👷 Everybody Talks About AI (And They Are All Wrong)
The AI adoption race is built on a benchmark that doesn't exist, and while you're busy trying to catch up, your actual job is full of problems nobody's fixing.
Thank you to Cube for sponsoring this issue.
If you’ve been in data for a while, something feels different now.
The job you spent years getting good at has quietly changed around you. Everything is different, now. Not just tools and pacing. What leadership expects from you barely resembles the job description you, barely resembles the job you signed for.
You are watching a job you love change in front of you, without feeling ready for what it’s becoming.
Here’s what I think is happening, and why most of what you’re reading about it is making it worse.
What’s your leadership experience
The Truth About AI in Data
You know how the famous saying goes.
AI is like teenage sex. Everybody talks about it. Everybody thinks everybody else is doing it. Nobody really knows how to do it. And if you are actually engaged in it, you do it really badly.
You just need to open LinkedIn, or read any of the big data engineering blogs, and you’d think you’re dangerously late. You’d think every other company already has autonomous agents in production, a fully deployed semantic layer, and AI-generated pipelines shipping while the team goes for coffee.
So you shift roadmap, add a tool or two for evaluation, and somebody ships a workflow nobody asked for against a deadline nobody set. When it goes badly, nobody understands why, because the whole thing is about catching up to a benchmark that doesn’t even exist.
And you know what? Nobody is ready for this. Everybody is doing stupid shit and calling it strategy.
Companies used to have one goal: make more money. This goal is tangible and measurable. You knew if you were winning. Now the goal is “adopt AI“. And when you cannot define success, you cannot recognize failure either.
This issue is brought to you by Cube.
The best way to support me and this Newsletter is to check out the Sponsor below and click the link.
Cube Is the AI-Native BI Platform for Your Databricks Lakehouse
Cube is the agentic analytics platform for teams on Databricks. One platform for internal BI, embedded analytics inside your product, and an AI Analyst that answers questions in plain English.
All of it runs on a governed semantic layer, so a metric defined once means the same thing in a dashboard, a customer-facing app, and an AI query. Databricks holds your data. Cube is what your team, your customers, and your agents actually ask.
The Cube team is running a free 45-minute webinar on June 30 (1:00 PM ET / 10:00 AM PT) with a live demo connecting Cube to Databricks and exploring the same governed metrics through a dashboard, an embedded app, and the AI Analyst.
If you’re on Databricks and deciding what to standardize on, this is the session to watch.
What the Race Produces
Data democratization
Every platform now enables agentic workflows. And since building is trivially easy, everyone builds their own thing. Your CEO probably has a workflow that pulls data from your BI tool. And your head of finance has one pulling from a different source, with a different definition of the same metric.
Call it democratization if you want. To me that’s decision surface area with no shared truth underneath it.
Give people access without defining what the metrics mean, and your org starts running on wrong numbers. At some point everyone’s dashboard shows something, and something feels like progress.
Don’t get me wrong, this problem existed forever. Now, it’s just appears at scale.
Staling innovation
There’s a deeper problem though, and this one bothers me more.
AI learns from existing code. It reproduces what the industry already built. AI can’t invent what nobody has built yet.
Things like DuckDB, Polars and Arrow came from people doing non-consensus work. A model trained on yesterday’s patterns would never be able to produce these.
If you outsource thinking to AI, the field converges. You get well-executed mediocrity, forever. I don’t think enough people are saying that out loud.
One thing that held though is code. Whatever AI generates, you can read, version, and test. The craft is the same.
The Mess Is Your Backlog
Here’s what I’m watching happen.
The more carelessly people adopt AI, the more broken processes pile up. All these AI-generated assets that I told you a second ago are the natural output of an industry that moved fast without thinking.
And they are your backlog. Try this process just once:
Go talk to someone who isn’t on the data team.
Find the process that makes them want to throw their laptop.
Ask what they’re doing with AI and watch what happens.
I find something worth fixing almost every week.
And you know what? The more people use AI carelessly, the more stupid shit piles up for someone like you to fix.
The Raise Of Governance
That’s also why governance matters more right now than it has in years. Keeping your org from drowning in wrong numbers is the work.
Semantic layers used to feel like a gimmick to me. A nice idea that never justified the overhead.
I’ve changed my thinking on that. When your CEO, your CFO, and your VP of Product are each pulling “revenue“ from three different sources, a semantic layer is the only thing that gives you a shared definition both humans and machines use.
One source of truth that doesn’t depend on who built the dashboard.
I’m working on a project that helps us solve this in my org, and I’ll post a proper write-up when I have results.
Final Thoughts
I don’t know where this is going. Nobody does, and I’m skeptical of anyone who says otherwise. But I’m not worried, and I don’t think you should be either.
I believe you are fine as long as you stay current, solve real problems and skip the hype.
Every new type of work has something interesting in it if you go looking, and the people who adapt are always the ones who go looking. The job title might change, and the tools will definitely change. But the skill of finding a real problem and fixing it doesn’t go anywhere.
If this whole thing goes sideways, I’m genuinely good at building, electricity, and plumbing, although that last one is not my favorite. That was never the plan, but it’s a real option in case of disaster.
—
Until next time,
Yordan
PS: If you’re an experienced engineer trying to figure out the next move, the premium library is where I put everything on navigating the leadership ceiling, stakeholder politics, and getting paid what you’re worth. Chek it here.
PPS: If you’re figuring out your next move, a promotion, a raise, or what your career actually looks like in this market, I do one-on-one coaching with data people. Learn more when you are ready.
Let’s Connect
Connect on LinkedIn: https://www.linkedin.com/in/ivanovyordan
Work with me: https://www.ivanovyordan.com/coaching
Start journaling: https://www.dearself.ai
Governance is key, I am finding, with my own buildout of Semantic layers. Who has access to change the context and definitions of the underlying data determines how the AI will be engaged.
As for my role changing...While I had heard of Semantic layers a year ago, I had barely scratched the surface.