit economics have a problem.

It's a pattern, not an outlier. The Wall Street Journal reports enterprises burning through annual AI budgets in as little as three months, with some bills doubling or tripling; Walmart has capped employee use of its own in-house agent too. Finance teams that spent a decade governing cloud spend have no playbook for token bills that scale nonlinearly.

The cause is the exact thing DevLoops removes: agents chain dozens of model calls and re-read the same codebase on every prompt, so you pay to ingest the same data thousands of times a day. Index it once, inject only the relevant nodes, and the curve bends back down — up to 85% fewer tokens, on-premise.