A customer told me recently that Microsoft Copilot is "the JCPenney of Agentic AI."
It landed as a joke, but the more I thought about it the more I found the comparison legitimate. 

JCPenney did not fail because it disappeared. It failed while it was everywhere - harboring more American malls than almost any other retailer, structurally impossible to avoid, and still unable to convert that presence into a business people wanted to use.
That is the exact position Copilot occupies inside the enterprise today: bundled into 440 million Microsoft 365 seats, with a reported ~20 million weekly active users against ChatGPT's 800 million-plus. It is a case study in a mistake enterprise software makes constantly: mistaking distribution for demand.

The Setup: Two Companies, One Structural Error

JCPenney's downfall is usually told as a pricing story: CEO Ron Johnson eliminated the coupons and clearance events its core shoppers loved, replaced them with a generic everyday low price model built for a customer JCPenney didn't have, and watched revenue drop 25% in a single year. He was removed after seventeen months.

But the pricing decision was a symptom. The underlying error was structural.

1. Anchor Tenant Economics

JCPenney's placement was, for decades, a function of real estate strategy, not retail performance. Mall developers needed a recognizable anchor to draw baseline foot traffic and justify leasing the smaller storefronts around it. Being the anchor guaranteed presence. It never guaranteed that shoppers walked in wanting to buy something.

Copilot's position inside Microsoft 365 is the same mechanism. It ships pre-installed across the seats organizations were paying for: Word, Excel, Teams, Windows. That is a licensing and bundling strategy, and it is a genuinely effective one for driving install-base coverage. It is not, on its own, evidence of a workflow anyone chose to adopt.

When a vendor cites seat count or install base as a proof point, ask the follow-up question JCPenney's board should have asked in 2011 - of the people who have access to this, how many are actually using it, and why?

2. Reinvention Without a Customer

Ron Johnson's mistake was not that he redesigned JCPenney. It was that he redesigned it around a customer psychology he assumed rather than the one he had. His core shoppers were value-driven: they wanted the ritual of the deal, the coupon, the clearance rack. He removed all of it in favor of a cleaner, Apple-inspired experience built for a different kind of buyer entirely, and never validated the assumption before shipping it enterprise-wide.

Copilot's agentic layer follows a comparable pattern: a conversational sidebar bolted onto existing productivity surfaces, branded as agentic, without being re-architected around the actual unit of work regulated industries run on: multi-step, auditable, cross-system workflows with clear ownership and compliance requirements. A chat interface answering questions inside Word is a meaningfully different product than an agent that can execute a prior-authorization workflow end to end inside a HIPAA-governed environment.
Labeling the former "agentic" doesn't make it the latter.

3. Diversification Without Focus

Michael Porter's stuck in the middle framework describes exactly what happened next at JCPenney: the company tried to run as both a discount retailer and a high-style specialty brand simultaneously, achieved neither position credibly, and lost ground to competitors who had chosen one lane and gone deep in it: Macy's on style, Kohl's on value.

Microsoft's current agentic product surface, Copilot, Copilot Studio, Copilot Cowork, Agent 365, is the same instinct applied to software architecture: a horizontal spread across every department and use case rather than genuine depth in any one of them. Breadth is not a flaw by itself. It becomes one when the buyer's need is narrow, technical, and high-stakes, which describes almost every regulated healthcare or financial services workflow StackAI is brought in to solve.

4. Kept Alive for the Real Estate, Not the Retail

After its 2020 bankruptcy, JCPenney was purchased not by a retail turnaround firm but by Simon Property Group and Brookfield Asset Management, two of the largest mall landlords in the country. The purchase decision was explicit: keeping the JCPenney anchor open protected the value of the malls built around it. The business was preserved as infrastructure, not because it was winning as a retailer.

This is the most uncomfortable parallel, and the most instructive one. Copilot's continued presence inside the enterprise increasingly reads the same way: sustained because it protects Microsoft's position in the seat, not because it is winning the underlying workflow. That is a perfectly rational thing for Microsoft to do. It is not, however, a reason for a buyer evaluating agentic AI.

Conclusion

Bundled is not the same as built-for-purpose. Being in every seat is not the same as getting the work done. JCPenney's anchor stores stayed open long after the business stopped being one people chose. The question worth asking about any horizontal AI assistant riding on the back of an existing productivity suite is the same one that should have been asked about JCPenney a decade earlier: is this here because people want it, or because someone needs it to stay open?

Want to learn more? Geta comparison between StackAI and Copilot studio. Talk to our team of AI experts by booking a demo.