When I was sixteen, a summer book left a mark on me that never faded. It was philosophy, not code: Alexandre Koyré's essay on our passage "from the world of the approximate to the universe of precision."
His thesis stayed with me. For most of history we lived in a world of "more or less" and "roughly." Science, and nearly everything it built, became possible only when we crossed into a universe of measurement and precision.
Many years have passed. The foundations I learned that summer have not moved.
AI marketing wants universal adoption, and I understand why: it grows revenue. But it also does something genuinely good: it democratizes creativity and puts it into more hands than ever.
I won't take anything away from that honest part of the mission. But the technical reality is sadder: a tool built to "help" has quietly left too much behind. Too many details, and not only technical ones: philosophical, social, and ethical ones too. We can no longer ignore them without drawing the lessons that let us grow.
The good news: many teams are already learning, fast. They understand AI as an extraordinary Swiss Army knife: in the hands of an expert MacGyver, it solves an astonishing number of problems. The human stays at the center, directing the tool, as it has always been.
But the next lesson is already waiting: how detailed must the thing we ask for actually be?
An archetypal sketch of a house may be enough to imagine it, to plan it, even to picture yourself living in it. The vibe is enough to dream.
It is not enough to build.
You don't pour foundations "more or less." You don't vibe-code the load-bearing wall. Even the term's own inventor draws that line: Andrej Karpathy, who coined vibe coding, now calls the serious craft "an engineering discipline." Meanwhile, AI agents have wiped production databases they were told not to touch, and vibe-coded apps have shipped with user data wide open.
And since rigorous logic is still not a domain AI has mastered, perhaps we should pay a little more attention to the order and the method with which we lay the foundations of this house.
That is what I have started to call Intentional Coding.
AI still belongs in the process. Intentional Coding just asks for method again, built on two old, very human ideas:
• FSOP, a full-stack optimization mindset: best practice (security, correctness, tests, maintainability) as the default at every layer, instead of bolted on at the end.
• ITIL-style discipline: the lifecycle rigor, order, and accountability IT service management has quietly taught us for decades.
Intentional, because you choose the structure you're building, at every layer. Nothing is left to chance.
The approximate is a wonderful place to dream. It is a terrible place to build.
Where does your team draw the line between AI as copilot and AI as author, and who's accountable when it's crossed?
The approximate is a terrible place to build—especially when compliance, data sovereignty, and security are on the line.
I am currently building a Sovereign European, AI-assisted GRC (Governance, Risk, and Compliance) and Cybersecurity platform designed precisely for teams that cannot afford to "vibe-code" their security posture.
If you are a CTO, CISO, or Engineering Leader navigating European compliance and AI security, let’s connect:
Follow me on LinkedIn for updates on our journey and early beta access.
Join our private waitlist contact@praesidai.fr
Top comments (2)
The concept of Intentional Coding resonates deeply, as it highlights the importance of rigor and precision in software development, especially when AI is involved. The distinction between "vibe coding" and intentional coding is particularly relevant in fields like security and compliance, where the stakes are high and approximation can have severe consequences. I appreciate how you've woven together philosophical and technical insights to make a compelling case for a more thoughtful approach to coding, and I'm curious to see how the principles of FSOP and ITIL-style discipline can be applied in practice to balance the benefits of AI with the need for human oversight and accountability. How do you envision Intentional Coding evolving in the context of emerging technologies like Explainable AI and edge computing?
The evidence is already in:
• Veracode's 2025 GenAI Code Security Report: AI-generated code introduced an OWASP Top-10 security flaw in ~45% of tests.
• A controlled METR study (July 2025): experienced developers were 19% slower on mature codebases with AI, while predicting they would be faster.
• CodeRabbit (December 2025, 470 pull requests): AI-coauthored PRs carried ~1.7x more issues than human-only ones, and 2.74x more cross-site scripting.
• GitClear's multi-year analysis: as AI assistants went mainstream, copy-pasted code overtook refactored code for the first time, and duplication climbed sharply.
Sources: Veracode 2025 GenAI Code Security Report · METR, "Measuring the Impact of Early-2025 AI on Experienced OSS Developer Productivity" (Jul 2025) · CodeRabbit, "State of AI vs Human Code Generation" (Dec 2025) · GitClear AI Code Quality research (2025).