If Your Data Engineer Can Be Replaced by an Agent, You Already Hired Badly

Somewhere, right now, a VP of Engineering is standing in front of her data team giving The Speech. "AI is not coming for your jobs. It's a tool. It's going to make you more productive. We're not planning any cuts."

Three weeks later, half the team is gone.

She wasn't lying. She just didn't know — or didn't want to say — what she meant.

AI is not coming for data engineers. It's coming for the work a lot of people were doing under the title data engineer. The layoff happened at hiring, not at automation.

Q1 2026: ~80k tech workers laid off. Almost half blamed on AI or automation. Stanford's research shows employment for software developers aged 22 to 25 has fallen nearly 20% from 2024. In the UK, entry-level tech roles down 46-53%.

Junior software roles collapsing. Senior engineering roles? Still growing.

But let's be honest — a chunk of that "AI-attributed" number is cover. Companies over-hired in 2021-2022 at zero-interest-rate valuations, inflated titles to keep people, and now need a narrative that isn't "we made bad bets". AI gives them one. Even Sam Altman admitted there's "AI washing" happening with layoffs.

Fine.

What's uncomfortable is: the engineers getting cut under either banner — real AI displacement or over-hire cleanup — are the same engineers. The ones doing work that never warranted the title in the first place.

Let's talk about what agents already do well.

Writing a DAG. Joining three tables and writing to a staging schema. Generating the dbt model. Drafting the test file. Scaffolding the ingestion job. Writing the Airflow operator. Generating the schema migration. Wiring up the source-to-staging boilerplate. Translating a SQL query from one dialect to another. Writing the YAML config nobody wants to write. All of it — agent-competent, fast, cheap. Not perfect, but good enough to pass code review if nobody's paying close attention. And let's be honest: on most teams, nobody was.

Tristan Handy at dbt Labs basically said it: dbt Copilot is already "very good" at all authoring tasks — building models, writing documentation, tests, defining metrics. And he's underselling it, because it gets better every quarter.

If that's 80% of what your "data engineer" does in a week, you don't have a data engineer. You have a pipeline technician wearing a senior-looking title. And yeah — that person is replaceable. Not because they're bad at their job. Because the job was always automatable. We just didn't have the automation yet.

Here's the work that's safe, and the reason is structural, not temporary.