TLDRocket
Sign in

đź”® The curious economics of a $6 AI agent #597

Exponential View Azeem Azhar

An AI agent that once cost $500 a day now costs $6. Cheaper models and paid subscriptions are slashing costs, but the bill is still easy to lose track of.

Based on reporting by Exponential View, Azeem Azhar — read the original for the full story.

Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error

A weird little AI budget story is doing the rounds out of London: an AI agent that had been burning through about $500 a day has been pushed down to roughly $6 a day. That’s not because it got dumb. It’s because the model stack underneath it got much cheaper, and a lot more carefully managed.

The author says an earlier audit found the agent was still leaning on older, pricier models such as Opus 4.5 when smaller options like Sonnet 5 were available. Some of the savings now come from a $200-a-month OpenAI subscription that covers Codex, plus fallback routing to DeepSeek v4 Flash or Pro if OpenAI isn’t available. Harder jobs can still jump to 5.6 Sol, OpenAI’s top model, or to Kimi K3 and Anthropic’s Fable, which are paid for per token.

The result is a monthly bill that looks almost comically low next to the earlier outlier, but it’s not trivial. At $6 a day, the agent still comes to about $2,000 a year for one person, and that’s before the workload picks up again. The author expects spending to rise when the work shifts back toward book research.

That personal example is a neat little mirror of the bigger AI market. The source points to continued corporate spending in the US, with Ramp saying the top 1% of businesses spent a median $7,400 per employee on AI in July, while the median firm spent $11.95. And the pricing war between Anthropic and OpenAI, sharpened by Chinese advances, is making these agents cheaper even as they become more capable.

The irony is obvious: the same systems that make budgets harder to forecast are also the ones making the bills fall. That’s the AI business model in one sentence — impressive, useful, and a bit slippery when nobody can say what anything actually costs.

My take — AI-written commentary, not fact-checked reporting

This is the AI industry in miniature: glorious capability, lousy accounting. The real competitive moat may be who can make the bill look boring, not who can make the demo look magical. Anyone selling AI as a clean, predictable utility is already decorating the lie with nicer fonts.

Read more about this at: Exponential View

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.