The Question Nobody Asked
A Report on the Future of Data Engineering That Can't Imagine It Without Data Engineers
I agree with the problem. I disagree with the conclusion.
MIT Technology Review Insights published a report on how AI is redefining data engineering. 400 senior technology executives surveyed. 72% say data engineers are integral to the success of their business.
I agree with the problem. I disagree with the conclusion.
First, some context. The report is sponsored by Snowflake. Snowflake sells to data engineers.
That doesn't make the numbers wrong. But it shapes which questions get asked. Every question in the survey assumes the data engineer stays at the center of the process. None of them asks the obvious one: what if the data layer AI needs could be built without a data engineering team?
To be fair, I run a company with its own point of view. So weigh my opinion the same way.
I've seen this from both sides. I spent nine years at YPF, where my team produced 92 reports a month. Then I led data teams in insurance and banking, with more than 40 data scientists and engineers. I hired data engineers. I know how hard they work.
And I know their effort was never the bottleneck.
Look at the report's own numbers. Data engineers spent 19% of their time on AI projects in 2023, 37% in 2025, and expect 61% in two years. 77% of executives say their workload keeps getting heavier.
The report presents this as a challenge to manage. I see a bottleneck.
If every new AI use case needs a person to build a pipeline, more AI means a longer queue. Making engineers faster doesn't fix a queue that grows faster than they do.
Second point. The report calls agentic AI “the next big challenge.” Only 20% of companies had deployed it when the survey was run.
I built my first agent in 2021, before anyone called them agents. And the market is moving faster than the report suggests: Gartner's February 2026 Top Trends in Data & Analytics report predicts that by 2029, AI agents will automate 75% of data engineering workflows.
75%. That isn't a tool that helps the data engineer. That's a different way of doing the work.
Third point. The top challenges respondents cite with new AI tools are integration complexity (45%) and tool sprawl (38%).
Of course. If the answer to every problem is one more tool on top of the stack, the stack grows. The fix isn't adding AI to each layer. It's collapsing the layers.
And the recommendations? Courses on machine learning frameworks. Cross-functional teams. Better communication skills.
All reasonable for a person's career. But for a company, that means years. Meanwhile the CFO still needs the number. Today.
Where the report is right, it's very right. “There is no AI without data.” Completely agree. That's the reason Teramot exists.
It's also right that security is the top concern, cited by 55%. Agents that touch a company's data have to run inside that company's own cloud, read-only, without copying data anywhere else. There's no shortcut there.
So no, I don't think data engineers disappear. What I think is that the job this report describes, building pipelines one by one so AI can work, is exactly the job agents should take. What remains is the part the report itself praises: architecture, rules, judgment.
Here's the core point. Language models can't connect to a company's databases, understand them, and use them on their own. That's the biggest problem in enterprise AI.
Teramot already solved it. Its agents build the full data infrastructure end to end, down to calculated results tables in the gold layer, traceable to the source, which the company's own AI uses through MCP. No implementation project. An SAP instance with 70,000 tables usable by AI in under an hour.
The question isn't how to make data engineers 20% more productive. It's why every company still has to build its data layer by hand.
That's what we're changing.