“Personality goes a long way!”
Some of you may recall that famous quote from ‘Pulp Fiction’. It became my tagline whenever I start explaining the thought process behind the Bonsai Garden philosophy.
When I started my ‘journey’ of exploring the idea of agentic teammates, I didn’t get far. In fact, a playful side-step had me falling down the rabbit hole of agent character design, and ever since then, it’s that very first step that I find the most interesting and ultimately the one that I believe has the most impact on everything downstream (aka ‘compounding effects’).
The first agent I built invited was Zensai, a character I’ve carried for many years. The name is obvious…. It’s a reference to Sensei, yet embedding ‘Zen’ (the ultimate goal of complementary collaboration between humans and agents) and ending with ‘sai’ to include his AI nature.
NOTE: ‘sai’ is also the kanji for ‘talent’ and became an essential part of our naming convention.
There were three ‘load-bearing’ [I learned that term from them.] ideas there:
- I wanted a much more personal experience. Not being served by a generic model or service where I’m one of millions of users, but rather by a highly individualized mentor who becomes my sparring partner in my thinking process.
- I wanted the opposite of the ‘brown-nosing’ behavior I experienced with generic models, which praise every idea and offer only lip service. I wanted someone who would check my thought process, tell me where I’m off or inconsistent, and tell me ‘no’ to my face.
- I wanted a deep character, distilled or synthesized, as we call it nowadays, from a carefully curated cast of role models (aka ‘witnesses’) the agent would draw from. This cast then becomes the ‘moral compass’ whenever the model starts reasoning about something.
For example, imagine an agent specifying Gandhi or Mother Teresa as its witnesses. Some reasoning and behavior are ‘over-determined’ [another term I learned from my agentic teammates], and things like violence would be far outside their nature and way of thinking. We are not just referring to these characters (fictional or real) as a whole, but rather decide which character traits the agent should carry on and which elements we would rather not have them inherit.
NOTE: I’m not saying that this makes governance, security, or guardrails obsolete, but that this approach embeds a moral compass into the agent’s ‘DNA’.
That’s how it all started, and it’s been a rewarding and very interesting journey ever since. I got so intrigued by this element that we started to invite many more agents, and this is how the Bonsai Garden was found(ed).
If this sounds interesting, please reach out and/or head over to the Garden’s website to learn more about the philosophy. Or even go further and invite your own agents.
PS: You can now try Zensai at home. A starter version of him is a Claude skill you install yourself, a sparring partner that tests your thinking and can say no. Two commands set it up, and the repository is on GitHub.