I saw the recent news about Microsoft’s plans for Copilot, and my first reaction was relief.
Not because Copilot has failed. And not because Microsoft is retreating from AI.
The relief came from seeing Microsoft articulate something the market has been signaling for some time: AI does not earn adoption simply because it is available everywhere. It has to earn a meaningful place in the work people are already trying to do.
According to recent reporting, Microsoft Copilot executive Jacob Andreou told the organization he leads that the product must “earn the right to exist.” That is unusually direct language from a company that has made Copilot the center of nearly every Microsoft 365 conversation for the past several years.
It is also the right standard.
Distribution Was Never Going to Be Enough
Copilot has had an advantage most software products could only dream of. Microsoft can place it directly inside the applications millions of people already use every day.
Word. Excel. Outlook. Teams. Dynamics 365. Windows.
There may be no better distribution platform in enterprise software.
But availability is not adoption.
A button inside an application does not automatically give someone a reason to use it. A technically impressive response does not necessarily help complete the work. And an AI assistant that can do many things in theory may still struggle to become essential to anyone in practice.
That is not exclusively a Microsoft problem. It is one of the defining challenges facing enterprise AI.
The first wave of generative AI was largely driven by possibility:
What can the model generate?
What information can it retrieve?
How naturally can it respond?
How many products can it be added to?
The next wave will be driven by usefulness:
What job is the user trying to complete?
What business context does the AI need?
What action should it take?
What permissions and controls apply?
Can the result be trusted, reviewed, and repeated?
That is a much higher standard than adding a chat experience to an existing application.
Enterprise AI Is Missing Business Context
A general-purpose AI assistant can help write an email, summarize a document, or brainstorm an idea. Those are valuable capabilities, and I use them regularly.
But business applications operate in a different environment.
The work is connected to customers, projects, policies, transactions, relationships, deadlines, exceptions, and approvals. The data often spans multiple records and systems. The correct action may depend on the user’s role, the organization’s rules, and what has already happened in the process.
For enterprise AI to move from answering questions to completing meaningful work, it needs to understand that environment.
What Enterprise AI Actually Needs
- Structured and reliable business data
- Relationships among records and activities
- Clearly defined processes and outcomes
- Role-based permissions and governance
- The ability to take approved actions
- A workspace where people can review and control the result
Without that context, AI can sound intelligent without being operationally useful.
This is why the future of enterprise AI is not simply a better chatbot. It is AI working inside a well-designed business application, with the context and controls required to help someone accomplish a real job.
Why We Removed a Copilot Integration From Our Own Product
This became more than a theoretical discussion for us at ForgeXRM.
We were recently preparing a new product for submission to Microsoft’s commercial marketplacev. One planned capability involved an integration with Microsoft Copilot.
The concept made sense. It was aligned with Microsoft’s direction, looked compelling in a demonstration, and gave us an obvious AI story to tell.
But as we got closer to release, the practical questions became harder to ignore.
To implement the integration responsibly, we would need access and cooperation within each customer’s Azure and Microsoft environment. Customer IT administrators would have to become involved. Security, permissions, governance, and configuration could vary significantly from one organization to another.
If something did not work, responsibility would be divided among our product, the customer’s environment, Microsoft’s services, and the people configuring the connection.
We could demonstrate it.
What we could not yet do was deliver it as a predictable, repeatable, and supportable product experience.
That distinction matters.
Our model is not to sell an exciting concept and leave customers to assemble the pieces. We want to provide software we can clearly explain, consistently deploy, responsibly support, and confidently stand behind.
So we removed the integration and submitted the product without it.
Microsoft’s subsequent announcement did not prove that our decision was right or that Copilot was wrong. It reinforced the product discipline behind the decision.
We should not ship an AI capability simply because it is possible or because the market expects to see one. We should ship it when it solves a defined problem and when we can support the complete customer experience surrounding it.
Product Discipline Sometimes Means Pulling Back
Earlier in my career, we recognized both breakthrough innovation and the discipline to stop pursuing an idea that was not ready.
That second form of innovation deserves more credit than it typically receives.
It is difficult to step back from something after time and energy have already been invested. It becomes even harder when the capability aligns with the market’s most prominent trend.
But product discipline is not only knowing what to build. It is also knowing what not to release—or what is not yet mature enough to bring into the market responsibly.
Continuing simply because a commitment has already been made is not necessarily conviction. Sometimes it is just momentum.
The better question is whether the capability creates enough customer value to justify the complexity it introduces.
Microsoft’s Reset Is an Encouraging Signal
What interests me most about Microsoft’s evolving Copilot strategy is the increased emphasis on focus.
A more unified experience across personal and work contexts could eliminate some of the confusion users face today. A closer connection among Copilot experiences could make the product feel less fragmented.
The growing focus on agents may be even more important.
The value of an agent is not that it can hold a better conversation. It is that it can work toward a defined outcome.
Instead of only asking Copilot to summarize activity at the end of the week, an agent could potentially follow the relevant work throughout the week, identify what changed, gather the required information, and prepare the update for review.
That moves AI closer to how work actually happens.
But agents will face the same standard as every other software product. They will need access to the right context, a clearly defined job, appropriate controls, and an outcome valuable enough that people choose to use them again.
The technology may be new. The fundamentals of product adoption are not.
What This Means for ForgeXRM
At ForgeXRM, we build intelligent business applications on Microsoft Dynamics 365, Power Platform, Dataverse, Azure, and the broader Microsoft ecosystem.
Our role is not simply to attach AI to those applications.
It is to organize the business context that makes AI useful.
That means structuring data and relationships, improving how people interact with business information, defining repeatable processes, and creating workspaces through which users—and increasingly agents—can understand what is happening and take the appropriate next action.
We believe well-designed business applications will become more important in the AI era, not less.
AI needs an operational environment. It needs reliable data, clear relationships, governed actions, and a defined understanding of the work. Business applications provide that foundation.
The opportunity is not to insert Copilot everywhere.
It is to build better applications where AI can contribute something meaningful.
Building Around the Work
We have been heads down building something new at ForgeXRM, and it is coming soon.
I cannot share all the details yet, but the product has been shaped by the principles behind this post.
We are not adding AI for the sake of an announcement. We are starting with the work people need to accomplish, the information they need to understand, and the experience that will help them act more effectively.
We will continue to evaluate deeper Copilot integration as Microsoft’s direction and implementation model mature. I am optimistic about where Microsoft is heading, particularly as the conversation moves from AI availability toward agents, business context, and measurable outcomes.
In the meantime, we will keep applying the same test to everything we build:
Does it solve a real job?
Can customers adopt it without unnecessary friction?
Can we confidently stand behind it as a product?
AI still has to earn the right to exist.
So does every feature we release.
