This is Part 5 of The Real Cost of Agentic AI series.
Summary
Microsoft’s Copilot vision is often talked about as if it’s a single, omnipresent assistant, but in reality “there is no one Copilot”. Different AI features and Copilot-branded products are gated behind specific licenses and add-ons, leading to surprises when customers discover certain capabilities aren’t included by default. This post examines how Microsoft gates AI features (across M365, Dynamics 365, Power Platform, etc.), why they do it, and the impact on customers trying to navigate these artificial walls.
What’s happening
Despite the cohesive marketing around “Copilot” as if it were one product, Microsoft has in fact fragmented AI features across many offerings. For example, if you have Microsoft 365 Copilot (the $30/user add-on for Office apps), you might assume you get all the nifty AI goodies. But if you read the fine print, some features are only available in certain contexts or SKUs. One recent scenario: Microsoft announced that any user can generate images via Copilot Chat without needing an M365 Copilot license – a nod toward making Bing Image Creator accessible in enterprise chat. Yet days later, clarification implied image generation would be limited unless you had the proper licensing, causing confusion.
Similarly, within the Power Platform, the so-called “AI-powered” features often require specific licenses: e.g. Power Automate’s AI capabilities might need an AI Builder credit allocation or a special add-on. In Dynamics 365, each module (Sales, Customer Service, etc.) has its own “Copilot” features, and they typically require that you not only have a license for that module, but in some cases, an additional enablement or preview flag (some started in trials or as free previews, leaving Microsoft free to monetize them later).
This is a per-user, per-agent add-on model, albeit framed as a single $50 license that includes both the base Copilot and the service agent. We can imagine this pattern repeating: today it’s a service agent, tomorrow it could be a sales Copilot agent, a finance Copilot agent, etc., each as a separate licensing component.
We also see feature gating via product tiers – an AI feature might only light up if you have the premium tier of a product. A concrete example: If you’re a Dynamics 365 Sales customer on an Enterprise license, you might not have the entitlement for all of the AI features promoted in release wave keynots that a Sales Premium license offers, or you might need to purchase an add-on.
In short, Microsoft’s approach has been to embed AI into every product but not universally unlock it – you must pay or upgrade to access each Copilot instance. This is what we mean by AI feature gating: the placement of licensing toll booths in front of various AI capabilities. So a customer might encounter messages like “to use this feature, please upgrade to the premium plan” or find that a button is greyed out because they don’t have the required license.
We’ve also seen Microsoft change these gates on the fly: something in preview might be free, but then a licensing guide update announces it will require an add-on SKU moving forward. This could even apply to administration related capabilities, such as Data Loss Protection (DLP) rules for Power Automate desktop flows (meaning RPA bots) requiring the use of Managed Environments and thus premium licenses from all users accessing them.
Why it matters
For customers and partners, AI feature gating can lead to confusion, misplanning, and missed expectations. If you believed the marketing and assumed Copilot would just be there, you could be in for a rude awakening when you realize each Copilot is separate. This matters when budgeting projects: you might budget for Microsoft 365 Copilot licenses, only to find out that to use Copilot in Dynamics 365 Customer Service, there’s another cost. Or you deploy Copilot Chat for free and users love generating images, but then a policy change requires M365 Copilot after all – suddenly you have to justify a new purchase to keep a “feature” that users thought was included.
It’s also impactful for user experience and adoption. Imagine rolling out a fancy new “AI-driven” feature in an app, only for users to hit a paywall (“Upgrade to use Copilot agent with this content”). That can sour the perception and trust in these tools. On the flip side, from Microsoft’s perspective this gating is about monetization and managing resource usage – they can’t afford (literally) to give every user unlimited AI if they haven’t paid for it. So they gate to ensure only paying (or trial) customers consume the expensive AI compute. However, this creates a hodgepodge licensing landscape: one that IT admins and procurement have to piece together.
