Just Now Possible

What happens when a customer reports a stolen credit card? The frontline answer is simple—freeze it. But underneath lies a cascade of follow-ups: dispute filings, fraud investigations, merchant communications, and proactive outreach to gather more details. Most AI support tools handle only the tip of the iceberg. In this episode, Teresa Torres talks with Jack Taylor (Product Engineer) and Ibrahim Faruqi (AI Engineer) from Gradient Labs, an AI-native startup building agents that automate the full scope of customer support in fintech. They share how they've architected a platform with three coordinating agents—inbound, back office, and outbound—all built on a shared foundation of natural language procedures, modular skills, and configurable guardrails. You'll hear how they: - Let non-technical subject matter experts define agent behavior through natural language procedures—no coding required - Architected a state machine orchestrator that manages turns, triggers, and skill selection across long-running conversations - Built guardrails as binary classifiers with eval pipelines, tuning for high recall on critical regulatory checks - Designed an auto-eval system that samples conversations for human review to catch edge cases and build labeled datasets It's a detailed look at how one startup is moving beyond simple Q&A bots to agents that can actually take action, coordinate across workflows, and handle the messy reality of customer support.

What if your small business could have a full marketing team—automated content calendars, customer segmentation, and channel-specific posts—without the headcount? In this episode of Just Now Possible, Teresa Torres talks with Chris O'Connor (CEO) and Jessica Valenzuela (Co-Founder) of Mowie, an AI marketing platform built for small and medium-sized businesses in restaurants, retail, and e-commerce. Chris and Jessica share how their hands-on experience managing marketing for overwhelmed business owners at a previous company led them to build Mowie—first as a concierge service, then as a fully automated AI product. They walk through their document hierarchy approach: how Mowie crawls the web to build a "dossier" about each business, infers customer segments and marketing pillars, and generates quarterly content calendars with channel-specific posts. You'll hear about the technical challenges of structuring unstructured data, the evolution from rigid schemas to loosely structured markdown, and how they use customer feedback—from calendar approvals to regeneration requests—as their primary evaluation signal. Whether you're building AI products that synthesize messy real-world data or figuring out how to keep humans in the loop without overwhelming them, this conversation offers practical lessons from two founders who built their product by doing the work first.

What happens when you combine a real customer problem, a no-code prototype, and a team willing to listen to every single call? In this episode of _Just Now Possible_, Teresa Torres talks with Steven Payne (Product Manager), Gabriel Stock (Senior Engineering Manager), and Philipe Steiff (Senior Software Engineer) from Perk—a company that helps businesses eliminate "shadow work" like travel booking and expense management. They share how they built a voice AI agent that calls hotels to verify virtual credit card payments, preventing travelers from arriving to find their rooms unpaid. What started as a hackathon experiment in Make.com became a production system handling over 10,000 calls per week across multiple languages. Along the way, the team learned hard lessons about prompt engineering for voice (numbers, pronunciation, and a very "Karen-like" first version), how to break a single monolithic prompt into structured conversation stages, and why listening to actual calls beats any amount of theorizing. You'll hear how they: - Built a working prototype without writing a single line of backend code - Structured the call into discrete stages (IVR, booking confirmation, payment) to improve reliability - Created two eval systems: one for call success classification, another for conversational behavior - Scaled from five calls a day to tens of thousands per week while maintaining quality This is a detailed look at building AI for real-time human interaction—where the stakes are high and the feedback is immediate.

What if you could get personalized sleep coaching—inspired by the same principles that cost thousands of dollars and have year-and-a-half waitlists—through a voice AI that checks in with you every morning? In this episode of Just Now Possible, Teresa Torres talks with Martin Siniawski (CEO and co-founder) and Ignacio (CTO) from Rest about how they built an AI sleep coach inspired by Cognitive Behavioral Therapy for Insomnia (CBTI) principles. The journey started when they noticed users of their podcast app were listening to content to fall asleep, explored sleep audio solutions, and eventually pivoted to an AI-powered voice coach when LLMs emerged. They share how they evolved from basic chatbots to a sophisticated voice-first system with memory, dynamic agendas, and RAG—all while navigating the tricky line between wellness and medical products. Their "one bite of the apple at a time" approach to building AI offers practical lessons for teams tackling complex, personal AI products.

Accounts payable inboxes can see 1,000+ vendor emails a day. Xelix's new Helpdesk turns that chaos into structured tickets, enriched with ERP data, and pre-drafted replies—complete with confidence scores. In this episode, Claire Smid (AI Engineer), Emilija Gransaull (Back-End Tech Lead), and **Talal A.** (Product Manager) walk us through how they scoped the problem, prototyped with “daily slices” (Carpaccio-style), and built a retrieval-first pipeline that matches vendors, links invoices, and drafts accurate responses—before a human ever clicks “send.” We dig into tricky bits like vendor identity matching, Outlook threading, UX pivots from “inbox clone” to ticket-first views, and the metrics that prove real impact (handling time, stickiness, auto-closed spam). We close with what’s next: targeted generation, multiple specialized responders, and more agentic routing.