[kanjun]

Hi there! I'm Kanjun, Josh's cofounder — thanks for this lovely take on Imbue :) I feel quite seen, and wanted to share some replies to your thoughts.

On Sculptor: You’ve nailed it on the key things to get right. Today, I’d say the product is still very early, and it can indeed get expensive if you kick off tons and tons of agents. Today’s research preview release (currently free for testers) aims to slowly grow our community of testers, so we can get the product to identify and fix issues well, and enable actual software engineering with AI agents. We’d genuinely love to work with you as we build this out if you’re game to deal with early-stage bumps and crashes: DM me on X (I’m @kanjun) and we’ll get you set up with Sculptor.

On policy and mission: Thank you for the treatment on SB-1047. We care a lot about distributing power, and that showed us firsthand the complexities of regulating entrenched interests. We are not currently a PBC because the legal entity doesn’t seem to greatly affect the behavior of companies that are PBCs. It seems like the bigger effect is how leadership thinks about ethics and decision-making, and so for now, we take our own ethical grounding and behaviors seriously rather than relying on the legal structure.

We’re neither e/acc nor EA — I don’t fully resonate with either's treatment of the dynamics created by increasing AI capabilities, and I’m generally wary of this kind of tribalism as it can hinder critical thinking. My current primary lens is that of power distribution — we want to ensure that power is as much as possible in the hands of humans, and in particular, prosocial humans.

On Minsky: Wow, I’d totally forgotten about his Sculpting paper! But upon rereading, we are referencing a similar feeling to how software engineering could feel in the future — less need for intense precision, more like sculpting. Our view of how that gets implemented necessarily differs from Minsky’s (he was ahead of his time).

Indeed, we started our journey as Generally Intelligent (and posted jobs on HN all the time, heh), with the intent of figuring out how to address what we saw as core sociotechnical problems of AI. This product is, we believe, a step in that direction. In the next month or so, I will put out a more detailed treatment of the ideas around power, tailored for an audience keen on diving deeper.

Thanks again for the encouragement!