Stacks/Sales
Production sales agents with thread-as-context (HN)
HN user hirewilliam describes a production sales-agent pattern: put the whole conversation and the original prospect research in context, give each agent a bounded task, and route on reply classification with a thin orchestrator. They explicitly err toward human review when confidence is low. Framework and model names are not given.
- Use case
- A production sales context: multi-turn conversations with real prospects who may reply days apart, covering research, drafting, sending, and parsing replies
- Category
- Sales
- Team size
- not stated
- Orchestrator / framework
- Agents own clearly bounded tasks end to end (research, draft, send, parse reply), with a thin orchestration layer that routes based on reply classification. The author says classification is the hardest and most important part to get right. Ambiguous situations are flagged for human review rather than guessed, because the cost of saying something wrong to a real prospect is high.
- Memory
- The entire conversation thread is treated as the context window: every reply, every prior message, and the research done on the prospect at the start is passed through, rather than processing each reply independently.
- Evals
- not stated
- Observability
- not stated
- Monthly cost
- not stated
- Models
- not stated
- Observability tools
- not stated
- Other tools
- Human review for ambiguous situations
- Source
- https://news.ycombinator.com/item?id=47685726
hirewilliam · Apr 6, 2026