Isenberg Hears AI Labs Pay Data Brokers $2M a Week for Data

Greg Isenberg, whose X bio lists him as host of @startupideaspod and CEO of @latecheckoutplz, says that during 48 hours in San Francisco he heard “Sooo MANY stories of data brokers making like $2M a WEEK selling company data to labs.” He made the claim in an X post on 14 October 2026, the first of 21 observations from time spent with “billionaires, data labs and AI founders and execs running $1B to $100B companies.”

He hedges it himself: “I can’t tell if it’s the beginning or a bubble, and I kept wondering which niches are left.” He names no broker, lab or deal, and says he is “Digging into it actively.”

Business Pill · THE MODEL AND THE HARNESS

A one-minute explainer of the model and the harness: the software around a model that decides what it can see and do. It teaches the general idea only and says nothing about any person or organisation in this story.

What Isenberg Says He Heard

The post had more than 160,000 views by the evening of 14 October, as X displayed it when we read it. It is a list of what Isenberg says he saw and heard, not reporting: the data-broker figure is a story he says he heard many times, given with “like” in front of it.

We group his 21 points into five business themes below, quoting his words. Where he gives a guess, a forecast or a feeling, we keep it as his.

What Was Posted on X

Greg Isenberg’s own post of 14 October 2026, the source of this piece: his 21 observations from 48 hours in San Francisco, starting with the data-broker stories he says he heard. Every point quoted above is from this post.

Data Is the Asset

Beyond the brokers, Isenberg says: “My guess is the value moves to data. If the software can be rebuilt quickly, what’s worth paying for in a legacy SaaS company is the years of customer data and workflows inside it.”

He adds that “Labs can get plenty of training data, but it’s harder to get real examples of agents failing inside companies. Whoever has those logs has something valuable.”

And he says “Some startups are worth more dead than alive now. A shutdown company with 5 years of Slack, email and code can get paid for licensing all of it.”

The Model-Swap Test

Isenberg says that at dinner someone asked the table, “If a new model drops next month, does your company get better or worse?” He writes that “HALF the table went quiet!” and calls it “the best question to ask about any AI startup right now.”

He describes the stack this way: “The model is the CPU, the harness is the operating system, and skills and plugins are the apps.”

On harnesses, he says “The best teams I talked to treat the model as swappable.”

He writes: “A vertical harness is basically four things: the connectors for that industry’s data, evals written by people who know the work, rules for what the agent can and can’t touch, and a clean screen for a human to review. The model is maybe 20% of it.”

Revenue Fast, Retention Unknown

Isenberg says “The bar people talk about now is hitting $1B in revenue in 12 months”, and that “What I didn’t hear much about is retention.”

His view: “AI revenue can come in fast and leave fast, and I think we’re going to see a lot of stories of AI companies with huge revenue and terrible retention.” We covered micro1 passing a $1B run rate on 12 October and Modal topping a $1B run rate on 14 October.

Agents Take On Software

Isenberg says “We’re maybe 12 months from an agent that can watch you work for a week and then do your job without any instructions.”

He says “People are decompiling software and rebuilding it with agents”, and that “It feels like a good chunk of legacy software won’t make it to the agentic era, especially anything built around people clicking through a lot of screens.”

He writes that some of the companies he saw “barely have a UI because the main user is an agent”, and asks: “when the most popular user of the internet is an agent, what startups should you be building?” We covered a Tencent Cloud agent fleet probing Alibaba’s Amap on 14 October.

New Risk Markets and the Fight for Attention

Isenberg says “Every agent with access to your systems is a new door into your company, and a lot of those doors were left open”, and that “I think agent insurance is coming. Once an agent makes an expensive mistake, insurers will want audits before they cover anyone.”

On distribution, he writes “Getting SIGNIFICANT distribution is HARD right now” and “Attention is what’s scarce!”

The Structural Read

As we read it, the $2M-a-week broker stories, the agent-failure logs and the dead startups that “can get paid for licensing all of it” are one claim seen three ways: in his account, the scarce input is no longer public training data but private records of how work is actually done.

As we read it, his dinner question, “If a new model drops next month, does your company get better or worse?”, is a test of where a company sits: a business that improves when the model improves is building on the harness; one that gets worse is competing with the model.

As we read it, his warning that “AI revenue can come in fast and leave fast” and his note that he did not hear much about retention are the counterweight to the $1B-in-12-months bar he describes: run rates show adoption, not whether customers stay.

Three Implications

FOUNDERS OF AI APPLICATION COMPANIES Isenberg says the best teams he talked to treat the model as swappable, and that one founder walked him through routing different models to different jobs.

OWNERS AND BUYERS OF LEGACY SAAS COMPANIES His guess is that the value moves to data, and he asks whether 2x ARR is still the right way to price such companies.

SECURITY AND INSURANCE COMPANIES He says every agent with access to your systems is a new door into a company, and that agent insurance is coming.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Agentic Harness War.

The framework puts it this way: “Value migrates model → harness → skill marketplace. Pricing power leaves the model and lands on whoever controls the surface and its reusable workflows.”

As we read it, Isenberg’s account of teams that treat the model as swappable, and of vertical harnesses built from connectors, evals, rules and a review screen, describes that migration from the builders’ side.

What Is Not Established

The piece rests on one person’s account. Isenberg names no data broker, lab, company or deal, and gives no source for the $2M-a-week figure beyond the stories he says he heard.

The forecasts in the post, such as agents doing a job “without any instructions” in “maybe 12 months”, are his guesses, and he labels several points as his own thinking.

The Bottom Line

Greg Isenberg says he heard many stories of data brokers making like $2M a week selling company data to AI labs, guesses that the value moves to data, and says the model is “maybe 20% of it” in a vertical harness.