Cost Analysis

Why AI Coding Burns Money

$100–$1,100 token burns in a single session are documented, not hypothetical. Here's exactly where the money goes — and how to stop it.

The Five Cost Leak Sources

Retry Inflation

35% of total waste
$35-$400/incident

Failed attempts compound token consumption. Each retry adds more context, making subsequent retries more expensive.

Context Waste

25% of total waste
$20-$150/session

Verbose error messages, stale conversation, and failed file reads consuming tokens without producing value.

Scope Creep

20% of total waste
$15-$200/incident

Agent modifies files outside the requested scope, then spends tokens fixing the unintended changes.

Unattended Execution

15% of total waste
$50-$1,100/incident

Agent runs overnight or during meetings with no human oversight, burning tokens on circular logic.

Orchestration Loops

5% of total waste
$100-$890/incident

Multi-agent workflows where agents agree with each other without doing work, consuming compute at scale.

Real Documented Incidents

$1,147
Overnight token burn

Agent in retry loop for 6h 36m. Zero usable output.

$890
Agreement loop

340 turns of agents agreeing. Zero code produced.

$340
CSS animation loop

67 retry attempts on a simple task.

How Governance Contains Costs

  • Per-task budget caps — execution halts at $25 by default
  • Per-session budget caps — hard ceiling at $50 per session
  • Retry limits — maximum 3 retries before human escalation
  • Unattended timeout — automatic halt after 30 minutes without interaction
  • Agreement loop detection — halts multi-agent workflows with no tool invocations
  • Scope enforcement — blocks file modifications outside the approved scope

Frequently Asked Questions

I got a surprise bill from my AI coding tool. Why?
Surprise bills happen because AI coding agents have no default financial circuit breakers. Without governance, a doom loop or retry storm can consume $100-$1,100 in tokens overnight with no human notification. Cost governance adds per-task ($25) and per-session ($50) caps that halt execution before your bill explodes.
How much does a typical AI coding session cost?
A governed session costs $8-$40. An ungoverned session with retry inflation can cost $100-$1,100. The difference is whether financial circuit breakers are in place.
Can I just set a budget in the API?
API-level budget limits stop ALL execution, including productive work. Governance-level limits are task-aware — they halt only when the cost-per-task ratio indicates waste, not productive computation.
Is vibe coding creating a maintenance nightmare?
Yes. AI-generated codebases accumulate technical debt 3-5x faster than human-written code because agents optimize for local fixes, not architectural coherence. Without governance, vibe coding produces black box codebases that become abandonware — too fragile to modify, too expensive to maintain.
What about Claude Max/Pro subscriptions?
Subscription plans cap your bill but not your waste. You still burn through rate limits on retries and context waste. Governance prevents the waste itself, making every token productive.