character: “Isaac“. Isaac lives in the early 80s. Isaac is an excellent architect, but he is a little set in his own ways. For example, he always says:
“I don’t trust this calculator thing; I will still do all the computation by hand.”
As a consequence, all the other architects in the firm deliver way more than he does because they use this innovative tool that he doesn’t trust and refuses to adopt.
Keep in mind that Isaac is still essential to the architecture firm he works for: he can never be replaced by a calculator. However, it is a shame that Isaac’s talent is not fully used to develop new projects, unleash his creativity, and showcase his amazing work, simply because he refuses to adopt a productivity tool.
The goal of this post is for you not to be that Isaac.
In this blog post, I will cover all the lessons that I have learned in the last 7 years of using, researching, developing, and adopting AI for both productivity and real-world applications of algorithms in research and industry within the Data Science world.
This will not be a technical guide on Claude Code, OpenClaw, Codex, or Cursor. This will be a set of guidelines that have helped me stay professionally relevant throughout my career and that I use on a daily basis to increase my productivity.
This is designed especially for students, early- to mid-stage Data Scientists, or even advanced data professionals who do not use AI at all and would like to know more about it.
If this is you, welcome on board. Let’s dive in!
Step 0: Breathe!
The people who are most overwhelmed by AI are the ones who work with it. If you feel that you are behind, and you don’t know enough, and everyone else is doing better than you, I want to give you this one piece of advice: take a deep breath.
AI is changing and evolving at a speed where knowing every new LLM out there and every possible vibe-coded, agentic “revolutionary” tool being published on LinkedIn is not only impossible but also unnecessary and ultimately meaningless for your career.
You do need to be up to speed and aware of your surroundings, but adapting to AI should still be enjoyable and not overwhelming. And the reason why I advise you not to panic is this: your analytical skills, critical thinking, algebra and statistics knowledge, and creativity are your strength, not your AI setup.
And this leads me to Step 1.
Step 1: What are we dealing with?
Let’s start with this:
Large Language Models are productivity tools.
That’s all they are. They allow you to ship things fast; they optimize your time by reducing the amount of coding; they automate your recurrent flows and loops.
What they are not is functioning Data Scientists who completely understand the logic of your business, who have all the pieces and information to navigate open-ended problems without definite and perfect solutions, who can very well weigh in on the sentiments of a certain product, and who can read between the lines of an ask to figure out the best solution end-to-end without constant human attention.
The reason why they are not capable of that is that LLMs are not creative; they sound extremely artificial, and they usually take silly routes unless you specifically guide them along the way.
So now that we know what they are, the next question is, “What can we do with them?”
Step 2: Don’t delegate the thinking!
Technically, we know that LLMs are amazing information retrievers but bad extrapolators. This means, for example, that an LLM can find information very quickly (and often correctly), but it won’t do a great job in writing a poem that doesn’t sound like it is a (mediocre at best) reshuffling of old poems.
And this does not involve “art” or “philosophy” problems only: LLMs don’t do well with problems that are not well defined. One viral trend showed that LLMs suggest you “walk” to the car wash if it is close to your house, not to waste gas [here]. This is just one of the many hallucinations that show that LLMs are not able to deal with somewhat open problems.
In other words, we must use LLMs to tackle very straightforward and specific asks. This is how they are trained and evaluated.
However, real data science problems do require some sort of creativity. You need to answer questions like: What is the angle that we are looking for? Who are we showing this to? What does our company need, and what would be beneficial to develop in this specific period of time? Can this solution integrate well with another problem that we are also dealing with? If we talk with Coworker X, they might have an interesting take; should we send them a message?
These questions belong to humans. LLMs won’t do a great job in considering them, fully understanding them, and implementing a solution. Sure, you can use them to get a hint and see if there is any obvious solution to your problem, but ultimately you can’t delegate the whole thinking process and go get a coffee.
My recipe is thus the following:
- Use LLMs to extract information and evaluate suggestions
- Use your brain to be creative and make sure all the requirements are met.
This will ultimately make sure that your personality and sharpness will emerge and that people will remember you for the way you work.
Step 3: Build your own system
Not everyone uses AI the same way, but every user has their own preferences. The new advance from simple ChatGPT to systems like Claude Code, Codex, or Cursor is the ability to build your own AI routines.
Through tools called “skills/commands,” “MCPs”, “loops”, and much more, you will have a real boost in productivity and also sound somewhat “unique”.
For example, in Claude Code, you can create skills that:
1. Will code like you, so everyone associates you with your code.
2. Will speak like you, so you don’t sound like an em-dashes-fanatic robot.
3. Will review things according to your metrics and tests, so that you make sure that your work is always in line with your desired quality.
These are just examples: the amazing part is that you can be creative and adaptive. Once you find something you keep doing over and over again, you can make it a routine in your favorite agentic tool, and your productivity will boost (the right way!).
However, I would still check that the code is really my coding style, that the text is really well written (Claude Opus 5 sounds awfully robotic), and really be careful with the way your routines get “interpreted” sometimes.
This leads us to the next point.
Step 4: Be Diligent
These new LLMs are not just better in benchmarks; they are better at gaslighting. If something is well written, it sounds convincing, “official”, and ultimately true. This might not be the case, though, and you won’t know unless you actually verify the output.
When you develop a design, you need to question it, bulletproof it, test it, understand the limitations, predict the possible questions that a PM or manager might ask you, and prepare your answers.
At the end of the day, you are responsible for your code. When something won’t work or when your pipeline will fail, people will come to you, and they will expect prompt answers, professionalism, and clarity.
The more of a beginner you are in your career, the more you need to spend time to make sure you understand your project thoroughly. Every piece of the pipeline needs to be validated, and the whole approach needs to be evaluated. At the end of the evaluation, you need to be 100% sure that there is no way to do this better than what you did (even if Claude tells you otherwise).
In short…
The most important thing I want to convey with this article is the following:
Don’t forget that LLMs are productivity tools.
But just like some great boxing gloves allow you to train better, they don’t make you a good boxer.
Don’t panic; take your time to do your actual homework, use AI to get rid of boring and repetitive tasks, and don’t blindly trust a robot to be the rockstar that you are 🙂
Before you head out!
Thank you again for your time. It means a lot. My name is Piero Paialunga, and I’m this guy here:
I’m originally from Italy, hold a Ph.D. from the University of Cincinnati, and work as a Data Scientist at The Trade Desk in New York City. I write about AI, Machine Learning, and the evolving role of data scientists both here on TDS and on LinkedIn. If you liked the article and want to know more about machine learning and follow my studies, you can:
A. Follow me on Linkedin, where I publish all my stories
B. Follow me on GitHub, where you can see all my code
C. For questions, you can send me an email at piero.paialunga@hotmail