Cathie Wood Tells Investors to Track Where AI Agents Spend Money, Not Just Who Builds Them

Cathie Wood wants investors to stop asking which chatbot is smartest and start asking where the money is actually going.

Speaking at Robinhood's Summit in Houston on Wednesday, Sept. 30, the ARK Invest CEO said, "We're probably going to be talking more and more about 'follow the agents,'" according to CoinDesk. She meant AI agents: software that completes tasks and makes purchasing decisions on a user's behalf, rather than just answering questions.

Wood has built ARK's reputation on telling investors to "follow the developers" as an early signal of where technology is headed. Her pivot to "follow the agents" is a bet that autonomous software, not human engineers, will soon be the better leading indicator.

The Numbers Behind the Pitch

ARK's own research gives the thesis some teeth. According to ARK's reports cited by Crypto Briefing, AI agents went from handling tasks averaging about 12 minutes in early 2025 to more than 180 minutes by early 2026. That's a real jump in how long software can operate independently before a human has to step back in.

ARK is projecting AI-driven software spending could hit between $3 trillion and $7 trillion, which the firm's analysts frame as 19% to 56% growth in that category. That's an enormous range, and it should be read as exactly what it is: a projection, not a locked-in number.

ARK is putting money behind the idea. During the week of September 21, ARK bought roughly $24.7 million worth of CoreWeave shares, betting that infrastructure providers will cash in as agent workloads demand more compute.

Wood also pointed to OpenAI's shift toward what she called "headless agents-as-a-service," aimed at charging more per user for inference rather than relying on flat subscriptions. ARK's reporting puts OpenAI's revenue run-rate in the tens of billions, with Anthropic showing similar growth, per Crypto Briefing.

Who Controls the Money Agents Spend

The harder question isn't whether agents will spend money. It's how.

Joseph Chalom, co-CEO of SharpLink and the former head of digital assets at BlackRock, laid out the stakes in the final piece of a three-part series on agentic finance last month. "A world full of intelligent agents means nothing if a handful of companies decide where your money can go," he wrote, as reported by CoinDesk and BigGo Finance.

Chalom's proposed fix: an agent should get a defined spending limit, like $500 to book a hotel, without open-ended access to a bank account. Users should be able to revoke that authority and see a log of every transaction. He also wants agents to be portable between financial providers, carrying their identity and permissions the way a phone number moves between carriers.

This concern cuts against both Big Tech and Wall Street consolidating control over machine-driven commerce. It's also, notably, an industry proposal from a crypto-adjacent executive, not a regulation. No federal agency has published rules on how much authority a consumer can legally delegate to an AI agent, or what happens when an agent overspends or gets hacked. The industry is largely writing its own guardrails right now.

Coinbase has already launched a payment protocol called x402 for machine-to-machine transactions. Coinbase CEO Brian Armstrong said on X that "Grok is the leading client for agentic traders on Coinbase currently," according to a comparison of trade coverage by Newscord. Stripe, Visa, Google and OpenAI are all building competing systems, meaning the fight over whether agent payments run on open blockchains or closed corporate rails is already underway, not theoretical.

The Enterprise Reality Check

Away from the speculative spending forecasts, some companies are already banking real savings from agents. Rhonda Baldwin, CIO at LaunchDarkly, said agentic AI tools helped the company forgo roughly $1 million in spending across two optimization projects, save another $120,000 by building an in-house asset-management system, and cut about $50,000 annually from tier-1 IT support. Kevin Rooney, CIO at West Monroe Partners, said an internal IT and HR support agent drove a 40% reduction in yearly managed service provider costs.

But not everyone is sold on the hype holding up at scale. Jeet Pattanaik, founder and CTO of Glokal AI, offered a blunt warning: "The pilot always looks great because it runs on the easy tickets," he said, noting that production environments bring in odd cases that require someone to actually check what the agent did.

Newscord's comparison of coverage noted that Tech Policy Press emphasized security and trust failures in agentic systems, while CIO and ZDNET framed the same trend around cost savings and speed. Both things can be true: agents are cutting real costs for some firms right now, and they're also operating with far less oversight than the humans they're replacing.

What's Unresolved

Wood's forecast depends on agents continuing to gain autonomy and budget authority at the pace ARK describes. Whether Washington, the banking industry, or Silicon Valley ends up deciding how much money a piece of software can move without a human checking in first is still an open fight, with Chalom's framework representing one side's wish list rather than settled law.