The Pig Knuckle Papers
The Shadow AI MythThe problem was never the sneaking.
If you've read anything about AI at work this past year, you know the Shadow AI story: employees secretly using ChatGPT, corporate IT horrified, sensitive data leaking into the ether, executives scrambling to regain control before everything catches fire. It's an interesting story. It just isn't the whole story.
The Scroller tl;dr
The evidence strongly supports one idea: employees are adopting AI faster than organizations are governing it. What it does not support are many of the overconfident claims in secondary reporting. The real competitive advantage isn't stopping Shadow AI — it's turning employee experimentation into secure, governed workflows that actually redesign how work gets done.
What the data says vs. what everyone says the data says
After digging through the research behind these headlines, we found the underlying evidence is often more nuanced than the reporting suggests. The broad conclusion is absolutely correct — employees are adopting AI far faster than organizations are governing it. But many of the statistics repeated across articles are being blended together from different studies, creating a narrative that sounds more definitive than the evidence actually supports.
That's worth understanding, because good strategy starts with understanding what the data actually says — not what everyone else says the data says.
Start with the study everyone cites
The cornerstone of this discussion is Microsoft's 2025 Work Trend Index, The Frontier Firm Is Born. It's a substantial piece of research: roughly 31,000 workers across 31 countries, combined with anonymized Microsoft 365 telemetry. As enterprise workplace studies go, it's one of the better datasets we have.
It paints a picture of organizations under strain. Sixty-eight percent of employees say they struggle to keep up with the pace and volume of work, while 82 percent of leaders believe this is the year they need to fundamentally rethink strategy and operations. Those two findings reinforce each other: employees feel overwhelmed, and executives recognize that incremental improvements are no longer enough.
That environment creates a powerful incentive for experimentation. When people are overwhelmed, they don't wait for permission to become more productive. They look for tools that help them survive the workday.
Where the reporting gets muddled
Real numbers, different studies, one story
Many articles present the 2025 findings alongside statistics claiming that roughly 78 percent of AI users already bring their own tools to work while only about 16 percent rely on employer-provided ones. Those numbers are real — but they don't come from The Frontier Firm Is Born. The 78 percent figure traces back to Microsoft and LinkedIn's earlier 2024 Work Trend Index; the 16 percent comes from a separate 2025 adoption survey entirely. Different studies, different questions, different years.
The 2025 Work Trend Index
~31,000 workers · 31 countries · + M365 telemetry
About transformation & capacity
Separate adoption surveys
different studies · different years
About who's using what
Real numbers. Different studies. Blended into one story.
The 2025 report is primarily about workplace transformation, capacity strain, and organizational readiness. The adoption surveys are about who is actually using what, and whether anyone sanctioned it. They complement one another remarkably well — but they answer different questions, and were never meant to be read as one comprehensive study that proved everything at once.
That doesn't weaken the overall conclusion. If anything, it strengthens it: independent datasets are pointing the same direction. Whether you approach the issue from productivity research or adoption telemetry, you arrive at essentially the same destination — employees are adopting AI faster than organizations are building the policy, training, and governance around it.
Describing behavior is not predicting consequences
Where we get more skeptical is when the conversation shifts from describing behavior to predicting outcomes. Most of these studies tell us that employees are using AI, why they believe they're using it, and how often organizations report seeing the behavior. What they generally do not tell us is how many significant security incidents resulted directly from Shadow AI, how much productivity actually increased, or how much value organizations ultimately created. Those are much harder questions, and the evidence is still developing.
That distinction matters, because it separates trend identification from causal claims. We have high confidence that employee experimentation with AI is accelerating. We have much lower confidence in anyone who claims to know precisely how much money organizations are making or losing because of it. And it's worth remembering that Microsoft has an obvious interest in selling governed, enterprise-grade AI, just as the adoption-survey vendors have their own — neither fact invalidates the research; it just earns the same healthy skepticism we'd apply to any sponsored study.
We can say, with confidence, that experimentation is accelerating. We cannot say, with anything like the same confidence, what it is worth.
Why "Shadow AI" is the wrong frame
The encouraging news is that the broad trend survives the scrutiny. Employees aren't waiting:
They're not waiting for executive committees.
They're not waiting for procurement.
They're not waiting for the annual technology roadmap.
They're solving today's problems with today's tools, because the work still needs to get done.
That's why "Shadow AI" is slightly misleading. It implies the central issue is employees sneaking around corporate policy. The more interesting story is that organizations were caught flat-footed by the speed of adoption. History says this shouldn't surprise us — cloud storage, messaging platforms, personal smartphones, and collaboration software all followed the same path. Employees adopted them before companies formalized them. AI is simply moving faster than any wave before it.
Experimentation is a signal, not a transformation
The organizations that succeed won't be the ones that eliminate employee experimentation — that battle is already over. They'll be the ones that treat experimentation as a signal rather than a threat: finding where employees are already getting value, redesigning workflows around those discoveries, and wrapping the result in governance that makes it secure, auditable, and scalable. That's a fundamentally different objective than trying to stop people from using AI — and it's where many organizations are struggling today.
Consulting firms increasingly point to an adoption-to-earnings gap: plenty of organizations report widespread AI usage, yet far fewer can demonstrate meaningful changes to operating performance. That isn't an indictment of the technology. More often, it's evidence that organizations automated individual tasks without redesigning the underlying processes.
There's a line making the rounds among practitioners, and it fits: "AI doesn't fix broken operations. It exposes them."
It's become a remarkably effective diagnostic — revealing inefficient workflows, redundant approvals, unclear ownership, and outdated processes with uncomfortable speed. Organizations often read those discoveries as technology problems when they're actually management problems that existed long before the first prompt was ever typed.
The risk isn't the experiment. It's mistaking it for the finish.
The biggest risk was never that employees are experimenting with AI. It's assuming that experimentation alone constitutes transformation. Giving people better tools without rethinking how work gets done is like dropping a new engine into a car with square wheels: plenty of noise and excitement, and you still don't go anywhere.
The companies that create lasting value won't be the ones with the most AI licenses. They'll be the ones that convert thousands of isolated acts of employee experimentation into governed, repeatable, enterprise-wide systems that genuinely improve the business. That's a much harder challenge than blocking ChatGPT. It's also the one that actually matters.
About the Pig Knuckle Papers: The Pig Knuckle Papers are a human-led, AI-assisted research series published by Alchemy Agentic. Each paper begins with a human question, uses Pig Knuckle, Alchemy Agentic’s flagship LLM-orchestration product, for deep research, and undergoes human review before publication. Keith Norton, co-founder of Alchemy Agentic, serves as editor and narrator. Paul Langtry is co-founder of Alchemy Agentic. ChatGPT may be used to shape approved research into the intended human voice, but humans review and approve every final piece.
Sources
- [1]The 2025 Annual Work Trend Index: The Frontier Firm is born — Microsoft
- [2]Microsoft Work Trend Index 2025 Shows Workplace Capacity Strain — Forbes / Moor Insights
- [3]AI at Work Is Here. Now Comes the Hard Part (2024 Work Trend Index) — Microsoft & LinkedIn
- [4]Tech Trends 2026: The agentic reality check — Deloitte Insights
- [5]The state of AI: How organizations are rewiring to capture value — McKinsey (QuantumBlack)