Controlled Demolition
On tearing down a professional identity built on certainty, one syntax error at a time
Most career advice for senior leaders is draped in comfortable lies. We are told that our soft skills and domain expertise are our ultimate shields.
I believed that, too — right up until I sat down at 9:00 AM facing a blank VS Code editor and realised that thirty years of strategic wins felt remarkably thin.
This is an adapted excerpt from a longer piece on the Data Consigliere site. Read the full version here.
In three weeks, I start an intensive Data Science bootcamp at Le Wagon. A deliberate move. The kind that looks like ambition from the outside and feels, from the inside, more like controlled demolition — tearing down a professional identity built on certainty so you can rebuild it on something more durable.
But the dismantling started earlier than I expected.
The Fog Doesn’t Care About Your CV
For three decades, my value has been tied to certainty. The actual job — not the title — was to walk into rooms full of ambiguity and make the fog legible for everyone else.
Then I encountered TypeError: 'NoneType' object is not subscriptable, and suddenly, the fog was mine alone.
This is what I’ve started calling the Expert’s Paradox: the very thing that makes you effective in a leadership role — the reflex to project competence — becomes the biggest obstacle when you sit down to learn something genuinely new.
A junior developer expects to fail. For someone who has spent decades being the person who clarifies things for others, failing at a basic loop doesn’t just sting. It feels, in some quiet, embarrassing way, like identity collapse.
I spent time fighting that feeling instead of writing code. I’d read the error message, feel a flicker of something close to shame, and go do something productive instead — answer emails, review a deck, anything. The syntax wasn’t the problem. The ego was the problem.
The shift I eventually had to make was simple but uncomfortable: I had to give myself permission to be bad at this, on purpose, for a while. Not as a strategy. Just as a fact.
Systems, Not Scripts
Once I stopped performing competence and started actually learning, I noticed the gap between what the internet teaches and what I actually needed.
The internet is full of what I’ve started calling Survival Python — the “Hello World” scripts, the basic loops, the snippets that live and die in a single Jupyter Notebook cell. They give you a quick dopamine hit: the code ran, the chart appeared. But in any professional context, “it works on my machine” is a liability, not an achievement. I’ve spent years telling other people that.
So I made an early decision to approach this differently. Borrowing project templates that impose structure from the start. Using modern tooling like uv to handle environments cleanly. Asking not just “how do I calculate this average?” but “how do I structure this so someone else could run it tomorrow?”
Thirty years of pattern recognition makes one thing clear: technical debt is a tax, always collected with heavy interest further down the road. Learning follows the same logic. What feels like a productive shortcut today is simply the seed of tomorrow’s frustration.
The Safety Harness
The most important thing I’ve done in this preparation period isn’t a framework or a tool. It’s writing unit tests.
Most people treat testing as an advanced topic — something you unlock once you’ve mastered the basics. I came to think that’s exactly backwards.
What I’m most afraid of, working in a new space, isn’t hard work. It’s invisible failure. A silent error in a spreadsheet that propagates through an analysis and surfaces three weeks later in a board presentation — that caution is professionally encoded after three decades.
Drawing on practitioners like Eugene Yan, I’ve started treating unit tests not as a chore but as an insurance policy for my own logic. Writing a test before or alongside the code does two things: it forces you to define what success actually looks like before you start — harder than it sounds when you’re still learning the vocabulary — and it lets you experiment without the paralysing fear of breaking something that was already working.
It is a way of saying: I may be new to this syntax, but I am not new to accountability.
Three Layers, One Direction
The biggest trap for senior professionals taking on technical learning is scope creep. We’re trained to see the whole system at once. We want to understand the neural network before we can reliably manipulate a list.
I forced myself into a sequence.
The first layer — variables, loops, functions, the basic logic of giving unambiguous instructions to a machine — is done. It sounds trivial. It wasn’t. The second — Pandas, NumPy, working with datasets as sets rather than rows — is also done. This is the layer that replaces the expert spreadsheet, and where the payoff starts to feel concrete. The third — PyTorch, scikit-learn, model building — starts in three weeks.
Most people rush to the third layer because it’s where the hype lives. But the models are only as strong as the plumbing beneath them. By taking the first two seriously, I’ve tried to ensure that when I reach the third, I’m still thinking like a strategist — not just pressing buttons and hoping for outputs.
What This Is Actually About
The value of this transition isn’t in the Python scripts I’ll write. It’s in the shift in judgment that comes from understanding the friction of the data — how hard it actually is to clean, how fragile a pipeline can be, where the assumptions hide.
There’s a version of this story where I frame all of this as a clever strategic move. An investment in future relevance. And it is that. But it’s also something smaller and more honest: I sat down at a blank editor, felt the discomfort of not knowing what I was doing, and kept showing up anyway.
That’s not the insight I expected to walk away with. But it might be the most useful one.
The goal isn’t mastery. It’s literacy. And literacy, it turns out, starts with being willing to look stupid for a while.
The full version of this article — including the Field Notebook with tools and references — is on the Data Consigliere site. If this resonated, share it with someone navigating the same transition.