David, this question comes from a place of curiosity and learning, and I do not mean it to sound like a jerk. If AI can do a lot of this processing on its own, what advantage does going with your own python workflows gives us? I know one argument is over reliance on AI and not understanding what it is doing is dangerous.
Are there other reasons where this process would be advantageous over an AI workflow handling the tokenization and analysis? Compute cost?? Speed?
This is really awesome stuff. My dilemma is allocation of my time when there's so much going on right now. Thanks again, and I hope it was okay to ask this.
This is a legitimate question, and I welcome it. At base, there are 3 primary reasons why learning the fundamentals is still the best way forward in 2026:
Reproducibility - AI tools built on LLMs can and will produce different outputs from the same data inputs. There’s a certain level of randomness built in.
In my experience, most business stakeholders really don’t like the idea of basing business decisions on AI outputs that are not reproducible. It just makes them nervous.
By learning how to analyze data using Python, you get past this by partnering with the AI to generate the reproducible Python code.
Accountability - If you’re using AI in analytics, you’re probably trying to leverage data in some way to drive business decisions in some shape or form.
Vibing your analytics is problematic because you don’t understand what the AI is producing (that is the original definition of “vibing”). So, what happens when the AI makes incorrect assumptions or hallucinates?
Good luck with the argument, “It’s not my fault! Blame the AI!”
By learning how to analyze data using Python, you can partner with the AI because you understand what’s going on.
Cost - Even if you ignore the very likely scenario that AI costs from leading vendors like Anthropic will go up substantially, knowing Python workflows will save on costs.
For example, partnering with AI to generate reproducible Python code. Re-running the generated code is essentially free compared to interacting with AI.
David, this question comes from a place of curiosity and learning, and I do not mean it to sound like a jerk. If AI can do a lot of this processing on its own, what advantage does going with your own python workflows gives us? I know one argument is over reliance on AI and not understanding what it is doing is dangerous.
Are there other reasons where this process would be advantageous over an AI workflow handling the tokenization and analysis? Compute cost?? Speed?
This is really awesome stuff. My dilemma is allocation of my time when there's so much going on right now. Thanks again, and I hope it was okay to ask this.
This is a legitimate question, and I welcome it. At base, there are 3 primary reasons why learning the fundamentals is still the best way forward in 2026:
Reproducibility - AI tools built on LLMs can and will produce different outputs from the same data inputs. There’s a certain level of randomness built in.
In my experience, most business stakeholders really don’t like the idea of basing business decisions on AI outputs that are not reproducible. It just makes them nervous.
By learning how to analyze data using Python, you get past this by partnering with the AI to generate the reproducible Python code.
Accountability - If you’re using AI in analytics, you’re probably trying to leverage data in some way to drive business decisions in some shape or form.
Vibing your analytics is problematic because you don’t understand what the AI is producing (that is the original definition of “vibing”). So, what happens when the AI makes incorrect assumptions or hallucinates?
Good luck with the argument, “It’s not my fault! Blame the AI!”
By learning how to analyze data using Python, you can partner with the AI because you understand what’s going on.
Cost - Even if you ignore the very likely scenario that AI costs from leading vendors like Anthropic will go up substantially, knowing Python workflows will save on costs.
For example, partnering with AI to generate reproducible Python code. Re-running the generated code is essentially free compared to interacting with AI.
Thank you. This was a much more satisfying answer than what perplexity gave me.
I use a Chromebook, but I believe this tutorial can be done with Google sheets with some modifications, at least according to AI lol.
Thanks again. I have no counter.