I got back into coding — without really becoming a coder again
AI did not turn me into a software developer. It restored a direct execution capability I had gradually delegated as my CTO / CTPO role expanded.
The desire to regain execution autonomy never disappeared
For several years, I had wanted to do certain things myself again: turn an idea into a product, fix a behavior, test a user journey, or ship an improvement without immediately depending on a team.
I did not plan a grand return to coding. That was a deliberate choice: I worked with brilliant developers, and my impact was usually higher elsewhere.
Product, data, strategy, architecture, technology watch, management, organization design, hiring, and trade-offs already filled the space. Returning to a sustained developer role would have carried a real opportunity cost.
Being hands-on does not necessarily mean opening an IDE
I have always tried to stay close to the actual work: understand constraints, challenge architecture, read metrics, clarify a product decision, or help a team unblock a situation.
That proximity is a form of hands-on work. It does not require writing every line of code yourself.
But I sometimes missed the final mile: the ability to move from a need to a deployed version when the scope was small enough for one person to understand end to end.
AI changes the cost of individual execution
I can now discuss the objective, possible solutions, trade-offs, and edge cases, then have the result implemented, tested, and corrected within the same loop.
My role remains very close to CTO / CTPO work: provide context, make priorities explicit, assess risks, arbitrate, and verify that the solution answers the real need.
The difference is that this conversation can now produce code and a testable state directly. Intent, implementation, and feedback sit much closer together.
On small projects, I can own the entire chain again
Rebuilding this website helped me regain that autonomy. I documented how I left a site builder to reconstruct xavier-barry.fr with Astro, GitHub, Netlify, and an AI-assisted workflow in my website rebuild field note.
With Mama Flow, the loop is richer: product, acquisition, subscriptions, personal data, and real operations. I explain how I divide the work between Lovable, ChatGPT, and Codex instead of asking one tool to do everything.
In these personal and contained contexts, I can follow the full thread: need, decision, code, checks, deployment, and observation of the result.
This practice also strengthens my work at scale
Getting closer to execution gives me concrete references for what these tools can do, the controls they require, and the new operating models they enable.
In a transformation engagement, I do not spend my days generating code. My use also covers analysis, synthesis, decision preparation, and continuity across many sources. I shared a concrete example in my article about daily AI use at Bedrock.
The point is not to reduce AI to a coding assistant, but to understand how it brings reasoning, decisions, and execution closer together.
This does not generalize to every codebase or organization
A mature codebase, critical infrastructure, or a large organization carries history, dependencies, and risks that one person cannot reasonably absorb through a conversation.
In at-scale engagements, I continue to work with experts. I rely on their system knowledge, judgment, and ability to sustain technical choices over time.
My goal is not to replace developers. It is to regain autonomy where it makes sense and improve my ability to collaborate with them everywhere else.
When friction disappears, we also need to recreate boundaries
Being the only human in the loop makes it possible to move forward at any hour without waiting for anyone. That lack of friction is an advantage, but it can also blur the line between when we can continue and when we should stop.
That opens another question: how do we protect free time, attention, and rest when AI dramatically increases our individual execution capacity? It deserves an article of its own.