Why we build AI-native, not AI-added

Most “AI features” released in the last few years share a pattern: an existing product gets a chat window bolted onto a corner of the screen. It can answer questions about the product, sometimes take an action on your behalf, and it demos well. But it rarely changes how the underlying product works.

At Lesteko we’ve tried to hold ourselves to a different standard: if removing the AI layer from a product would leave the core workflow unchanged, we haven’t actually built an AI-native product — we’ve built a regular product with a chatbot attached.

What AI-native means in practice

For each of our five products, we start with the same question: what does this tool do differently if AI is a core assumption, not an add-on?

• Datoteka doesn’t just let you ask about your files — its organization and search are built on the assumption that meaning, not folder structure, is the primary way people find things.
• Zvezek doesn’t summarize notes on request only — it continuously builds structure across your notebook so that asking a question is answering against something already organized.
• Analitik treats natural-language queries as a first-class interface to streaming and relational data alike, not a translation layer bolted in front of SQL.

The cost of doing it this way

This approach is slower. It’s much easier to ship a chat sidebar in a sprint than to rethink a core workflow around a new capability. We’ve shipped features later than we wanted to, more than once, because the “AI-added” version was ready first and we chose not to ship it.

We think that tradeoff is worth it. Software that people actually rely on daily earns that trust through structural changes, not surface-level ones — and that’s the bar we hold every Lesteko product to before it ships.