Why we are building Kunobi

Increasingly, we take an idea to an AI agent before we discuss it with another person. We share unfinished thoughts, explain the background and explore possibilities. The agent may come to know our work better than the people we could collaborate with.

What other people bring

Sharing with people introduces something else. They have their own interests and experience. A colleague may challenge an assumption. A partner may want something quite different. We have to explain ourselves, listen and sometimes negotiate. That can be uncomfortable. It is also how an idea grows beyond what we could have imagined alone.

Much of cultural progress depends on that exchange. Someone encounters an unfamiliar perspective, questions something they took for granted and carries the discussion further. I want AI to help more people participate in that process.

An agent can challenge us too. But a person brings their own stake in the outcome. When colleagues choose a supplier, purchasing may see an attractive price while operations remembers the cost of maintaining the previous installation. Both concerns deserve a hearing.

Some differences call for better evidence. Others require a decision about what matters most.

Sharing all our conversations would give others a great deal to read. It would not necessarily give them a reason to engage. A proposal they can question, or a discovery they can use in their own work, offers a more useful starting point. The reasoning should be there when someone needs to examine it.

This matters between coworkers, across disciplines and outside work. Friends planning something together face the same basic challenge: each person knows things the others do not, and each has something to contribute.

Desktop software expanded what individuals could produce; the web and shared documents changed how others could contribute. AI raises that question again. The idea of augmenting human intellect has a long history, and I think its collaborative ambition deserves as much attention as individual capability.

The freedom to keep choosing

The growing attention around shared AI environments, including ChatGPT Space and Claude Cowork, is encouraging. More people are exploring how AI can participate in ongoing work. Those environments also ask us to invest in a place: our knowledge accumulates there, colleagues join, and workflows develop around it.

Over time, changing the intelligence we use can mean rebuilding that whole arrangement. Exporting documents helps, but the relationships and custom tooling around them may be much harder to carry forward. A convenient choice becomes a dependency that involves other people too.

We should be able to rent intelligence without surrendering control of our knowledge, working relationships and tools.

An assistant needs relevant context to help. Giving it that access should leave us able to choose another provider later and continue the work. That expectation should apply to Kunobi too. The work should remain useful beyond whichever application helped produce it.

Our freedom to choose also shapes how we collaborate. Working alone with an assistant can narrow the perspectives we encounter. Building our collaboration inside one provider's environment can narrow our practical choices. Both carry a risk of monoculture. Choosing several models does not, by itself, bring other people's experience into the room.

Building on each other's work

We are building Kunobi as a shared workspace where people and their chosen assistants can contribute to ongoing work. A colleague's finding should help someone else take the next step. A decision should remain available to question and revisit. Each person should be able to participate while choosing what to share and keeping their preferred AI.

That freedom also needs to be accessible. Many new capabilities first reach people who can install and configure them. We want to make worthwhile advances easier to discover and use, and help people share what works. A choice that requires expertise someone does not have is difficult to exercise.

These are the two convictions behind Kunobi: collaboration and freedom. Can people accomplish something together that none could have reached alone? Can they keep building on that work as their needs and their choice of AI change?

In future posts, I'll explore how these convictions shape the product. That includes making contributions easier to share, opening capable tools to more people, and the choices that open models, local computing and peer-to-peer collaboration offer for privacy and control.