Grok Bot Review
Grok Bot was my first encounter with a harness (besides Claude Tag) that pulled me up a level—instead of interacting with an agent 1:1, going back and forth with it to get work done, Grok Bot automatically spun up a team of agents and delegated work to them. I just told my ‘chief of staff’ agent what I wanted, and it wrote the prompts and fed them to the other agents on the team. The team of agents talked to each other and the chief-of-staff agent kept me updated about progress and asked me clarifying questions.
3rd party integrations were easier to setup that I’m used to—getting Github, web search, and others on Grok Bot’s of first class supported integrations felt like something you wouldn’t need an engineering degree to pull off. I am really glad for that—that puts a lot more power into a lot more peoples’ hands and I think Grok Bot could be especially great for folks that manage a ton of communication surfaces, e.g. marketing.
I struggled to get good results with brown field coding projects. The agent team made a lot of poor decisions, got stuck, and needed a fair amount of handholding and direction to get to where I wanted them to go. I’d still rather use Claude Tag and Code for coding. I wonder if I’d get better results using Opus instead of Grok for the Grok Bot model, but I haven’t tried that yet.
All that said, I love the abstraction of an orchestrator agent managing a team of specialist agents each with a prompt, context, memories, tools, and permissions to operate with. This is the way. As much as I like Claude Opus, man, please barf your word vomit responses to my chief-of-staff (agent), I don’t have time for it.