Why Data Products Break the Binary
In my previous column, I wrote about the mental health AI industry heading toward a chasm between enthusiastic early adopters in research labs and the overwhelmed humans who have to use AI that isn’t making them feel better.
The framework I proposed wasn’t “cross the chasm faster.” It was “build transparency as infrastructure from the beginning.”
But I’ve been thinking about a deeper problem: The chasm framework itself was never designed for data products. Geoffrey Moore’s Crossing the Chasm emerged from the technology adoption lifecycle as a model designed for products that remain relatively stable once they are shipped. You build a widget. You find early adopters. You refine the widget.
You cross to the mainstream. The product is the fixed variable; the market is what you fit to. This assumption underpins the entire product-market-fit (PMF) religion that Y Combinator codified and venture capital (VC) scaled. Find fit. Achieve fit. You either have it or you don’t. Binary. Done. But what happens when your product’s primary ingredient is data, and data is not a widget?
DATA IS A FLOW, NOT AN EVENT
Meredith Whittaker, Signal’s president and cofounder of the AI Now Institute, has a phrase she first uttered on June 5, 2024, at Axios AI + NY, and it should haunt every AI investor.
These companies, she said, are “basically promising God and delivering email prompts.”
She expanded on the thought, noting that hundreds of millions of dollars are being spent to train models under relentless pressure to show returns. It reflects the current zeitgeist perfectly: an industry controlled by what Whittaker calls “a handful of surveillance giants” that are “largely unaccountable,” promising transformation while delivering incrementalism at extraordinary cost.
The infrastructure numbers are staggering. Sam Altman sought $5–$7 trillion to reshape the global semiconductor industry, an amount exceeding the gross domestic product of Germany. TSMC executives allegedly found the proposal so disconnected from reality that they started calling him a “podcasting bro.” Wholesale electricity costs have increased by up to 267% near some data center hotspots. Consumers will pay $16.6 billion between 2025 and 2027 to secure future power supplies for facilities that may become obsolete as chip architectures evolve.
Meanwhile, Sasha Luccioni, climate lead at Hugging Face, warned: “Even if the energy is renewable (which it isn’t guaranteed to be), the quantity of water and rare earth minerals required is astronomical.”
This is PMF as religion: Build it massive enough, make the promises extravagant enough, and you’re too big to fail. But the framework assumes the product stabilizes. Data products are being created to feed greedy large language models, yet the data itself remains dynamic, a quickly changing dimension that requires continuous context. Treating data as an event codifies the past as the predictive future.
THE BINARY THAT NEVER WAS
As Emily Bender, the computational linguist who co-authored the influential “Stochastic Parrots” paper with Timnit Gebru et al., put it: “We are working at a scale where the people building the things can’t actually get their arms around the data.”
This is the crux. Traditional PMF assumes you can get your arms around your product. You can define it, test it, ship it. But when your product’s primary ingredient is data that changes continuously and reflects human behavior, market conditions, scientific discovery, and cultural shifts, you cannot fix the variable.
The PMF framework asks whether the market wants this product. Data products require a different question: How are humans interacting with this information, and how is that interaction changing both them and the information itself?
This is why the Crossing the Chasm binary fundamentally misunderstands the role of data products. There is no stable shore to reach. The “product” is a conversation, not an artifact. Consider what I described in my mental health column: millions of people using ChatGPT as their therapist, pouring out trauma at 2 a.m., asking whether they should leave their marriages. The danger arises when “identity coupling” occurs, and someone stops seeing AI as a tool and begins to experience it as a relationship.
This isn’t a PMF problem. You can’t A/B test your way out of it. The interaction itself transforms both the human and the system’s meaning. Data products don’t fit markets; they co-evolve with the humans who use them.
WHY WE NEED CYNEFIN
Dave Snowden’s Cynefin framework distinguishes between complicated problems and complex ones. Difficult problems have corrected answers that experts can discover through analysis. Complex problems differ fundamentally:
Cause and effect can only be deduced in retrospect, and there are no correct answers.
The PMF framework treats markets as complicated. Analyze the segments, identify the beachhead, execute the playbook. But data products operate in complexity. The instructive patterns can only emerge if the leader conducts experiments that are safe to fail.
PROBE. SENSE. RESPOND.
This matters because founders are burning out in pursuit of a binary that doesn’t exist. “The Untold Toll” report, based on a survey conducted by Startup Snapshot in 2025, found that 72% of founders experienced mental health impacts, including anxiety, burnout, and depression. The Global Entrepreneurship Monitor reports that fear of failure among aspiring entrepreneurs has increased from 44% in 2019 to 49% in 2021. Nearly half of the people who see good opportunities won’t act because they’re afraid.
The PMF mythology contributes to this. It promises certainty: Follow the steps, achieve the state, be safe. When the state never arrives because data products don’t stabilize into a “fit,” founders internalize the failure as personal inadequacy rather than a framework mismatch.