SQL: The Skill That Never Expires

Every few years, something new arrives that promises to change data forever.

Hadoop. Spark. NoSQL. The modern data stack. Now AI.

Each wave brings new tools, new job titles, and a fresh round of blog posts telling you what to learn next. It is exhausting. And most of it is noise.

Here is what does not change: SQL.

SQL has been the language of data since the 1970s. Every major platform built in the last decade -- Snowflake, BigQuery, Databricks, Redshift, DuckDB -- runs on SQL. dbt, the tool that reshaped how analytics engineers work, is just SQL with version control. Looker's LookML compiles to SQL. Even the AI tools that "generate insights" are writing SQL under the hood.

The stack changes. The foundation does not.

Why people skip it

The modern data ecosystem makes it tempting to reach for abstractions early. Drag-and-drop BI tools. Low-code pipelines. AI assistants that write queries for you. These are genuinely useful. But they create a dangerous gap: you can produce output without understanding what is happening underneath.

I have worked with analysts who could build a Power BI dashboard but could not explain why their numbers were wrong. They could not debug a join. They did not know what a window function was doing. The tool worked. The number was wrong. They had no way to tell the difference.

That gap is a career ceiling.

What SQL mastery actually looks like

It is not about memorising syntax. It is about understanding how data moves.

You need to know how joins work -- not just that LEFT JOIN keeps all rows from the left table, but why that matters when your grain is different across tables. You need to understand aggregation, and what happens when you GROUP BY the wrong thing. You need window functions: ranking, running totals, lag and lead. You need to read a query plan and have some intuition for why it is slow.

Most importantly, you need to be able to look at a result and ask: does this make sense? That question is not answerable without understanding what the query actually did.

AI makes this easier, not unnecessary

This is where I push back on the pessimism.

AI tools -- Claude, ChatGPT, Copilot -- are genuinely good at writing SQL. Give them a clear problem and they will produce something reasonable, often faster than you could write it from scratch. I use them every day.

But here is the thing: AI does not know your data. It does not know that your orders table has one row per line item, not one row per order. It does not know that your date field includes nulls for cancelled transactions. It does not know that the metric your stakeholder calls "revenue" is actually net of refunds and your finance team defines it differently.

You have to know those things. And when the AI writes a query that looks right but produces the wrong number, you have to be able to catch it.

AI raises the floor. It does not replace the ceiling.

The people who get the most out of AI-assisted SQL are the ones who already know SQL well. They can prompt clearly, review output critically, and fix what is wrong. The people who struggle are the ones who cannot tell whether the output is correct.

The practical path

If you are starting out in data, here is what I would focus on:

Start with the basics until they are automatic. SELECT, WHERE, GROUP BY, JOIN. Do not move on until you can write these without thinking.

Then learn window functions. ROW_NUMBER, RANK, SUM OVER, LAG. These are where analysts separate from people who just pull data.

Then learn to think about grain. Before you write any query, ask: what is one row in this table? What is one row in my result? If those are different, what does that mean for my aggregation?

Then learn to read execution plans. Even a basic understanding of why a query is slow will save you hours.

After that, the specific platform -- Snowflake, BigQuery, whatever your company uses -- is just dialect. The concepts transfer.

The career argument

Data tools have a lifecycle of maybe five years before something better arrives. SQL has survived fifty years of something better arriving.

Every data role I have hired for or worked alongside -- analyst, analytics engineer, data scientist, analytics lead -- required SQL. Not as a nice-to-have. As the baseline.

The people who build strong SQL fundamentals early in their careers have options. They can move between platforms, between industries, between roles. The people who skipped the fundamentals and went straight to the tools find that when the tools change, they have to start over.

You do not have to choose between fundamentals and modern tools. The point is that one without the other is fragile.

Master the SQL. Use the AI. Know the difference between the two.