Notes from building OnlyLeads
What we believe about outbound, and the specific problems we hit and solved while running it for real: deterministic pipeline, LLM where it earns its keep, a human on every message.
Campaigns, sequences, and a sorting hat, without going on autopilot
Repeatable campaigns for Black Hat and beyond, multi-step sequences, and smart routing across senders, all still gated by human approval. Here's how we kept the discipline while adding the power.
Read →Never pitch your product to a competitor's employee
Two embarrassing failure modes (messaging a competitor's staff, and leaking the wrong product angle) and the guardrails we built so neither happens again.
Read →The bug that taught us to snapshot the truth
A draft can outlive the facts it was written from. Here's a quiet data bug we shipped, how we caught it, and the principle we took from it.
Read →You approve every message, and that's the feature
The review queue isn't a safety compromise on the way to autopilot. It's the product. Here's why we'll never make it optional.
Read →We removed the bullshit agent layer
Every "AI SDR" sells an autonomous agent as the magic. The agent is the bug. Here's what we built instead, and why it costs a fraction to run.
Read →