The week before Ignite has its own weather. The agendas leak, the session catalogue goes live at some antisocial hour, and my feed fills with people reading tea leaves in session titles — does "unified" mean they finally merged the two things, or just the two logos? I've done this long enough to have a ritual about it. Before the keynote tells me what to be excited about, I write down what I actually want from the week. Then, afterwards, I check my list against what shipped. It keeps me honest, and it stops the keynote from doing my thinking for me.
So here's the 2026 list. What I'm hoping Ignite delivers, and — because I've watched a few of these now — what I'm bracing to sit through on the way there.
What I actually want
Governance that arrives as a default, not a workshop. By this point in 2026 nobody argues that data governance matters; the EU AI Act's high-risk obligations turned "we really should catalogue that" into "we are legally required to be able to explain that." What I want from Ignite isn't another governance product with another name. It's the unglamorous plumbing getting more automatic — lineage that populates itself, sensitivity labels that travel with the data across Fabric and Purview without a six-week services engagement, access reviews a human can actually complete in an afternoon. Put plainly: I want the control that stops an AI answer from surfacing the salary column to become a setting, not a project.
A Copilot I can check. The Copilot features that have survived in my work are the ones that show their working — the DAX assistant that hands you the measure to read, not the black box that hands you a number to trust. So I want fewer "ask your data anything" demos and more of the plumbing that makes an answer verifiable: show me the query it wrote, the columns it touched, the definition it picked. A Copilot that shows its reasoning is a colleague. One that just produces confident output is a liability with good manners. Give me the first, everywhere, and I'll happily let it draft.
Cost controls that treat me like an adult. Fabric's capacity model is genuinely powerful and, on the wrong Tuesday, genuinely frightening — a runaway workload can turn a sensible bill into a board conversation before anyone notices. I want budgets I can set and alerts that fire before the overage rather than in the post-mortem. The ability to cap a workload, not just watch it burn and reconcile the damage later. This is the least glamorous thing on my list and probably the one that would save the most real money.
OneLake staying honestly open. The best thing about OneLake was always that it's a lake you can point other engines at — one copy of the data, many tools. I want that openness deepened, not quietly narrowed: Iceberg and Delta interop that actually holds under load, shortcuts that don't fall over at the edges, so "one copy of the data" stays true and never drifts into "one copy, as long as you only ever read it with our things." Openness you can rely on is worth more than any single feature announced on top of it.
And the unsexy one: lifecycle. When I wrote my Fabric watchlist last year, the item I cared about most wasn't a headline capability — it was proper Git integration, deployment pipelines, real environments. The daily work of shipping data changes safely. Some of that landed; I want the rest. Nobody will cheer for it in a keynote, and everybody will notice the week it saves them.
What I'm bracing for
The agent avalanche. Ignite 2024 was the year everything became an agent, and I don't expect 2026 to be shy about it. I'd put money on "agent" being said more often than "governance," and on at least one tool I already use being reintroduced to me as an agent. Some of it will be real and useful. A good share will be a familiar feature under a newer, more fundable noun. My job that week is to tell which is which, calmly, while the room is still applauding.
Default-on. Something will ship switched on by default, and the announcement will frame it as empowerment rather than a decision made on my behalf. Doing nothing about it will not be neutral. I'll be reading the fine print on tenant settings while everyone else watches the demo.
The demo that doesn't survive Tuesday. Every keynote has the flawless run on the clean, curated dataset. It's genuinely impressive and it is not the job. The job is the following week, on the real estate, where the data is messy and the names lie and the thing has to work anyway. I'm not cynical about the demo — I just know it's the trailer, not the film.
The honest part
None of this is me waving Ignite away. Microsoft ships real, load-bearing things at these events every year, and being reflexively above it all is exactly as lazy as swallowing the hype whole — it just feels smarter. The difficult, useful thing is telling the two apart in real time, which is precisely why I write the list before the keynote instead of after. A wish list made in advance is a test I can't retrofit to make myself look prescient.
So the list is on the fridge. I'll report back once the keynotes are done with the honest scorecard — what actually landed, what was a rename, and what I'm going to be using by December. Until then, I'm going in hopeful, coffee in hand, and bracing for the word "agent."