I first tried Microsoft AI in 2023, when the company introduced Copilot as its “everyday AI companion.” ChatGPT was all the rage, and the idea was genuinely exciting: AI inside Word and Outlook, no more window-switching, no more cutting and pasting.

The reality was underwhelming. Copilot felt bolted onto Office rather than built for how people actually worked — awkward, inconsistent, unforgiving if you didn’t phrase the prompt just right. We bought licenses and gave it a serious trial. The reaction was mostly a yawn.

Then Came The Metamorphosis

Over the next three years, nearly everything changed.

Microsoft put AI everywhere — Word, Outlook, Teams, PowerPoint, Excel, GitHub, Azure, Dynamics, Power Platform, Defender, Sentinel, Intune, Entra. No company has moved faster or broader to plant AI inside the tools people already open every day. That was the easy win.

But shipping AI everywhere and making AI good everywhere are two different jobs. On strategy, Microsoft earns an A. On experience, it’s closer to a B-minus.

The Model Isn’t the Advantage. The Real Estate Is.

Microsoft doesn’t need the best model. It needs the best address. It already lives inside your inbox, your meetings, your documents, your identity system, your security stack, your cloud. That’s not a feature. That’s an occupation.

A chatbot can write you a memo. Copilot writes that memo pulling from yesterday’s Teams call, last week’s deck, and the numbers sitting in Excel — Microsoft 365 Search finds the source material, and Copilot Pages turns the conversation into a persistent workspace colleagues keep developing together. One is a party trick. The other understands how work actually moves through a building.

AI Everywhere, Consistency Nowhere

Copilot drafts in Word, summarizes in Outlook, recaps in Teams, and in GitHub increasingly just writes the code itself. It runs the infrastructure in Azure, sharpens security in Defender, Sentinel, Intune, and Entra, and automates workflow in Power Platform and Dynamics. No competitor owns this much surface area. Nobody’s close.

But ambition isn’t the problem. Inconsistency is. Copilot is often useful, but not usually as sharp, conversational, or intuitive as ChatGPT or Claude. And the naming is a maze: Copilot Chat, Microsoft 365 Copilot, Copilot Studio, GitHub Copilot, Security Copilot, plus a dozen app-specific variants. Microsoft calls this a platform. Users experience it as a pile of similarly-branded products with different prices, limits, and personalities.

Some integrations earn their keep — Word and PowerPoint are made for this, and GitHub Copilot has a job and does it. Excel is trickier: the AI is only as good as the structure underneath it, and most spreadsheets are not structured. Which points to a harder truth: Copilot rewards people who already know the tool. AI doesn’t erase expertise. It taxes its absence.

Copilot’s Real Enemy Is Your File System

Microsoft’s biggest strength doubles as its biggest exposure. Copilot — and its agents — can only find what’s findable. Feed it a SharePoint site with inconsistent naming, stale documents, and permissions nobody’s audited since 2019, and it surfaces that mess at the speed of AI. Sometimes it can’t find the right file; worse, sometimes it finds the file too many people already had access to.

Copilot doesn’t create your permissions problem. It exposes it — fast, and in front of an audience. Fix the data governance, the access controls, the retention policy, and the file hygiene before rolling Copilot out widely, and before you turn an agent loose on it. Otherwise you’ve bought a Ferrari and parked it on gravel.

Nobody Knows Yet if the Math Works

Microsoft has sold the enterprise on AI. It hasn’t proven every seat is worth the invoice. Some employees live in Copilot and save real hours; others open it once a month to rewrite an email. Blanket licensing across an entire org, without knowing which is which, is a hard number to defend to a CFO.

The deployments that actually work start with sharper questions:

• Who spends the most time producing, reviewing, or hunting for information?
• Which recurring task is actually shrinkable?
• Where does AI cut delay or error, not just add novelty?
• Whose data and permissions are clean enough to trust it with?
• How will you know, three months from now, whether it worked?

Microsoft made AI easy to buy. It didn’t make it easy to justify. That part’s still on you.

The Quiet Smart Move: Going Model-Agnostic

Microsoft’s sharpest strategic bet gets the least attention: it’s stopped tying its fortunes to any single model. No one model will win everything — one writes better, one codes better, one reasons better, one simply runs cheaper and faster inside a locked-down enterprise environment.

Microsoft wants to be the layer that picks the right model for the right job, invisibly. The end user shouldn’t need to know which model answered — they just need the answer to be accurate, secure, governed, and sitting inside the app where the work already lives. That role, quiet arbiter of models, may end up worth more than owning the flashiest chatbot on the market.

Round One Goes to Microsoft

Distribution is won. AI sits inside the tools millions of organizations already use, pay for, and trust enough to buy more of. Round two is harder: Microsoft has to turn that footprint into an actual AI operating system for work — one that understands the organization, moves among its applications and data, and performs work rather than merely suggesting it. That means making Copilot less confusing, more consistent, and undeniably useful — not just present.

The winning formula was never just Copilot. It’s:

Copilot + clean data + correct permissions + real training + carefully chosen workflows.

Skip any of those, and Copilot becomes an expensive button people forget exists. Get them right, and Microsoft’s head start becomes very hard to catch.

Microsoft won the distribution battle. The experience battle is still being fought — one badly-permissioned SharePoint site at a time.

Have ideas, questions, or thoughts to share? I’d love to hear from you—feel free to reach out anytime at g.*******@***et.com. And if you want more insights like this, subscribe on our website to get regular tips on cybersecurity, IT management, AI tools,  and more delivered to your inbox weekly.