Everyone writes the Fabric adoption post at the six-month mark, once the story's been tidied into a clean arc of triumph. I want to write the honest one, from close enough to the start that I still remember the confusion — a real journal of the first ninety days, wins and surprises and bumps included, before hindsight sands off the interesting parts. If you're about to start this journey, this is the map I wish I'd had: not the demo version, the Tuesday version.

Days 1–30: the honeymoon is real, and partly earned

The first month genuinely does feel good, and I don't want to be so world-weary that I pretend otherwise. The integrated experience — data landing in OneLake, a lakehouse spun up in minutes, a report on top of it the same afternoon — collapses a lot of setup that used to take weeks of provisioning and plumbing. The first time you go from raw data to a working report without stitching four separate services together, the productivity feels almost unfair. That part of the pitch is true.

But here's the first surprise, and it's the one I'd most want to warn you about: the ease of creation is a trap wearing the costume of a benefit. By the end of month one we had a proliferation of workspaces, lakehouses, and draft semantic models, because making them was so frictionless that we made them without thinking. The honeymoon productivity and the future governance mess are the same property of the platform, experienced a month apart. Enjoy the speed; just know that every easy creation is a small future debt, and start a light inventory from day one so you're not archaeologising later.

Days 31–60: reality, mostly in the shape of governance and capacity

The second month is where the platform stops being a demo and starts being your platform, with your mess in it. Two things bit us, both predictable in hindsight, neither obvious in the moment.

The first was governance debt arriving faster than expected. All those frictionless creations from month one had to be reckoned with — which model is canonical, who owns this workspace, why do we have three versions of the same dataset. Nobody had done anything wrong; the platform's ease had simply let sprawl outrun structure. We spent a good chunk of month two doing the governance we should have been doing lightly all along, and the lesson landed hard: with Fabric, governance can't be a phase-two project, because phase one creates the mess phase two has to clean.

The second was capacity behaving in ways the demos never show. Our capacity started throttling, and the cause wasn't dramatic — a couple of models refreshing far more often than anyone needed, a pipeline reprocessing all of history nightly out of habit. Learning to read the Fabric Capacity Metrics app went from "something we'll get to" to "the thing standing between us and angry users" overnight. Budget real time for understanding the capacity model before it teaches you the expensive way.

Days 61–90: what actually stuck

By the third month the novelty's gone and you can finally see what's real. A few things stuck, and a few things quietly didn't, and the honest sorting is the most useful part of this journal.

  • OneLake and shortcuts genuinely changed our habits. Referencing data in place instead of copying it everywhere is the change I'd least want to give back. It cut redundant pipelines and, more importantly, cut the "which copy is right" arguments that used to eat our mornings.
  • Copilot found its real, narrow lane. The month-one fantasy of "everyone just asks questions" faded exactly as I expected it to; what remained was genuinely useful — drafting measures, explaining inherited work — for the analysts who could check it. Useful junior, not oracle. That settled into a stable, honest place by day 90.
  • The semantic model became the centre of everything. Every governance problem, every Copilot quality issue, every "why don't these numbers match" traced back to model definitions. Ninety days in, I'm more convinced than ever that the model — not the lakehouse, not the AI — is where the real work and the real value concentrate.

What I'd tell my day-one self

If I could send one note back to the version of me starting this, it wouldn't be about features. It'd be three sentences. Govern from day one, lightly, because the platform's greatest strength — how easy it makes creation — is also the thing that will bury you if you let creation outrun ownership. Learn the capacity model early, on purpose, rather than late, via an incident. And resist the month-one fantasy that the AI changes the fundamentals; the fundamentals — clean models, clear ownership, agreed definitions — matter more on Fabric, not less, because everything else got so easy that they became the only hard part left.

Fabric's onboarding is genuinely the smoothest I've used. That smoothness is exactly why the discipline has to come from you — the platform will happily let you build a beautiful mess at record speed.

The honest verdict at 90 days

Would I adopt Fabric again, knowing what these ninety days taught me? Without hesitation — but I'd do the second month's work in the first. The platform delivers on the integrated-experience promise, the productivity is real, and the direction of travel is unmistakable. What the demos don't tell you is that its greatest strength and its central risk are the same thing: it makes everything so easy that the only real challenge left is the human discipline of governing what you so effortlessly create.

That's not a criticism of Fabric. It's the actual job now. The platform handles the plumbing it used to take a team to manage; what it can't handle — what it quietly hands back to you — is deciding what's true, who owns it, and what all of it means. Ninety days in, that's where I spend my time, and I suspect that's where I'll spend it from here. The tool got easy. The thinking got concentrated. That's the real story of the first ninety days, and I'd rather you heard it now than discovered it in month two like I did.