I used to work on the Power BI team at Microsoft. Here are my thoughts on the future.

I’m a former Microsoft SDE who worked on Power BI and left the company a few months ago. I’m also one of the few who also used to be a customer of our own application.

You are right to question the future of BI software

This has nothing to do with how good or talented the team is. Far from it. Power BI was one of the best-organized and most painless development environments I’ve had in over 8 years of my career across 3 Microsoft teams and 7 different companies.

The truth of the matter is that the changes we are facing are just larger than Power BI

I remember a meeting where one of our engineers built a Power BI report with tables and pie charts showing which feature was in which stage of rollout, and in what region, so we could check if we forgot to push a feature out to Japan, the EU, or America on schedule.

As someone who loved this program for years, my heart sank when I asked the question:

“I hate to bring this up, but is Power BI really the right solution to this problem?”

In hindsight, knowing what I know now about agents, it seems stupid. If I wanted to solve that problem today, we could set up chat and automated email alerts that DM the responsible engineer about their feature and escalate it to their manager if things aren’t rolled out on schedule.

Today we could probably make the agent roll the feature out on schedule itself, adapting to any managerial calls from the top down to delay the release due to unexpected bugs and regressions. Maybe removing the need for this report altogether.

There are just too many new and better ways to solve the same problem now

Your job as report creators, and my job as an engineer, is to “solve the problem,” and you are under no loyalty obligation to keep doing it the same way.

You need to see yourself as a Data Analyst instead of a Power BI Developer

Your job has always been to process, relay, communicate, and most importantly be responsible for the information coming into the business. That job will never change, because the job is bigger than any one tool, even AI.

The thing is, not every company has or will ever need a data analyst. But at some point the CEO doesn’t want to worry about whether the monthly sales bar chart includes purchases made on accounts with checks that haven’t cleared, or whether it’s a live DirectQuery or a daily refresh. Someone still has to own whether that number is actually right.

Is Power BI doomed?

When I joined, people knew my experience making reports in Power BI. I had to tell them I honestly had no idea how measures worked.

I avoided them. Instead, I wrote all my table results and logic ahead of time as a SQL view so all the report’s DirectQuery had to do was a SELECT * FROM TABLE, leaving all the insane calculation logic on the backend.

Even as someone who built reports and worked on the product, I was bypassing half of it. Power BI is a product with 10,000 features where 100% of customers only need at most 10% of them. The chances that you have SQL, Redis, MongoDB, and Kusto data stores all at once are basically 0.

But it has to be that way because every customer needs a different 10% of those features, which is why it exists.

The one-stop-shop solution for everything was extremely attractive in the past because it meant everyone could look at the same thing and see how it would solve their problem.

This isn’t just Power BI. It includes Dynamics (I also worked there), and you can think of plenty of other “all-in-one” solutions.

What we are seeing is people building their own stuff to strip out the 90% they don’t need

Not every report needs to be interactive. Not every report needs conditional formatting. What I worked on was Visual Calculations, which few of you used.

I once made a company use Power BI because I estimated that building our own React and D3 web page to serve that one report would take about 6 months at best.

Now I would estimate it at maybe a week. The calculation is much different today.

So, you decide to vibecode your own HTML/JS/D3 report?

Be careful not to vibecode yourself into tech bankruptcy.

One of the black magic things about Power BI was its ability to take a DAX statement like SUM(Sales[Amount]) and, on DirectQuery, translate that on the backend into any flavor of SQL, KQL, or MQL. And do that while staying performant.

I have no idea how it works.

That’s on you now.

Your one click to filter all the items on a bar chart just kicked off 17 SQL queries, one for each visual, each making 3 layers of OUTER APPLY, and now it has locked up the database.

Maybe Claude can fix it. But these production problems are going to happen, and if you can’t find anyone else to blame, if you can’t find Microsoft to blame, they are going to blame you for it.

Behind those 10,000 features are hundreds of thousands of hours of engineering effort trying to deal with every edge case and every performance consideration, and we still got it wrong. Claude will still get it wrong, and you will also get it wrong.

Still, that doesn’t mean inheriting an entire all-in-one system when you don’t need 90% of it. If the all-in-one tool is bloated and vibecoding from scratch is a trap, what’s left?

No one’s dream car involves bespoke handmade airbags

If we look at Power BI like a car, it was a car built to satisfy the most people possible.

AI now makes it possible to build exactly the car you want, down to remaking every single component of it.

Just because you can doesn’t mean you should.

Do you want your own bespoke handmade airbags in your car?

Maybe you’re willing to take the risk of them not working on yourself, but what about handing it off to other people?

You tested that it works when hit at a 90-degree angle at 30 mph, but what about 26.77 degrees at 25 mph?

You just want them to work.

Even when they don’t work, you want someone to blame, like Takata. Make it someone else’s problem.

The future is likely modular components that someone else is responsible for

You bolt various components together to make your dream product.

I’ve been following the people at Convex. Among all the AI Doomer and AI-Bro garbage, it’s been a breath of fresh air: they point out that over the course of engineering history, as the build gets easier, the products themselves get more complicated.

Responsibilities shift. What was once the job of making Power BI reports that hooked one of many SQL systems to a single bar chart is now split into pieces.

Your job is to figure out how to Lego someone’s reliable and performant SQL Server query system onto someone else’s platform-agnostic bar chart visual. Use AI to build the custom glue and UI shell in a week, not to reinvent the airbags. Each part, much like the tires, airbags, and seats in your car, is made by someone or some company that has dedicated itself to the massive complexity of that one domain to make it their problem and not yours.

Can Microsoft/Power BI adapt?

Maybe. Those are some of the smartest, longest-running people I have seen in a company famous for reorganization and turnover.

But waiting for them to figure it out is putting your future in their hands and hoping for the best.

I can really only advise doing what you think is the best thing for yourself, which is what I decided to do when I left.