Hey Arun, Thanks for the feedback. That’s fair. We’ve added several safeguards: • The proposed reply goes to the rep for feedback, then another Claygent reviews it before sending regardless of whether feedback is provided. The prompts are based on 600+ Lemlist conversations, and the tagged rep remains responsible for approving the final response. • Two Claygents evaluate ICP fit. Non-ICP accounts are routed to Slack, while Claude reviews them daily to catch missed opportunities. • Our Apps Script triggers are time-based, though we continuously monitor and replace sources that may degrade. Apps Script is also useful for parsing newsletters from Gmail. • Everything is logged in Google Sheets - companies, contacts, sources, scores, and outcomes so we can measure performance and improve governance.
Hey everyone, I've been building an outbound engine where the only human step is the call. Curious if anyone's gone down a similar road. Intake: trade newsletters, RSS, article pages, Google search, Instagram posts and reels, YouTube podcast transcripts. Through Appscript we get fresh information daily. Clay: pulls the operators out of each item, verifies domains, dedupes, scores ICP, finds contacts, writes personalised messages based off whatever the actual signal was. Reply handling - the bit I'm most pleased with: a reply comes in on email or LinkedIn, Claygent drafts a response, it lands in the rep's Slack channel. They approve it, replace it, or just tell the bot "shorter" or "she's asking about pricing, be direct" and it redrafts and sends the reply in the same thread. Coaching happens in the channel, not another tool. Lead only reaches a human when there's a conversation worth having. Apps Script, Clay, Claude, Lemlist. No custom infra. What looks fragile to you? And is anyone else using Apps Script as the glue into Clay, or is there something better for scheduled ingest?
