Humans should always be in the loop. LLMs are extremely capable but a computer can never be held accountable. Context and decision-making must be transparent. Many LLM tools are built to reduce human agency, suggesting that we contribute a few tokens once in a while. Teambook is built with the opposite idea that we humans should use our new capabilities to increase our agency.

Your data is your data. This means both portability and security. You should always be able to easily take your data and leave. And allowing agents to act on our behalf implies high levels of privilege, so security is a primary concern.

Workflow portability matters. LLM providers have begun to understand that models are not the moat once believed, and now they're doing what they can to secure the market with restrictive terms-of-service and subsidized pricing. We've seen this playbook before: eventually the external capital dries up and prices increase. Models should serve your workflows, not vice versa.

Chat is the wrong interface for an advanced tool. Linear conversations were a familiar way to introduce LLMs, but human thought takes many forms: linear or organic, textual or visual, static or interactive. Interfaces should be flexible to match our thinking.

Collaboration is fundamental. Ideas come from everywhere and our best work is done together. These new bicycles for the mind can take us further than ever, but going with a team takes us further still.