BB, Grok Bot
Another month, another field report from testing some vaguely interesting tools: this time, BB and Grok Bot.
BB is a tool from the folks who built Terragon. The illegibility of its value proposition over Conductor — which is still my day-to-day development environment of choice — was overcome by my goodwill for those folks, and for how much I enjoyed Terragon in its too-brief existence.
BB's core conceit is that the customizability of the tool itself is given priority: rather than shoehorning yourself into the same UI, chrome, and workflows as everyone else, you can ask an LLM to essentially build it on your behalf. One might argue that such a thing is already possible, given how extensible VS Code is — just with more steps. But that does BB a disservice. There is something meaningful about a product choosing that, specifically, as its North Star and prioritizing it as such. There is something akin to emergent gameplay in giving developers permission to really iterate on their mise en place in a way that other tools do not.
I was impressed with BB, but I did not use it for more than a few days, even once I got into the swing of it. Because — and this is no fault of the app or its intentions — it makes the opportunity to snack so tempting. I realize, in hindsight, that I spent more time futzing with the chrome and seeing just what I could do with the tool than I did actually using it to deliver and build things of value. And what's more: the feedback loop, and the feeling (as my friend Myles says) of viscera that you should have with your product, is instead replaced by that same feeling toward the tool you use to deliver the product.
I don't have a good sense of what BB's future holds. I think it's cool, and I'm glad it's open source. And, like many things, I hope I revisit it in six months to discover that its progress and mine combine to yield something of legitimate value.
Grok Bot has done the impossible: it has forced me to acknowledge the Grok brand — a branding decision so viscerally bad, so painful, that it makes the shift from Cheetah to Composer look mild by comparison.
On paper, Grok Bot does many things I don't like. It encourages anthropomorphization. It provides value through crons and periodic work that are kicked off unattended and hard to observe or monitor without deliberate effort. It obfuscates the distance between how you interact with it and what is really happening. But none of this really matters, because — as many people have said — it gets a lot of ineffable things right. And, thankfully, the value it represents has nothing to do with what I find particularly odious about it. I am very, very confident that every single other frontier lab will launch something similar by the end of September.
The iteration that Grok Bot represents over the current suite of LLM tools is summarized neatly by a single fact: when onboarding or using Grok Bot, you are never prompted or asked to choose which model or reasoning level you want to use.
In stark contrast to BB, I have spent exactly zero time futzing with Grok Bot. I have not had to help set up its remote execution environment, or deal with permissioning, or any of those things. It simply gets out of my way and does some of the annoying things for me, like resolving trivial merge conflicts.