It matters because it increases the complexity (remember the days of a single Office license? Now we have base licenses plus multiple Copilots and add-ons). For partners or consultants implementing solutions, it’s an extra layer of requirements to check. A solution might require turning on a certain AI feature, which in turn requires confirming the client has the right licenses – if not, that’s an awkward conversation about “we need an extra $X to get what was demoed”.
Furthermore, feature gating can hinder innovation on the customer side. Teams might not even try something if they aren’t sure it’s included in what they already pay for. It adds friction – needing approvals to unlock a feature slows down experimentation. Another impact is inequity among users: perhaps only certain departments get the fancy AI features because the company can’t afford to license everyone. That can create haves and have-nots within the organization, which has its own cultural and productivity implications.
Finally, surprises are never fun in enterprise IT. Nobody likes finding out mid-project that you need to go back for more budget. It’s always better to know upfront. So it matters a great deal to get clarity on what’s gated where, so you can plan properly and manage user expectations. (We often find ourselves simply mapping out these gates for clients – a sanity-saving step.) If Licensing.Guide achieves anything, hopefully it’s to eliminate the phrase “Wait, that’s not included?!” from our readers’ vocabulary.
My perspective
Microsoft’s fragmented Copilot strategy is a double-edged sword. On one edge, I understand it – different products, different value props, and yes, Microsoft is a business that will charge for premium capabilities. But the other edge is sharp: it cuts users with confusion and a sense of bait-and-switch. My take is that Microsoft isn’t being as upfront as they should be about the “no one Copilot” reality. They paint a future where Copilot is everywhere, but don’t emphasize that each instance might be separate.
In my hands-on dealings, I’ve had clients say “We have Copilot, why can’t it do X in this app?” – only to explain it’s a different Copilot needed. I sometimes feel like I’m re-living the Microsoft 365 E1/E3/E5 upsell explanations, but now with AI: “Well, you have this Copilot, but not that Copilot.” It’s a bit ironic because AI was supposed to simplify our work, yet it has complicated our licensing.
I tend to be contrarian when the hype is too high – and Copilot hype is sky-high. So here it is: Copilot is not a single product, it’s a brand across many siloed offerings. Accepting that is step one. Step two, we push Microsoft to reduce unnecessary gates. Some gating is understandable (you can’t give away GPU-intensive services for free), but does every little AI assist need a separate SKU? Perhaps over time they’ll bundle more (maybe Copilot will become more unified once adoption hits critical mass).
Until then, customers must be vigilant. My advice is to catalog the AI features you think you’ll need and cross-check them against your licenses. Use Microsoft’s documentation (and independent sources) to find the catches. For instance, if you plan a project around Power Platform’s AI Builder or the new Canvas AI copilot, make sure you’ve accounted for those credits or licenses.
In my consulting practice, we literally make checklists of “features vs. license required” so nothing sneaks up. Microsoft’s not going to put a big red sticker on a slide saying “this feature requires additional purchase” – that’s on us to figure out. I also encourage folks to leverage trial periods and preview programs to the fullest: not only do you get to test functionality, you can often test the licensing (Microsoft sometimes opens gates during preview, and then you see them close when the trial ends – that’s insightful). If something critical to you is gated by a pricey license, that’s a signal to evaluate how badly you need it or if there’s a creative alternative. Sometimes, a clever use of what you already have can approximate a gated feature (for example, using Azure OpenAI directly for a chatbot instead of waiting for a formally licensed Microsoft product).
As a final take, I’ll say: Don’t let the gating drama discourage you, but don’t ignore it either. There is real value in these AI features, enough that Microsoft feels justified charging for them. Often, they are worth it if properly applied. The key is going in with eyes open and budget aligned. And if Microsoft’s approach frustrates you, voice that feedback – they do occasionally listen, especially if uptake is slow due to licensing. In summary, expect the gates, understand why they’re there, and plan accordingly. (And if you need a map through this maze, well, that’s what advisors are for!)
