Hello Reader,

In 2016, I graduated with my master's degree. At the time, I only knew a bit of coding in R from working on my thesis. Yet even that modest skill was my major selling point I'd make sure to mention in every job interview!

Fast forward to 2026: LLMs can now generate working Python code in seconds. It makes that situation from 10 years ago feel almost ancient. Honestly, it's a little embarassing how I felt like GOD knowing just some basic programming back then 🤣.

AI is reshaping many jobs. Data scientists, data analysts, and other tech professionals are no exception.

I recently sat down with my friend Lan Chu, a data scientist/ AI tech lead at a major bank in the Netherlands, to talk about exactly that (full interview below 👇).

Lan has a strong background in data science and deep expertise in natural language processing. Today, she focuses on building AI systems, specializing in the design and implementation of data and AI pipelines, as well as responsible AI practices. I'd highly recommend you check out her awesome work and connect with her on LinkedIn and Medium.

TL;DR​
Is learning to code still matters in 2026? Her take: yes—more than ever. AI can generate code fast, but it also creates context bugs, rework, and hidden tech debt if you don’t understand the fundamentals. Coding skills are essential for debugging, guiding AI, and building reliable systems. That said, “just being good at coding” isn’t enough anymore. The real career moat is being able to design, ship, and own end-to-end AI products, with strong fundamentals, solid engineering habits, and systems thinking.

---

Thu: So… tell me a bit more about what you do at work these days?

Lan: In the past years, I've been helping various teams at Rabobank with automating and accelerating their financial research, audits and reporting. These teams spend thousands of hours a year reading and searching many financial documents, but by building high-quality search tools and AI chatbots I managed to reduce that time to mere minutes.

I mostly code with Python (like many other data scientists) and occasionally SQL, R and Typescript if needed.

---

Thu: How much has AI changed the way you work, day to day?

Lan: A lot! Honestly, I can’t imagine my life without AI coding assistants anymore 😅. I feel like coding agents are still a bit underrated. But my relationship with AI depends heavily on context, whether I'm starting a new project from scratch or adding features to an existing codebase.

In my experience, AI can generate 80% of new features in minutes, but the remaining 20% often requires hours of refinement, debugging, and cleaning up duplicated code or removing bad designs to get it right.

When I know the codebase well, AI is a genuine net positive. I can guide the design, understand every line of code, and debug when needed.

But for completely new projects, I still do the “old” way: experimenting first in notebooks. Once I have a working prototype, I try to structure the codebase with the help of AI.

Most AI bugs I encountered are not logic bugs, they are context bugs. It will often miss the existing methods and architecture to be able to use existing components.

I once asked an AI to add user feedback storage. The database connection already existed, but the AI completely missed it and reimplemented everything from scratch 🤦🏻‍♀️.

---

Thu: So far, which AI coding tools are your favourite?

Lan: I've been using different AI coding assistants and agents. Early on I was using ChatGPT and Claude through its web UI. Then, I tried Cursor and stuck with it for a while.

At work, I only have access to Github Copilot but I am not a fan!

Last year, I made a switch to Claude Code I have tried out quite a few, Github copilot, Codium, Cursor and Claude Code. Claude Code it is the clear winner for me! I've had genuine "wow" moments using it, something I didn't experience with previous tools.

Not because it writes better code, but because its agent is so good at understanding and navigating the codebase and taking actions.

I love the terminal design, its ability to edit files, run commands, and make commits. Claude’s agents are so good with reasoning and calling tools. It pulls in the right files from my codebase when needed.

I've had issues with Cursor missing important pieces, but Claude Code actually understands which parts are relevant. You can also turn on "Plan" mode before implementation and use CLAUDE.md as a system prompt for your project.

For me, it is number one!

---

Thu: Nowadays, anyone can vibe code a working application without knowing how to code. If someone asked you, “Hey, should I still learn to code today?” what would you tell them?

Lan: Yes, absolutely - if you are serious about a career in Data and AI.

Not learning to code because of AI is like saying we don't need to learn arithmetic because we have calculators 🤯.

What's changed is accessibility. Previously, it took years to become proficient at coding. But with AI, coding becomes much more accessible, which means more people should learn how to code, not fewer.

A personal example: I recently worked on a project using TypeScript. Existing templates and AI helped me get the app running quickly, but as soon as I added more features, things started breaking and I couldn't debug. After hours of frustrating back-and-forth with AI, I finally stopped and took a proper TypeScript course.

Everything changed. Once I understood the syntax and fundamentals, debugging became easier. I was no longer guessing.

With AI coding assistants, what I fear most is losing debugging skills - the exact skills you need to catch AI mistakes. If you want to debug effectively, you must understand the code. There's no shortcut!

Vibe-coding without fundamentals is just guessing.

If you're starting out, I'd suggest:

• Learn Python - take a proper course
• Master the basics: data types, data structures, control flow, functions
• Use formatters, linters, and pre-commit hooks early on
• Build real projects and stick to those principles

These habits compound fast!

But if you ask me whether “being great at coding” alone is a long-term career moat? I doubt it.

I think the industry is expecting people to be full-package these days, being able to design and build real products end-to-end, NOT just being good at coding.

So unless you love coding and are ready to compete in a talent war in Silicon Valley, I'd invest in skills like system design, MLOps, and cloud engineering.

---

Thu: Now that we are generating code with AI, how do we avoid creating technical debt and maintain engineering standards?

Lan: People can ship faster than ever, and unfortunately, bugs are appearing more often. AI has lowered the barrier to writing code, but it hasn't lowered the bar for building reliable software.

I recently read that even in big tech, bugs that used to appear once a year now appear weekly!

Before we had code-assistants, a large part of implementing a feature was actually not writing the code itself, but understanding the requirements and context.

In data science this time was spent on data exploration, prototyping and creating a good understanding of the problem that you are trying to solve.

Now, even with a minimal description for a feature any coding assistant can write something decent. But the AI is just guessing what you want, and if you haven’t done much thinking yourself, you also won’t be able to tell if the AI did a good job.

If you want to ship AI-generated code, my advice is:

• Don't skip fundamentals: Still write requirements, review code, and create unit tests yourself. When reviewing AI code, focus on whether requirements are met and edge cases are covered.
• Plan and design documents first, code later: Planning beats prompting. Have AI create design docs in a /designs folder for each feature. Next time an AI needs to modify the code, it can look up the original design and understand why the code was written that way.
• Be explicit about what to do and NOT to do, and explain why: Use commit hooks and enforce standards.
• Establish a culture of code ownership and code review: Whether AI or a human wrote the code, if it's in your codebase, you and your team own it.
• If possible, join a team where engineering standards are already established and there's no way to bypass them. That's the fastest way for good practices to become habits.

I believe today's AI tools are just the warm-up. We may be nowhere near their full potential.

While we await the next breakthroughs, what I can say is this: AI is powerful, but it's not yet a substitute for genuine expertise.

So, do learn to code!

---

I hope you enjoyed this post and found it useful!

Have a great week! 😊

Thu

---

P.S: Work with me:

If you want a comprehensive course from Python fundamentals to building AI applications, check out my Python for AI Projects course. It’s packed with everything you need to build solid fundamentals and transform your skills in 2026.

​Learn more about the ​course.​