The short version: AI will not replace data engineers. But engineers who learn to use AI will replace engineers who don’t.
Below, we unpack the full conversation, including the semantic layer debate, the death of the dashboard, and the one skill that still matters most in the age of AI agents.
Watch the full episode here: https://youtu.be/2FQkyF89HVk
AI won’t replace data engineers. Engineers using AI will.
Julian opens with a line that should be printed on every engineering manager’s wall. AI on its own is not the threat. The real shift is happening between engineers who treat AI as a collaborator and engineers who treat it as a competitor.
This framing matters because it changes the question every data team should be asking. It is no longer “is AI coming for my job?” It is “am I one of the engineers who will use AI to do the work that used to take a whole team?”
For Julian, this is non negotiable inside his own team. Learn the tools. Adapt. Embrace them. Take the time you need, but the destination is not optional.
Why some engineers quietly resist AI
Not every engineer is racing to adopt AI. Some are dragging their feet, and Julian has a sharp theory about why.
It is not about capability. It is about identity.
Engineers take pride in their code. They want to maintain it. They want to defend it in code review. They want it to be appreciated as craft. AI threatens that relationship by producing similar output without the human touch.
The resistance, in other words, looks like a tooling problem. In reality, it is an identity problem. Once you see it that way, the path to adoption stops being a training exercise and starts being a conversation about what good engineering work actually means in 2026.
This is exactly why the question of whether AI will replace data engineers misses the real story. The engineers most at risk are the ones who keep asking the question instead of trying the tools.
Does AI actually need a semantic layer?
This is the question that splits the room at every modern data conference. Julian gives a more useful answer than yes or no.
Today, yes. AI agents benefit from the structure, the definitions, and the shared meaning that a semantic layer provides. He references a recent dbt study where AI accuracy sat in the high 90 percent range with a semantic layer, and dropped to around 70 percent without one.
In two years, perhaps not. As compute gets cheaper, context windows grow, and models get sharper, the gap closes. The same way we once threw more infrastructure at problems we could not solve in code, we may soon throw more compute at problems the semantic layer used to solve.
The lesson? Build for today and design for tomorrow. The semantic layer is your safety net right now. Don’t assume it is permanent furniture.
Local AI models are closer than you think
Julian is bullish on local AI models running on everyday hardware. He has been testing open source Chinese models on his own laptop, and while they trail the leading APIs, they are catching up faster than most people realise.
His mental model is simple. What today’s APIs can do is what your laptop will do in two years. For data teams in regulated industries, this matters. Privacy, compliance, and cost will all push more workloads to local inference.
That changes the calculus on every architecture decision being made today. Build with portability in mind.
Where AI is actually working in data engineering today
Theory aside, where is AI delivering real value inside a data team right now? Julian points to three places, all of which engineers have historically avoided.
Tech debt. AI is finally making it possible to address the backlog every team knows about but never has time to fix. The business case for refactoring used to be hard to make. With AI accelerating the work, the calculus changes.
Writing tests. Test driven development is the practice every engineer says they believe in, and few actually follow. Writing tests is slow, repetitive, and feels lower status than shipping features. AI agents excel at crawling every edge case and writing the coverage your team always meant to write.
Data governance. Documentation, column lineage, permission audits. Boring, important, and almost never done well. AI can take you 70 percent of the way there with minimal effort, leaving humans to handle the judgment calls.
If your team is looking for a wedge to start using AI, start here. The work is low risk and the productivity gains are immediate.
What the data industry is getting completely wrong
Julian saves his sharpest critique for the rise of low code tools.
The pitch is compelling. Lower the barrier to entry. Let anyone build a data pipeline. Democratise the work. The reality, in his experience, is uglier.
Critical pipelines are being built without version control. SQL is being shipped without unit tests. Personal credentials and raw keys are sitting inside scripts in production. None of this would be acceptable in software engineering or civil engineering. Yet in data, somehow it is.
His call to action is direct. Bring engineering practices back into data work. The cost of ignoring this is bigger than most teams realise, and it compounds quietly until something breaks.
For teams already working code first with version control and tests, this section will feel like validation. For everyone else, it should feel like a warning.
The death of the dashboard
The most provocative moment in the episode comes when Julian is asked what belief he held three years ago that he no longer holds today.
His answer? Dashboards.
For two decades, the dashboard has been the bread and butter of analytics. Beautiful charts. Filters. Drill downs. The promise that an executive could glance at a screen and make a better decision.
Julian’s view now is that most dashboards are on the way out. AI agents will not just show what happened. They will explain why it happened. They will look at granular data, cross reference other sources, and recommend specific actions.
The dashboard, in this future, becomes a notification. The agent says, “this dipped, this is why, here is what I suggest, want me to handle it?” That is a fundamentally different product than a chart on a page.
If you build dashboards for a living, this is worth sitting with.
The one skill that still matters most
If AI can write the code, test the code, and ship the code, what does the engineer actually do?
Julian’s answer is context.
The engineer carries context from meetings the AI was not in. The engineer reads the market in ways no dataset captures. The engineer understands the politics, the priorities, the constraints, and the personalities. That is the leverage AI cannot replicate.
This reframes the data engineer of the future. Less typing. More thinking. Less code production. More judgment about what should be built in the first place.
Will AI replace data engineers? The honest answer.
So, will AI replace data engineers?
No. Not in the sense most people fear. The role is changing, the workflow is changing, the deliverables are changing. But the human in the loop, the one with context, judgment, and business sense, remains central.
What is being replaced is a specific way of working. The lone engineer typing every line of code. The dashboard as the primary output. The low code shortcut as a sustainable architecture. The semantic layer as a permanent fixture rather than a stepping stone.
If you are a data engineer reading this, the message is the same one Julian gave his team. Learn the tools. Adapt. Embrace them. You will not be replaced by AI. You may be replaced by an engineer who uses it better than you do.
Watch the full episode
The full conversation goes deeper on every one of these themes, plus a spicy closing question Julian leaves for the next guest about cloud cost reduction.
Watch the full episode on YouTube: https://youtu.be/2FQkyF89HVk
Subscribe to the Meltano podcast for new episodes on data engineering, AI, and the future of the modern data stack.
Frequently asked questions
Will AI replace data engineers?
No, AI is unlikely to replace data engineers outright. Engineers who learn to direct AI agents will replace engineers who do not. Context, judgment, and business understanding remain the core leverage of the role.
Does AI need a semantic layer?
Today, yes. A dbt study referenced in the episode shows AI accuracy in the high 90s with a semantic layer compared to around 70 percent without one. In two years, as compute and context windows scale, the requirement may soften.
Are dashboards becoming obsolete?
Most static dashboards are. AI agents will increasingly move from showing what happened to explaining why and recommending what to do next. The dashboard becomes a notification, not a destination.
What is the biggest mistake the data industry is making?
Adopting low code tools at the cost of engineering practices. Critical pipelines are being built without version control, without tests, and with credentials baked into scripts. The fix is to bring software engineering discipline back into data work.
What skill matters most for data engineers in the age of AI?
Context. The engineer’s biggest leverage is the business knowledge, market understanding, and meeting context that no AI agent has. Code production is being automated. Judgment is not.