Why we're building Kaiso
AI coding tools are getting faster. They're not getting smarter about your project. We think the tool should be smarter than the user needs to be.
Where Kaiso began
Kaiso started from a pattern we kept seeing: AI tools can generate impressive code, but they fail silently. A data pipeline that produces plausible-looking output with subtle errors nobody catches. A web application that handles three user flows perfectly and has no error handling on the fourth. The problem isn't that AI is bad at coding. It's that AI coding tools have no architectural memory — no understanding of the whole project, no sense of what's consistent and what's missing.
We started by building tools for data analysis pipelines, where this problem is sharpest — wrong results look exactly like right results until you know where to look. That experience crystallized the vision: an IDE that carries the engineering intuition most builders don't have.
The vision
Most AI coding tools assume an experienced developer is supervising — someone who knows when the AI's suggestion is architecturally wrong, who notices when a pattern is incomplete, who carries a mental model of the entire project. Kaiso builds that mental model itself.
At its core is a semantic graph — a continuously updated, navigable map of your entire codebase that captures not just code structure but domain meaning: what your variables represent, how data flows through your system, what patterns and conventions exist. When you ask for a change, the AI explores this graph the way an experienced engineer would — starting from the big picture, drilling into relevant subsystems, tracing implications before writing any code.
The feature we're most excited about: Kaiso detects what's missing, not just what's broken. If your pipeline handles numeric data thoroughly but skips date fields that flow through the same path, Kaiso sees the asymmetry and asks about it. If your web app has error handling on some routes but not others, Kaiso flags the gap. These are the observations a senior engineer would make in a code review — and exactly the things that non-expert builders have no way to catch.
Marcelle Bonterre After nearly fifteen years building machine-learning and data-intensive systems, one pattern kept repeating: the gap between what AI can do in a demo and what it can do for someone building real software alone is enormous.
I've watched the vibe coding wave with a mix of excitement and concern. People with no traditional engineering background are building remarkable things with AI. But they have no way to know when the AI is leading them astray — when the code works but the architecture is accumulating invisible problems. Every other tool assumes you have twenty years of intuition. Kaiso is the tool that gives you that intuition on day one.