Eric Worrall Argues Artificial Intelligence Will Win Green Energy War
Blogger and self-described professional software developer Eric Worrall published a June 17, 2026 essay on Watts Up With That reporting he used Claude to "vibe code" a simple game in about 30 minutes, at an estimated token cost of roughly $2. Worrall says his first attempt got Claude stuck in an endless loop that could have run up a bill of "thousands of dollars" had he not caught it, and he estimates the single session burned enough energy to power "a street of houses." He argues that production multi-agent "vibe coding" sessions, where teams of AI agents work in parallel, could push energy use up to the equivalent of a small town, and predicts AI will eventually "vibe" much of the economy, including physical manufacturing. These figures are Worrall's own estimates, not measured telemetry, and the piece is a single-source opinion essay, not a benchmarked study.
The anecdote is a useful practitioner data point on two under-discussed risks of agentic AI coding: opaque billing exposure and energy cost, both reported here from a single first-person account rather than measured telemetry. Worrall's core caution - that a stuck agent loop can silently escalate from a low single-digit dollar cost to a bill of "thousands of dollars" - is a concrete illustration of a risk every team running autonomous or semi-autonomous coding agents should have guardrails against.
What happened
Eric Worrall, an Australian blogger and self-described professional software developer, published an essay on Watts Up With That titled "Why Artificial Intelligence will Win the Green Energy War," with an accompanying video and playable build. He reports using Claude to generate a simple game's code in about 30 minutes total, a task he estimates would have taken half a day without AI assistance. He estimates the token cost at around $2, covered by his existing AI subscription. He writes that an earlier attempt got stuck in an endless loop that drove up consumption, and that had he not recognized and stopped it, the bill could have reached "thousands of dollars"; he notes Claude has default spending constraints that can be disabled by the user. Worrall estimates that single session burned energy roughly equivalent to powering a street of houses, and argues that real-world multi-agent "vibe coding" sessions, where teams of specialized AI agents work in parallel on larger projects, could push energy use up to the equivalent of a small town.
Technical context
The essay is anecdotal: it does not publish measured power draw, GPU-hours, or a cloud/API billing statement, and the token-cost and energy figures are the author's own estimates. Worrall names Claude as the tool used and mentions Google Gemini, DeepSeek, and xAI as competing systems he expects could do similar work.
Industry context
The piece feeds into a broader practitioner conversation about hidden cost vectors in generative workflows: token consumption, API bill shock, and the operational energy footprint of multi-agent systems. Worrall's closing prediction, that AI-driven "vibe production" could eventually extend from software into physical manufacturing, is offered as his own forecast rather than a reported plan by any company.
What to watch
- •Reproducible, measured GPU-hours and power-draw figures tied to real multi-agent coding workflows, versus author estimates like this one.
- •Provider-side spending caps and quotas (Claude and others) that prevent a stuck agent loop from generating runaway bills, and whether those safeguards can be inadvertently disabled.
- •Tooling that exposes per-agent compute and energy estimates as multi-agent development workflows scale.
Editorial analysis
This is a single-source, first-person opinion essay from a blog known for climate-policy commentary, framed around a provocative "AI will win the green energy war" thesis; the specific dollar and energy figures are the author's own estimates, not independently verified measurements, so they should be read as illustrative rather than benchmarked data.
Key Points
- 1A developer's anecdote shows AI 'vibe coding' can cut a half-day task to 30 minutes for about $2 in token cost.
- 2A stuck agent loop could have generated a bill of thousands of dollars, illustrating a real billing-safety risk in agentic coding.
- 3The author's estimate that multi-agent workflows scale energy use toward a small town is a personal projection, not measured data.
Scoring Rationale
Solid-tier practitioner anecdote: verified against the primary source and factually accurate as reported, but it remains a single-source, first-person opinion essay with author-estimated (not measured) cost and energy figures, published on a blog with a climate-policy opinion bent. Kept at the low end of Solid given genuine practitioner relevance (agentic billing risk) offset by the anecdotal, unverified numbers.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
