I started building Tachiko Work because spreadsheets are incredibly useful, but once they become real operational systems, their meaning gets buried in cell positions, formulas, exports, scripts, and manual review. I wanted a workspace where the structure itself is explicit: typed data, stable identities, formulas, dependencies, validation, and semantic changes all live in one model. That also changes how AI can work with spreadsheets. Instead of letting a model directly manipulate cells or files, GPT-6 Astra can reason about the user’s intent and propose a typed semantic change. Tachiko then independently calculates the consequences, validates the result, and asks for approval before applying it. For this challenge, I’m exploring that boundary between probabilistic reasoning and deterministic execution: Astra reasons; Tachiko verifies. Tachiko Work is still a public pre-alpha, and I’d especially love feedback on this approach to AI-native structured work.

Tachiko Work

Semantic spreadsheets where AI proposes, Rust verifies.

Semantic spreadsheets where AI proposes, Rust verifies.