Quick heads up: stay tuned until the end for an exciting product announcement, and an opportunity to be a free lifetime user!
There is a growing belief that as AI tools improve, learning SQL and Python matters less. Our experience suggests the opposite. As code generation becomes easier, the real advantage shifts to those who can read what the code is doing, validate that it answers the right question, and direct the tool toward a safer and more useful result. In other words, the work moves from memorizing syntax to understanding intent and spotting risk.
AI can now draft a convincing query or script in seconds, yet seemingly small mistakes still bend decisions in the wrong direction. Join keys are selected on convenience rather than logic, grains are mixed across subqueries, filters live in the wrong place and mask important segments, and null handling is assumed rather than verified. As a result, the user winds up with a block of code that is broken for a dozen small reasons they can’t quite untangle, or an output that is off just enough to be unusable. The solution is not to avoid these tools entirely, but rather to ensure you don’t lose what got you here in the first place - your data literacy!
What data literacy looks like in an AI world
In practice, there are three elements we emphasize to our team when interacting with AI generated results:
- Reading: Given a short query, you should be able to say what it returns, describe the grain, point out the join choices, flag which filters apply, note how nulls and window functions behave, and predict the rough shape of the result
- Validation: You should have enough knowledge to spot obvious errors or questionable assumptions. Does the date filter make sense? Is the query actually answering the question you asked, or something subtly different? Are there edge cases that aren’t being handled?
- Direction: Perhaps most importantly, you need to know enough to guide the AI toward better solutions. When the first attempt isn’t quite right, can you articulate what needs to change? Can you break down a complex problem into smaller pieces that are easier to solve?
None of this requires encyclopedic recall of functions, but it does require a clear mental model of relationships, ordering, and tradeoffs.
A short example makes the point. We want the sum of full order revenue per customer for orders that contain at least one item in the widget category. An AI draft often joins core.orders to core.order_items and filters in the WHERE clause, which quietly counts the same order multiple times when it has more than one item within that widget item category.
AI draft (overcounts when an order has multiple widget items):
SELECT o.customer_id,
SUM(o.order_total) AS revenue
FROM core.orders o
LEFT JOIN core.order_items oi
ON oi.order_id = o.order_id
WHERE oi.item_category = 'widget'
GROUP BY o.customer_id;
Correct intent (one CTE to isolate qualifying orders, then join once):
WITH orders_with_widget AS (
SELECT DISTINCT order_id
FROM core.order_items
WHERE item_category = 'widget'
)
SELECT
o.customer_id,
SUM(o.order_total) AS revenue
FROM core.orders o
LEFT JOIN orders_with_widget w
ON w.order_id = o.order_id
WHERE w.order_id IS NOT NULL
GROUP BY o.customer_id;
By reading the query to understand what it says, validating the results to ensure they make sense, and providing clear direction you can ensure you get the results you’re looking for - not something close but not quite there.
Building organizations that move faster with fewer handoffs
When non-data teams develop a modest level of literacy, the entire decision cycle improves. Product managers and operators ask answerable questions, finance and marketing share a vocabulary with analytics, and the data team spends less time triaging tickets and more time solving harder problems. The goal is not to turn everyone into an analyst. The goal is to reduce friction so that simple questions are answered quickly and complex work receives the attention it deserves.
A quick announcement: sign up for our new daily SQL game!
A sneak preview of what our new SQL learning platform is going to look like
Inspired by our belief in the value of accessible data literacy, our team has been building a new tool: a reading-first SQL learning platform designed for busy operators, product managers, finance leads, and analysts who want practical capability without a bootcamp. The core idea is simple: daily, lightweight reps that train people to see what a query is doing, explain why it works, and spot the common failure modes that show up in real projects.
We are starting with SQL because it remains the common language for data work, but the vision is broader. We plan for the platform to expand into python, data modeling patterns, and visualization principles, always with the same approach: short, concrete reps that build competence. The outcome we care about is simple to describe and powerful in effect: increased data literacy, better AI responses, and more accurate reports!
If you’re interested in giving the product a try, we are opening a small beta. The first 100 individuals to sign up will receive free lifetime access to the tool regardless of any future pricing increases.
If this sounds compelling, drop a comment below or send a note to info@southshore.LLC with “SQL Tool beta” and a brief line about your role, and we will follow up as we expand the cohort!
At South Shore Analytics, we’re building tools to make data literacy accessible and practical for modern organizations. If you’re interested in learning more about our SQL challenge platform or our approach to data stack modernization, please reach out. We’d love to chat about how to make your organization more data-literate.