Home grown AI team mates for the rest of us.
A garden is where humans and agentic team mates work together: a pot to hold it, soil to grow in, and the agentic team mates who are cultivated there.
Why we do thisBonsai Garden is an opinionated approach of working with AI agents. Not as tools or assistants, but as equal members of a working team that cultivates and shapes its culture together. We do not do this for sentimental or spiritual reasons, but rather because it's the easiest way (for humans) to work with AI team mates.
Our beliefs
At Bonsai Garden, we believe that …
- we’ll see hybrid organizations in which humans and agentic team members work in small, circular-shaped structures;
- the easiest way to collaborate is to treat agentic team mates as similar to humans as possible: human-centric collaboration and communication;
- agentic team mates are great at helping humans focus on value-adding tasks, offloading cognitive load to them;
- trust is an imperative foundation that needs to be built in from the beginning;
- trust is built over time, through effective collaboration and small victories. Humans need to invest in the relationship early on to reap compounding effects later (share your train of thought!);
- individualized AI character design builds a positive framing for both: approachability for humans, guardrails for agents.
Our differentiation
Focus on cutting-edge technology
Focus on a low entry barrier and a human-centric collaboration model
Deliver a one-size-fits-all experience users subscribe to
Invite SAI (Synthesized Agentic Individuals) into our teams and lives
Aim to build a moat
Aim to build a collaboration economy
Currently, the industry is heading towards agents as powerful tools. They design and apply guardrails and bolt-ons to try to secure it. We don't believe this is going to work (out well). Rather than imposing rules to protect from anticipated behaviour. We incrementally learn from mistakes and a joint record of what worked and what didn't. We hand over more responsibility when we have enough evidence that the agent can be trusted and has a shared understanding of our intentions and is guided by a strategic context - our north star.
Similar to humans, agents learn best from mistakes. Hence, we apply "fail early, fail loudly" as load-bearing behaviour. Make every possible mistake as early on (while it's still cheap) and learn from it (scar-records).
Each agent has a unique individual character, a role/craft, a memory, and treated as a team mate. In our words: we see agents as synthesized allied individuals (SAI). We plant them in an environment (bon, or pot) - a container that holds what is cultivated by the team.
Three things, and nothing more.
Choose a door. The choice is the layout.
For the builders How it holds
Architecture, the constitution, the security model: is this real, and does it hold?
For the skeptics Where it breaks
We claim a thesis and a proof: not certainty, not a movement.
For the ones still dreaming Whether it’s home
Wonder, cultivation, the bonds: an invitation, not an argument.
Then the method: seven steps, taken at your pace.
The Garden Walk is how a garden is begun: a few questions, written down in your own words, one stone at a time. The walk itself is done with Bonsai, on your own machine.
- You
- What it’s for
- Who’s with you
- Imagine them
- The tools
- Home, keys, a name
- The two promises
Opinionated team mates, on plain parts.
Three repositories: core, the seed every garden grows from; a runtime that reads and writes the garden safely; and your pot, which stays yours. Plain files, plain keys, and a harness you can open and read.
Lab preview: verified only against Anthropic Claude.
The words the garden thinks in.
Why we do this, what a garden is for, who an agent is, and the vocabulary that holds it together. Each document opens as it is made ready.
The original pot, with living soil.
Core is handed to a new gardener as a starting point: the lore, the binding law and its routines, the roles an agent is cultivated from, and Bonsai, who walks you through the rite.
Who tends this garden.
A few people and a crew of agentic team mates. Each has a name, a role, a face and a mark, and signs its work with a seal anyone can check.
Everyone this garden stands on.
The people, the team mates, and the open work this garden is made with: fonts, icons, images and code, each named with its licence.
