@J T 1) lead flow thats next on the list for rebuild. 2) seo automation that researches, drafts, publishes blog content as well as a companion email & social media posts and GBP update. 3)And I’m in the middle of a social media content build that indexes source information I submit throughout the day, develops the weekly strategy, drafts the posts, creates the images, critic passes fail everything, pushes to slack for human approval, has a learning agent watching 24h & 7d post metrics that rewrites the strategy for the next run, and has an ops manager overlooking the whole thing that sends me weekly health updates via Slack. All orchestrated in make.

Very nice. Considering doing the challenge but my time is already limited with my current business and other build projects I’m in the middle of. How much time per day are you committing to it?

@Omotola Shogunle Thank you

Welcome welcome! What type of service business are you running and how are you applying AI or automation in it now?

I'm finding the same thing within my own business. I think the capabilities will get better with time but right now passing my judgement down through skills and .md files still doesn't cut the cake for the majority of tasks I've tried to automate. At some point I still have to be in the loop, even if its the last step.

Update: I was able to fix some of the checks by editing the prompt and making the critic_checklist descriptions more concrete for the LLM to test against. Claude may have also answered my question on the outliers I was frustrated with. I was running harness tests on the edits, and my focus was on getting every check to pass consistently above a 50/50 pass rate. I was chasing a ghost on some of those subjective checks. The solution: The hard-fail checks will stay hard. If something triggers a hard fail, it still loops back to the drafter for revision. But soft fails will no longer gate the pipeline. Instead, soft fails will be passed forward to the human-in-the-loop review step with warnings, descriptions of the failure, and the same fix instructions that would have been sent to the LLM revision round. What I found was that some of the subjective judgment checks for brand voice were failing about half the time, but when I reviewed the outputs myself, I would have passed the majority of them. Since nothing goes live without human review, those soft-fail items do not need to stop the whole process. They just need to be surfaced clearly so I can make the final call before publishing. At the end of the day I may have been trying to delegate too much of the decision making.