Multi-agent pipeline that researches accounts, drafts sequences, and enrols them, with the rep in the loop.
Outbound at Lumapps was a manual motion. SDRs across EMEA and NORAM spent 2 to 3 hours a day on account research, prospect lookup, and email drafting that converted at average rates. Pipeline was hard to forecast and only scaled with headcount.
I designed and shipped a multi-agent pipeline wired into Salesforce: lead fetch, account and prospect research agents, a sequence generator, and a push-to-Outreach step that lands cadences in the rep workflow. Reps approve before anything sends. Evaluations and reporting run weekly. I owned the whole program: architecture, rollout, evals, and the change management that gets a sales team to trust an agent.
The agent was the easy part. Sales-approved outbound meetings tripled (80% more per rep, since the team also grew), and each rep got 30 to 60 minutes a day back. That only happened after I put approval, evals, and reporting in front of the model. Trust is built with the loop, not the prompt.
Forward deployed engineer is the hottest title in AI. If you want it, aim at the part that never goes stale: owning the outcome. Advice for aspiring FDEs from someone about to wear the badge.
Towards AI A rep told me I had built half the product. So the agent now drafts the full 10-touch sequence and drops it into Outreach, ready for the rep to review and send. Nothing goes out on its own.
Towards AI Honest write-up of building the AI sales agent. What moved the metrics, what didn't, and the surprising amount of plumbing under a working agent.