Andreessen Horowitz said in a post on X on 8 October 2026 that “Agents burn nearly 5x the tokens people do, up 14x in six months”. The chart attached to the post plots the 7-day average token usage on OpenRouter, a service that routes developers’ requests to many AI models, split into agentic, human and mixed traffic.
Box chief executive Aaron Levie quoted the post on 10 October with a forecast of his own. Agents spawning other agents, agents working in the background and agent swarms, he wrote, “will consume 1,000X more tokens than people prompting agents 1 at a time.”
The chart is titled “Agents Are Using Far More Tokens Than People” and subtitled “7-day average token usage on OpenRouter, by type”. It draws three lines, labelled Agentic, Human and Mixed, and its source line reads “openrouter.ai/rankings”.
Two points carry labels. At the last point of the agentic line the chart reads “Agents: 7.3 trillion tokens, 14x increase”. Earlier on, an annotation reads “Feb 6, 2026: Agentic tokens surpass human ones”.
The horizontal axis runs from “Sep 05” to “Aug 07”. The human and mixed lines are drawn but carry no value labels, so the chart itself states only one level: the agentic figure at its final point.
The one level labelled on a16z’s chart: agentic traffic at 7.3 trillion tokens, a 7-day average on OpenRouter, at a final point marked “Aug 07”. The other figures are a16z’s own labels and post text.
Business Pill · COST PER TASK
A one-minute explainer of cost per task: why the price of finishing a job, not the price per token, is the number that matters. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read it, a16z’s chart moves the measure of AI demand from people to processes: on OpenRouter, the line it labels is agentic traffic, at 7.3 trillion tokens on a 7-day average and a 14x increase, and it marks 6 February 2026 as the day agentic tokens passed human ones.
What Levie Added
Levie’s post names three patterns: “Agents spawning other agents, agents operating in the background on every task, and agent swarms”. He wrote that together they “will consume 1,000X more tokens than people prompting agents 1 at a time.”
He also wrote that AI used “as a chat system that can only work for you at the pace that you can prompt it will seem like a relic within a year or two”. He closed by pointing to “how much more compute and infrastructure build out will be needed to sustain this.”
The 1,000X is Levie’s forecast. The chart labels a 14x increase; the “nearly 5x” agents-versus-people figure appears only in the post text. Levie runs Box, and we covered his September conversation with a16z’s Martin Casado on why agent swarms break enterprise access control.
Each number compares different things: the first two are a16z’s figures on OpenRouter traffic, the third is Aaron Levie’s forecast.
Whose Data This Is
The chart cites openrouter.ai/rankings as its source, and a16z has backed OpenRouter. In a newsletter post dated 19 August 2026, a16z wrote: “Stripe has signed a deal to acquire OpenRouter.” The same post describes a16z’s seed and Series A investments: “We jumped at the chance to do the seed deal, and when it came time for the A, we took a big swing.”
The chart’s three lines split OpenRouter usage into agentic, human and mixed traffic, and only the agentic line carries a value label. The post’s “nearly 5x” compares agents with people; the chart’s “14x increase” describes agent tokens alone.
Levie’s 1,000X is a different comparison again: agents spawning agents, background agents and swarms against “people prompting agents 1 at a time”. It is his forecast, not a figure from the chart.
As we read it, the source matters as much as the multiple: the chart cites OpenRouter’s rankings, a16z says it invested in OpenRouter at seed and Series A, and its own 19 August post calls tokens “a new, universal medium of value exchange”.
a16z on X, 8 October 2026
“Agents burn nearly 5x the tokens people do, up 14x in six months”
Three Implications
TEAMS BUDGETING FOR AI a16z’s post compares agents with people on tokens burned, and its chart labels agentic traffic on OpenRouter at 7.3 trillion tokens on a 7-day average.
READERS OF USAGE CHARTS The chart does not state how a request is classed as agentic, human or mixed, and its axis ends at “Aug 07”, two months before the post.
INFRASTRUCTURE PLANNERS Levie wrote that agent swarms and background agents point to “how much more compute and infrastructure build out will be needed”; the 1,000X behind that is his forecast.
The analysis puts it this way: the token “is simultaneously the unit of cognition the model produces, the unit of compute the data center serves, the unit of price the lab charges, and the unit of value the enterprise extracts.”
As we read it, a16z’s chart measures the second of those roles from one router’s vantage point: compute served, split by whether an agent or a person sent the request.
What Is Not Established
Neither the post nor the chart says how a request is classed as agentic, human or mixed. OpenRouter’s rankings page, as we read it on 11 October 2026, lists token totals by model and by app but did not show that classification method in its text.
The chart’s axis ends at “Aug 07”, two months before the 8 October post, and the post does not say why the later weeks are not shown or which dates the “nearly 5x” and “six months” refer to. The chart covers traffic that passes through OpenRouter, which is one routing service among the ways developers reach models.
Business Engineer Framework
Tokenomics: The Economics of AI
A Business Engineer structural read of the token as the atomic unit of the AI economy, cascaded through the stack.
Andreessen Horowitz’s post says agents burn nearly 5x the tokens people do, up 14x in six months. Its chart, citing OpenRouter’s rankings, labels agentic traffic at 7.3 trillion tokens and a 14x increase at a final point marked “Aug 07”. a16z says it backed OpenRouter at seed and Series A, and Stripe has signed a deal to acquire OpenRouter. Levie’s 1,000X is a forecast, not a measurement.
A note on sourcing. We read a16z’s 8 October 2026 post on X and its attached chart, Aaron Levie’s 10 October post, and a16z’s 19 August 2026 newsletter post on OpenRouter and Stripe in full, and OpenRouter’s rankings page on 11 October 2026. The chart figures are a16z’s labels, based on OpenRouter data; we have not checked them against OpenRouter’s raw data. a16z says it invested in OpenRouter at seed and Series A. Levie’s 1,000X is his forecast. Nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur, HackerNoon and Search Engine People. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
