Human-in-the-loop: why AI agents need a leash
Why we always require human confirmation before AI actions affect real data — and how we designed the flow.
Product thinking, engineering decisions, and the messy reality of building an AI-first company from Helsinki.
Why we always require human confirmation before AI actions affect real data — and how we designed the flow.
How we chain tool calls so Leiko completes multi-step tasks instead of stopping halfway.
The reality of bootstrapping from Helsinki: Y-tunnus, VAT, invoicing, and the quiet grind.
How we trained our OCR pipeline to handle Finnish receipts, K-Market formats, and low-quality photos.
How we balance information density with calm — showing what matters without cognitive overload.
How we track, cap, and optimize token usage so AI features stay sustainable at our scale.