By Teresa Torres · producttalk.org · @ttorres on X · LinkedIn
This podcast episode features ShowMe's co-founders discussing how they build AI digital sales reps for inbound teams, treating agents as full teammates that run voice and video calls, demo products, and follow up over days. They explain how a single sales conversation is decomposed into specialized sub-agents to manage latency and model limits, coordinated by a deterministic workflow layer. The conversation covers why a realistic avatar changed prospect engagement, and how customer-driven eval loops, starting with full review and tapering to about 5%, maintain quality for revenue-critical interactions. It is a practical look at what it takes to build agents that actually sell.
01Key takeaways
- Decompose a complex conversation into specialized sub-agents to manage latency and model limitations.
- Use a deterministic workflow layer to coordinate multi-day buyer journeys before attempting free-form orchestration.
- Review every agent conversation early, then taper sampling as quality stabilizes, keeping a small ongoing review set.
- Turn customer feedback into automated regression tests to prevent prompt changes from breaking earlier fixes.
- Use confidence scores and frustration signals to decide when a human should take over a revenue-critical call.
- Ground sales agents in company-specific transcripts and playbooks, since generic prompting underperforms.
02Key sections
- Origin and MVP
- The founders describe spotting a website conversion gap at a prior company and launching a voice agent with product videos and a simple RAG knowledge base.
- Avatars and trust
- Adding a realistic avatar changed how prospects engaged, with video-call design cues acting as affordances that build trust.
- Multi-agent architecture
- The system splits conversation, evaluator, and creator agents, with an orchestration layer routing a sale through greeting, qualifying, and pitching stages.
- Teaching sales skills
- Company-specific playbooks are built by ingesting real call transcripts and training materials rather than relying on generic prompting.
- Evals and handoffs
- Confidence scoring, frustration detection, and customer-driven regression tests guide human handoff and keep production quality high.
03From the post
“Listen to this episode on: Spotify | Apple Podcasts What happens when you treat an AI agent not as a chatbot, but as a full teammate on your sales team – one that can jump on video calls, demo your product, make phone calls, and follow up over days? In this episode”
04Frameworks mentioned
Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.