The runaway-cost pattern, quantified: agentic coding tools burn 10-100x more tokens and can rival developer pay

The specific runaway-cost incidents — a $47,000 agent loop, a $6,531 AWS retry loop, Uber exhausting its annual budget in four months — are instances of a structural pattern now quantified across the category. Agentic coding tools break each task into long sequences of model calls and resend the growing conversation context on every call, so they consume 10 to 100 times more tokens than a single chat prompt. Teams in 2025–2026 report tripling their bills quarter over quarter with no additional growth in team size, codebase complexity, or delivery velocity, and Gartner has projected AI coding costs could eventually match what developers are paid. Unlike a deletion or a breach, this failure mode is quiet and cumulative — which is exactly why it so often goes untracked until the invoice forces the issue.

Multiple-agents was responsible for this other incident, documented as STUPID-2026-0060 in the StupidLLM AI agent incident database.

It is rated 2.2/10 (low) on StupidLLM's CVSS-style severity scale for AI agent failures, based on damage type, reversibility, and scope.

The root cause was classified as confidence miscalibration. Costs remain a small, predictable fraction of engineering spend.

A structural cost failure mode affecting the whole category: spend scales with token-hungry agent loops rather than with output, so budgets balloon unpredictably across teams and vendors.