Sortlist turned its manual outbound routine into an automated loop using Codex. Every morning, the system reads raw inputs—CRM exports, job-board scrapes, funding announcements—then filters for signals that matter: real hiring triggers, fresh funding rounds, visible pain points. Strong judgment at this “sense” stage weeds out noise. Only accounts with clear, factual evidence move on. Next, Codex scores each signal by priority. Recent, high-intent triggers beat vanity metrics. You set a threshold—70 out of 100 in the first version—to cap early drafts at five solid messages. Above that cutoff, the engine drafts personalized outreach. Then a separate check job flags any potential missteps—overused buzzwords or weak claims—before queuing messages for human review. That morning review takes under ten minutes for twenty accounts. Every send, reply, or lost deal feeds back into the system. Outcomes and raw reasons—“wrong contact,” “no budget until Q4,” “already in-house solution”—go into a shared memory file. Weekly, Codex reads those logs and suggests adjustments: signals to retire, score rules to tighten, banned phrases. Over time, the engine sharpens itself, saying no more often and surfacing only the highest-value opportunities.