Perspectives Technical journal
Perspectives
Analyses and field notes from an intervention CTO on complex systems.
The most innovative solution is sometimes the simplest
A solution's complexity measures neither the difficulty of the problem nor the quality of the answer. Innovating sometimes means removing, not adding.
AI, when it solves. Not when it decorates.
AI is neither a goal nor a default feature: it's an engineering tool. Its value is measured by the problem it solves, not by its presence in the stack.
Temporary solutions often end up in the final product
Temporary shortcuts help you learn fast. But with no recorded reason and no review date, they have a surprising property: they become permanent.
A reversible technical decision doesn't deserve three weeks of debate
The more costly a technical decision is to reverse, the more it deserves study. When it's reversible, moving forward often teaches you more than debating.
Keep every option open, and you eventually stop moving forward
Keeping several technical options open protects you from a bad decision — until the point where it mostly starts to cost you time.
Optimising too early can kill invention
In R&D, applying the final product's constraints too early can eliminate an idea before you've even understood what it made possible.
Experimental failure is data
In R&D, a negative result isn't wasted time: if it rules out a hypothesis or sharpens the next experiment, it has just reduced the uncertainty.
Every component works. Yet the system fails.
Several perfectly compliant sub-systems can together produce non-compliant behaviour: some properties emerge from the interactions.
A working PoC doesn't prove you have a product
A proof of concept reduces one specific uncertainty. It proves neither the robustness, nor the reproducibility, nor the manufacturability of a product.
The problem isn't always where you're told to look
When every component is compliant but the system misbehaves, the cause often lies in their interactions, not in any single part.