I have a prediction: "data scientist" is going to become a very uncool job title.
I say this with some affection because I used to have an even worse one: econometrician. And I learned this lesson long before anyone was talking about AI.
When I first showed up on the 101st floor of the World Trade Center to work on a trading desk at Lehman Brothers, I thought I was pretty cool and pretty smart. I had an HP-12C calculator. I knew reverse Polish notation. I could do bond math. I had spent a lot of time learning how to do things that most people couldn't do.
Then I got to the desk and saw the guy sitting next to me working in Lotus 1-2-3. My world changed. It wasn't that he understood bonds better than I did. He had a tool that let him manipulate numbers, test assumptions and answer questions in seconds that were painfully slow to work through the way I had learned. My expertise still mattered, but a big part of what I thought made me special suddenly wasn't so special.
I think we're at one of those moments again.
For a while, data scientist was one of the coolest things you could be. You knew Python and R. You knew how to build models and manipulate enormous datasets. You could answer questions that the person sitting next to you couldn't answer.
GIS had its own version of this. If you understood coordinate systems, raster and vector data, spatial joins, remote sensing and satellite imagery, you possessed another specialized superpower. I don't think either one goes away. I think something more interesting happens: they become so ubiquitous that we stop thinking of them as specialties.
That's what happened with spreadsheets. There was a time when manipulating financial data on a computer was a specialized skill. Today, Excel is open on practically every desk in corporate America. Nobody introduces themselves as "the person who knows how to use a spreadsheet." The spreadsheet became infrastructure. I think we're watching the same thing happen to data science and GIS.
For most of the history of enterprise computing, there has been a huge gap between having a question and getting an answer from data.
You might want to know which facilities are at risk from a storm, where you are most exposed to a supply chain disruption, or which parcels satisfy twenty different conditions. But knowing the question wasn't enough. You had to find the data, get access to it, understand the schema, clean it and join it. If geography was involved, you needed someone who understood the spatial relationships. If time was involved, someone had to align the temporal relationships. Then someone had to build the model and turn the result into something useful.
That complexity created entire professions: data engineers, data scientists, GIS analysts, remote sensing specialists, database administrators and BI analysts. They existed because working with data was hard.
AI is changing that, but I think there is a much more interesting consequence than simply doing the same work faster: what happens when people stop being afraid to ask the question?
There is a tendency right now to think AI means putting a chatbot on top of enterprise data. Ask a question, get an answer. Fine. But the hard problem isn't whether the model understands your sentence. It is whether the model understands your world.
Ask which distribution centers are at risk from a storm and suddenly the system needs to understand what a distribution center is, which ones are yours, where the storm is going, what inventory is inside those facilities, which shipments and customers depend on them, what roads connect them and what "at risk" actually means. The English is easy. The context is hard. And more and more of that context is spatial and temporal.
One of my cofounders recently made an observation that stuck with me. We were talking about a large energy company and its GIS organization, and he basically said, "You know those people aren't going to have the same jobs much longer, right?"
It sounds harsher than he meant it. His point was exactly the opposite of saying geography doesn't matter. Geography matters so much that it can no longer live inside a specialized department.
A pipeline has a location. A truck has a location. A storm has a location and a trajectory. A shipment has an origin, destination and route. A satellite image describes a place at a particular time. So does a forest, an oil field, a battlefield and a supply chain. Once machines begin reasoning about the physical world, spatial and temporal context isn't a special category of analysis anymore. It is simply part of the data.
So what happens when an agent can find the imagery, understand the coordinate systems, identify the relevant datasets, connect them to enterprise data and run the spatial analysis? Geography didn't disappear. The specialized interface to geography did.
I have started thinking differently about what my cofounders and I did at SeerAI over the last six years. We weren't sitting around trying to forecast what enterprise computing would look like in 2026. We had a pretty strong conviction about where it had to go, and we started building for that world before most of the pieces were there.
We assumed it eventually wouldn't make sense to separate "geospatial data" from the rest of enterprise data because everything in the physical world exists somewhere and at some time. We assumed companies couldn't keep solving every new problem by moving another copy of their data into another centralized repository. And we assumed machines would eventually need to understand not just data, but context: what something is, where it is, when it existed, what it is connected to and what other information describes it.
Most importantly, we thought the interaction with data would eventually flip. Instead of starting with a dataset and asking, "What can I do with this?", you would start with a question and ask, "What data do I need to answer it?"
Six years ago, some of that felt pretty far out there. So we built federated data infrastructure, knowledge graphs and spatiotemporal compute around it. Then foundation models and agents exploded, and suddenly the reason for putting those pieces together became a lot more obvious.
The future caught up. We don't have to forecast a world where an executive, scientist or operator can ask complicated questions across enormous amounts of heterogeneous data without first knowing which database to query, which GIS tool to open or which specialist to call. That world is arriving now.
And this is the part I think matters most: removing all that machinery between the person and the question doesn't just make the old questions easier to answer. It changes what people are willing to ask.
For years, people learned what questions they were allowed to ask based on the limitations of their tools. If answering something required six months of data engineering, you stopped asking. If it required processing three petabytes of imagery, you stopped asking. If it required reconciling twelve databases owned by six departments, you stopped asking.
Eventually, you learned not to ask in the first place. You knew what asking meant: open a ticket, call the data team, find the GIS person, get budget, explain the question six times, wait three months and perhaps discover at the end that you had asked the wrong question.
We became conservative about curiosity. I think that is one of the biggest hidden costs in enterprise data. We measure infrastructure, software, people and time. We don't measure all the questions that never got asked because somebody assumed the answer would be too expensive, too difficult or impossible to get.
This is why our line at SeerAI, "answer questions you didn't know you could ask," has taken on a different meaning for me. I used to focus on the answer. Now I think the important word is ask. Remove the friction and you can ask a question, get something back, change an assumption, add a dataset, test another idea and try again. You can follow your intuition and be wrong cheaply. You can experiment with things that never would have justified a six-month data science project.
That changes what the human superpower is. The scarce skill isn't knowing the most obscure GIS command or being the fastest Python programmer. It is curiosity combined with judgment: knowing enough to ask an interesting question, recognize when the answer isn't quite right and keep pulling the thread. The experts don't become less valuable. The best ones become more valuable because they can spend less time operating the machinery and more time thinking.
Which brings me back to Lehman. Lotus 1-2-3 didn't make understanding finance irrelevant. It made it easier to change an assumption, run another scenario and ask "what if?" Thirty-something years ago, I walked onto that trading floor carrying an HP-12C and discovered that a new tool had changed not just how quickly we could calculate, but how we could think.
My cofounders and I have spent the last six years building for another one of those moments. And now that it is here, I don't think the biggest change is that we're going to get answers faster. It's that we're going to ask a hell of a lot more questions.
Use the tool. Or be the tool.