AI adoption is a business decision disguised as a technical one.
Organizations are investing heavily in AI adoption without a reliable way to know if it is working. The evidence is not scattered -- it is remarkably consistent across independent sources.
This is not primarily a technology problem. It is a governance problem -- one with a clear historical precedent. The pattern echoes the Internet boom of the late 1990s, not because AI and the internet are the same kind of technology, but because both created the same organizational condition: enormous pressure to commit resources before anyone had agreed on how the decision to commit should be governed.
The common retelling of the dot-com crash focuses on measurement: companies reported vanity metrics like traffic and user counts instead of profitability, and the truth caught up with them eventually. That is accurate as far as it goes, but it treats the symptom as the disease.
The deeper failure was structural. Most dot-com-era companies had no governance mechanism that required anyone to ask, before capital was committed, whether a given initiative had a viable path to a defensible business outcome. Measurement failures were downstream of that absence -- if there is no gate that requires evidence before advancement, there is no organizational reason to produce honest evidence in the first place. Companies that survived the crash (Amazon, eBay) were not simply better at reporting numbers. They had decision discipline earlier in the process, which honest measurement then reinforced.
The same secondary effect shows up in AI adoption today. When there is no governance checkpoint before an AI initiative begins consuming resources, there is little organizational incentive to surface uncomfortable findings once it is underway -- admitting slow progress or unclear ROI can feel like it reflects poorly on whoever sponsored the initiative, regardless of whether the initiative itself was ever set up to succeed. That is a real and consistent pattern worth naming. But it is a symptom of the missing gate, not the root cause.
| Internet Boom (1995-2001) | AI Adoption Now (2024-2026) |
|---|---|
| "You must go digital or die" | "You must adopt AI or fall behind" |
| No governance gate before resource commitment | No governance gate before resource commitment |
| Speed prioritized over structured decision-making | Speed prioritized over structured decision-making |
| Vanity metrics substituted for outcome evidence | Adoption percentages substituted for outcome evidence |
| Reality arrived at the crash, not before it | Reality is arriving now, while the window to act is still open |
Across conversations with AI leaders managing active initiatives, three patterns show up consistently enough to be worth stating plainly.
Most AI governance conversations happen after an initiative is already underway -- as a compliance review, a security check, or a retrospective. Very few organizations have a gate that operates before commitment, when the cost of saying no is lowest.
This is the mechanism, not the root cause: without a governance structure that makes "no-go" and "pivot" normal, expected outcomes of the process, every honest status update becomes a career conversation instead of a decision point. Fix the structure, and the pressure not to report honestly loses its organizational logic.
The dot-com era's harshest lesson was timing: measurement discipline built after the crash was too late to matter for the companies that needed it most. The AI adoption cycle has not reached that point yet. The organizations building governance discipline now are building it while it is still a competitive advantage rather than a recovery effort.
I currently lead AI transformation work inside a mid-size organization, across two active fronts: a broad enterprise AI adoption effort, and a parallel AI governance and risk effort covering dozens of active AI initiatives.
The pattern described in this paper is not theoretical to me. Leadership regularly asks whether AI investment is paying off, and the honest answer, most of the time, is that no one has a structured way to know -- not because the organization lacks tools, but because no checkpoint exists earlier in the process to make that question answerable. Separately, a governance review of AI initiatives across the organization surfaced a meaningful amount of AI activity running without any formal review at all -- not from bad intent, but because no gate existed to catch it.
That lived experience is the reason this research exists.
The problem is structural, not technical. Governance built before commitment, not measurement applied after the fact, is what determined which organizations survived the last major technology adoption cycle with their credibility intact. The same choice is available now, earlier in the cycle than it was available then.
GreenfieldworkAI is a research effort examining exactly this question -- how organizations should govern the "should we build this?" decision before resources are committed. Over time, this research is intended to inform both advisory work and a platform built around these principles. For now, the goal is simply getting the framework right.
If any of these patterns match what you are seeing in your own organization, I would welcome the conversation.