I have no idea how long it would have taken me to firgure out even a part of what Jed showed us during the challenge. I knew a fair amount about lead sources and available databases, but I've found that's where the issue starts; the problem isn't knowing that the data exists or even knowing where to get it; the problem is sifting the valuable information from ALL the information. That's what we learned. I haven't become an "expert" by any means, but it was an extraordinarily valuable use of (rougly) 90 minutes a day for five days to learn some real skills.

Is it just me, or is the Clauder Browser working a lot fast than just last week?

After spending a couple of days iterating on lead sourcing & data enrichment in cowork, I'm finding that upon iterating, cowork seems to drift from some of the foundational infrastructure design rules & principals that I originally design into it. Despite building clear markdown files, overarching instructions, and edits to these files during iteration, I still notice that it makes up completely new infrastructure at times vs. iterating on existing. I think once I prove the output I want, I might need to recreate some of the more engineering heavy components of my build in code. Curious if others have encountered this?

Manus is good but their token costs calculations are obscure, and there isn't much room available for optimizations besides prompt length and other basic items. I ended up re-creating the initial sourcing pipeline in OpenClaw. New system runs entirely in OpenClaw using Sonnet 4-6 for Enrichment and Analysis, but using lower tier models for basic and routine tasks and has dramatically reduced costs New features: - Sourcing (moved from Manus to OpenClaw. initial pass that gathers the first set of leads using ICP criteria) - Ingest (discards non qualified leads) - Enrich (deep pass that enriches qualified leads and verifies emails. Scores quality for instantly enrollment) - Campaign enrollment (qualified leads get assigned to an Instantly campaign) - Reply (alerts me via WhatsApp whenever there is a reply) - Metrics (pulled from Instantly via API) - Insights (new tab that collects performance such as click rates and things that could be useful for improvements) - Analyst agent (reviews metrics and insights. Proposes updated to copy, campaigns, delivery, etc to increase conversion) - Guardian agent (maintains health of the system. Flags issues and warnings) - Self- Healing agent (reviews flags from guardian, metrics and runs Sonnet for improvements. Self corrects if it finds an issue during any of the pipeline runs) Next steps: - Increase volume. Currently pulling about 25 leads every other day for testing. Will increase and run agent daily

@Jed Morris You speak of having a regres or progres that is free that you have set up and put everything on your desktop as to not be reliant on chatgpt or claude to store you data. Could you explain more in depth of how I could implement this on my PC?