Building Lorikeet: How AI Humility and a Dual-Agent Architecture Are Redefining Customer Support

This episode of Just Now Possible features Lorikeet, a startup building AI customer support concierge agents for regulated industries. The team describes how months of work on the wrong ideas, such as reflection tools and dashboards, gave way to a simple request from a healthcare customer to clear their inbox. Lorikeet now runs two agents: a Concierge that handles tickets end-to-end and a Coach that helps customers configure and improve it. The conversation covers the case for AI humility, meaning defaulting to human handoff under uncertainty, and customer-defined guardrails. For PMs building AI products, it is a concrete look at designing trust, escalation, and configuration into an agent.

01Key takeaways

• Validate the real pain before building: the team's early tools missed what customers actually needed.
• Design AI agents to default to human handoff when uncertain, treating humility as a core principle.
• Make guardrails domain-specific, since generic rules can break in specialized contexts like regulated industries.
• Let customers define what good looks like (evals and guardrails) before asking them to write procedures.
• Use AI to diagnose failure modes in traces and suggest fixes, shortening the improvement loop.
• Integrate with existing support tools like Zendesk and Intercom rather than forcing replacement.

02Key sections

Origin and early missteps

The team first explored reflection tools and information dashboards that did not solve a real problem. A healthcare startup's request to simply clear the inbox redirected their focus.

Dual-agent architecture

Lorikeet splits its product into a Concierge that resolves customer tickets and a Coach that handles configuration, testing, and continuous improvement.

AI humility and handoff

A core principle is that the agent should hand off to a human when uncertain, and guardrails must be domain-specific, as a cannabis company's tickets showed.

Configuration UX evolution

The interface moved from a workflow builder to a conversational Coach, and the team is now flipping the workflow so customers define what good looks like before writing procedures.

Resolution in the loop

Human agents can unblock the AI on a specific issue without taking over the ticket, enabling human-AI collaboration.

03From the post

“Listen to this episode on: Spotify | Apple Podcasts What does it take to build an AI customer support agent that actually knows when it can't help — and says so? In this episode of Just Now Possible, Teresa Torres talks with Jamie Hall (Co-founder & CTO), Xharmagne Carandang (Product Engineer)”

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.