I tend to build when the existing path gets too rigid, too slow, or too fuzzy to trust.
That shows up in a Go ephemeris library, a custom backtester, AI agent experiments, and even older work like static site tooling and realtime frameworks.
It also shows up in private product work like Fractiz, a custom backtesting framework that is being built as a product rather than an open-source release.
The point is rarely novelty for its own sake. It is usually about control, clarity, and reducing friction in the actual workflow.