The Future of Data Science in an AI World

There is a question I’ve been thinking about for the last six months. I’ve heard versions of it from Data Science leaders at every company.

If a non-technical PM can ask Claude or ChatGPT to write SQL, build a chart, run a regression, and write up the findings - what’s left for the data scientist?

Some have responded by quietly downsizing analytics teams. Some are pretending nothing has changed. Most are confused.

Here’s where I’ve landed.

AI doesn’t eliminate data science. It changes the process

Take a few steps back and look at what data science actually does.

I remember when the data scientist title first emerged at LinkedIn and Facebook in the late 2000s. The role didn’t exist before cloud and large-scale technology companies created enough data to make it valuable. Before that, it was data analysts and business intelligence. The shift wasn’t that DA and BI disappeared - it was that a new tier emerged on top, one that could turn massive data into product, optimisation, and direct business outcomes.

We’re at another inflection point now.

For years at Airwallex, we organised our data science org into two job families: analytics and algorithm.

DS analytics improve decision quality and reduce decision latency for product, ops, and GTM teams. They take messy business questions, structure them, find the answer, and help the right people act on it faster.

DS algorithm drive direct optimisation. Their models for risk, fraud, ranking, and lifetime value change product behavior, pricing, allocation, risk decisions - in real time, at scale.

It was a clean split that worked. Then AI happened.

Neither of those jobs goes away in an AI-native world. Both get harder, broader, and more valuable - if the data scientist evolves with them.

What AI actually shifts

Ad-hoc analysis is now accessible to anyone, not just the coders.

That’s a real shift. But it’s also where most leaders get the wrong takeaway.

Easy access to analysis doesn’t translate into higher decision quality. It often does the opposite. More charts, more interpretations, more confidently-wrong narratives. Without governance, without context, without the discipline that data scientists bring, you get a lot of noise.

The volume of analysis goes up. The signal-to-noise ratio goes down.

The remaining moat is scale, science, and rigor

What differentiates data scientists from a PM running ad-hoc queries is the same thing that always did: scale, scientific method, and rigor.

A PM can answer one question. A data scientist builds a system that answers a thousand questions correctly, repeatedly, under different conditions, with proper context, evaluation and governance.

A PM can run a regression. A data scientist understands when the regression is wrong, what’s confounded, what’s missing, and what experiment would actually answer the question. They know when to trust the data, when to challenge the framing, and when to design a study that isolates causality.

That’s the science part. Not just statistics - the discipline of forming a hypothesis, designing the test, controlling for what could mislead you, and being honest about what the result actually means. AI doesn’t replace that discipline. It amplifies it for the people who have it, and exposes the people who don’t.

The teams that invest in rigor, science, and scalable systems compound. The teams that don’t get drowned in their own ad-hoc output.

Where DS is heading: systems, not individual analyses or individual models

Here’s the shift I’m making at Airwallex.

The data science role of the next decade is not running analyses or training one-off models. It’s building the systems that produce reliable, contextualised, governed decisions and actions at scale.

Two flavors:

DS Analytics becomes systems of decision.

Are we building reliable AI systems that help the CEO, CTO, CRO, and other decision makers make better and faster calls? That means owning the AI analytics harness - the semantic layer, the eval infrastructure, the context retrieval, the validation gates. The data scientist is no longer the person running the analysis. They’re the person making sure the system that runs the analysis is trustworthy.

DS Algorithm becomes systems of action.

ML models already do this for risk, ranking, pricing. AI extends what’s possible - more autonomous optimisation, more end-to-end workflows, more agents driving real outcomes. The algorithm DS evolves from training a model to designing the agent’s reasoning, tool use, and guardrails.

In both cases, the work moves up the stack. From analysis to system. From model to harness. From “I answered the question” to “I built the thing that answers the question correctly forever.”

DS and DE are blending

The other shift: data scientists and data engineers are converging.

Data scientists need to excel more on production data pipelines - versioning their work, writing testable code, building evaluation pipelines, owning governed artifacts. The model or analysis is no longer the deliverable. The pipeline that runs it reliably is.

Data engineers need to move up the stack into the application layer - reasoning about business semantics, decision quality, what the data is actually being used for. Building pipes that move data is no longer enough. The work is increasingly about building the systems that consume the data.

Both are now systems builders. Any data talent worth hiring needs to be able to build scalable, scientific systems that drive business value. That’s the job.

The org structure will follow. We’re already restructuring around capability rather than legacy job family - applied AI scientists, agent system builders, members of technical staff. Different names. Same underlying truth: the work has changed.

The bottom line

AI didn’t kill data science. It killed the version of data science that was just running queries and writing decks.

Great data scientists were never about that anyway. They were about science and rigor - building systems that compound decision quality over time. AI gives them better tools to do exactly that, on a much bigger surface area.

The question isn’t whether your data science org survives the AI era. It’s whether your data science org can evolve fast enough to lead it.