Getting ready for Microsoft Ignite 2026 in San Francisco

I’m flying to San Francisco on Saturday for Microsoft Ignite 2026, and I’m looking forward to a week of talking about data, AI and what happens when people start relying on the things we build. Ignite runs from 17 to 20 November at the Moscone Center. I’ll be speaking three times, spending time with the Microsoft data community and sharing what I learn here and on LinkedIn as the week unfolds.

The question I’m taking with me is one I keep coming back to in my training and consulting work. What evidence would give you enough confidence to put an AI tool into someone’s working day? That applies whether you’re giving a team Microsoft 365 Copilot for the first time or letting managers ask a Fabric data agent questions about their numbers. It’s also where my Ignite sessions start.

My Microsoft Ignite 2026 session on testing Fabric data agents

My session is called Can You Trust Your Fabric Data Agent? Test It and Prove It. I’m giving the same 15-minute lightning talk on Tuesday and Wednesday, then hosting a longer table talk on Thursday. If you’re at Ignite and working with Fabric, Power BI or AI adoption, I’d love to see you there.

Pick either lightning slot and add it to your Ignite schedule. Lightning talk seating is open, so come a little early if you can. For Thursday, search for TT1780 in the Ignite session catalogue and reserve your seat. Open the lightning talk listing or find the session catalogue on the Ignite website.

A fluent answer is not the same as a correct one

Here’s the sort of problem I want to get into. Say a sales manager asks, “How much did we sell last quarter?” The agent returns a precise figure. It looks sensible. But did it use the financial quarter or the calendar quarter? Did it count orders placed or revenue recognised? How would the person asking know which definition it chose?

Anyone who has spent time reconciling two Power BI reports will recognise the problem. A fluent answer can make the discrepancy harder to spot.

What the lightning talk and table talk cover

In the lightning talk, I’ll cover how to start testing a Fabric data agent before people depend on its answers. You’ll see a before and after: an agent working from a weak schema and undefined business terms, then the same questions put to a fixed version, scored against a set of known correct answers with the generated T-SQL on screen. If you want to read ahead, Microsoft documents how to evaluate a Fabric data agent against a ground truth set.

At Thursday’s table talk, we’ll have time to get into the awkward cases. What questions would go in your test set? Who signs off the expected answers? What do you do when an answer is plausible enough to make it into a board pack? If you’ve had a convincing wrong answer from a data agent, bring the question that produced it, with the sensitive details removed. I’d like to hear how you caught it.

What I’m hoping to get to during the week

Monday and Tuesday. I’m aiming to start with Ignite 101 on Monday and get my bearings. On Tuesday, I’ll be at the opening keynote, then I want to spend time on Fabric IQ and hear how other teams are approaching Copilot adoption. There’s usually a difference between the questions you hear in a keynote and the ones you hear around a table. I want to hear both, especially where the people who own the data and the people rolling out AI have different views of what’s ready.

Wednesday. After the morning lightning talk, I’m looking at AI return on investment, FinOps and what smaller IT teams can realistically take on. That last one matters to many of the organisations I work with. The same team that supports Microsoft 365, keeps reporting running and deals with day-to-day incidents is often asked to run the AI pilot as well. I’m interested in what those teams chose to measure, how much time the work took and what they decided to stop doing. I’m also hoping to catch up with people at the Ignite Celebration that evening.

Thursday and Friday. Thursday starts for me with the table talk. Beyond that, I’m looking at agent governance and how charities and other organisations with tight budgets are making AI useful. On Friday, I’m hoping to catch sessions on frontline work and Microsoft Purview. I want to see what teams are doing to make AI useful in the middle of a shift or a customer conversation, when people have very little time to check an answer. That changes what you need to test before rollout.

I’ve left some gaps in the diary for the Hub and for conversations after sessions. I’m also hoping to record a couple of episodes of Putting the Human into Technology while I’m in San Francisco. I’d love to talk with people who’ve taken an AI or data project into day-to-day use and can tell me what they had to change along the way. The story of what happened after the pilot is often the one I most want to hear.

What I’ll share as Ignite unfolds

I’ll post short updates here and on LinkedIn during the week, with a fuller write-up once I’ve had time to go through my notes. When something catches my eye, I’ll look at who can use it today, what’s in preview, what it takes to get it running and what it could cost to keep it running. I’ll also share the questions and examples from conversations that make me rethink how I’d scope a Copilot pilot or test a Fabric data agent.

If you’re following from the UK and there’s something you’d like me to look out for, drop me a note. A specific question about how something would work in your organisation is particularly welcome.

If you see me in San Francisco, come and say hello

Whether we’ve worked together, met at a user group, crossed paths on a course or only know each other through LinkedIn, do come over and say hello. I’ll be around the Hub, at my sessions and out and about between them. Tell me what you’ve come to learn, what you’re building or what you’ve seen that’s worth making time for. And if you can make one of the Fabric data agent sessions, I’ll see you there.

Not at Ignite? Start with the questions that matter

You don’t need a conference pass to work out whether your organisation is ready for Copilot or agents. If your team is weighing up where to start, my AI Readiness Assessment shows you which questions to answer before you choose a pilot.