Eric Siu explains why vague goals like "grow my product" produce bad AI outputs. Instead, set concrete targets: increase conversion rates from 1% to 1.5% in eight weeks, or achieve 50 ms page latency. Claude remains the best AI writing model, though GrokBot works better when embedded in platforms like Hermes.
When it comes to marketing, don't think that you can just type /go for everything. You got to make sure, hey, whatever you're trying to do from a marketing standpoint, maybe you're trying to lift your conversion rate from 1% to 1.5% in the next 8 weeks. That's verifiable. Or maybe you want 100% of your webpages to average 50 milliseconds. That's verifiable. But when you say, I want you to just grow my product, that's not very verifiable, right? Poor inputs, you get bad outputs at the end of the day. GrokBot as a harness tends to be very short. It doesn't do the best job of writing long-form content. When we have this as a skill and it runs in Hermes or runs inside of Codex, the output is a lot better. Claude is probably still best for writing, maybe not Opus, but maybe for Babel. But that depends on you if you want to burn the tokens for that.
