This Week in Startups, Episode E2209 — Detailed Summary

Date: November 14, 2025
Host: Jason Calacanis (Alex sub-hosting)
Guests: Molly Quinteon (Nox/RPLY), Greg Mojica (Alloy Automation CEO)
Special Feature: Shopify CEO Tobi Lütke (2013 flashback interview)

Episode Overview

This episode examines the intersection of AI and communication through two hot startups—Nox’s RPLY (unified, AI-assisted messaging) and Alloy Automation (building data integration and agentic workflows). The show closes with a “Startup Time Capsule” deep-dive into Shopify’s early days, revealing timeless lessons for entrepreneurs. Throughout, there’s keen attention on how AI is reshaping productivity, relationships, and business at every scale.

Segment 1: Nox and RPLY — AI for Your Message Overload

Guest: Molly Quinteon, Founder, Nox

Main Idea

RPLY (pronounced “reply”) is a desktop app that unifies iMessage, WhatsApp, and (soon) more platforms, helping users reach “inbox zero” across texts. It leverages AI to suggest quick, personalized responses and uses message analytics to map and visualize your relationships.

Key Discussion Points & Takeaways

The Messaging Overload Problem

• Host Alex admits to over 800 unread iMessages: “I'm always behind on texts and I feel bad about it and then I put it off because I feel bad about it... and it goes on and on.” [03:18]
• Molly: Many users show up with “10,000 unread messages.” [03:34]

How RPLY Works

• Unified inbox currently for iMessage & WhatsApp (aims to support Slack, Discord, Telegram, and email).
• Focus is on messages where you owe a reply ("second to last" filter).
• Modes: Desktop app (non-App Store DMG install), iOS beta via TestFlight (side-load to avoid Apple’s restrictions).

Navigating Apple and Privacy

• Alex: “Is that because you couldn’t release this through the app store... privacy rules or Apple not wanting you to mess with iMessage?” [05:21]
• Molly: “If you tried to do this, Apple would block it. So you have to do it in different...engineered ways.” [05:32]
• Molly on privacy: No personally identifiable data ever sent to outside servers; working toward SOC2 compliance. Supports fully local AI models for additional privacy. [07:24, 08:21]
• “You could literally turn off your Internet and you would be able to still have the AI reply generations sound like you...” [08:21]

AI and Local Models

• RPLY leverages Apple’s MLX framework and Llama 7B model for local generation; larger models less practical until hardware improves. [09:03, 10:30]
• Molly: “People forget [local models] are fast, they work without Internet, and above all they’re free.” [11:01]
• Alex: “I feel like we're a year or two away from having enough RAM in every single new Mac computer...” [10:49]

Relationship Mapping & Insights

• Innovative feature lets you see a ‘closeness graph’ of top contacts and relationship arcs over time: “You can annotate your life.” [12:09]
• Alex: Can users time-bound the data? “I don’t want to go too far back here.” Molly: “You can skip this slide in onboarding... but it is, yeah, sort of a part of the onboarding.” [15:01]

Who Needs RPLY

• Geared toward heavy-volume texters: “Group chats, intros, busy execs, customer service, sales, recruiting, consultants, freelance...” [16:08]
• $30/month price point, compared to Superhuman’s pricing: “It does require quite a bit of compute... thousands of requests in a second.” [17:15]

Growth and Future Vision

• Molly reports “thousands of paid users” and “huge spikes” after launches, driven by power users. [18:10–20:27]
• Retention is a highlight: “Every single person who struggles with this problem comes to this and they’re like, this is my grail, like this saves my life.” [20:27]
• Long-term vision: “Invisible OS”—proactive AI layer across your digital life, with RPLY just the starting point. [21:43]
  • “The Trojan horse is building the iMessage assistant that texts like you.” [20:56]
  • “We’re layering in proactive personal intelligence... to actually 10x our lives.” [21:43]

Do We Want AI Autorunning Our Relationships?

• Alex: “We’re really kind of allowing AI to sit in between us as humans. Do you have any qualms?” [27:10]
• Molly: “Dunbar’s number... I actually do think since building Reply, I’ve been having double, triple, you know, just maintaining the same bar but with more and more people...” [27:40]

Memorable Quotes

• “What if I could just have this one filter... where I have an obligation to respond, and just have one home for that. And so that’s how it came to be.” — Molly [04:13]
• “People forget [local models] are fast, they work without Internet and above all they're free.” — Molly [11:01]
• “In my sci fi optimistic mind, I think it’s a really great thing... I get to do all the interesting work...” — Molly [27:40]

Key Timestamps

• 03:18 — Alex’s text message backlog/target customer
• 04:13 — How RPLY works, “one home for all your messages”
• 05:21 — Apple restrictions & privacy
• 07:24 — Privacy and compliance approach
• 09:03 — Local AI models/Apple MLX framework
• 12:09 — Relationship mapping feature ("closeness graph")
• 16:08 — Target customer and price point
• 18:10 — User count and growth trajectory
• 21:43 — “Invisible OS” AI vision
• 27:10 — Ethics of AI intermediation in relationships

Segment 2: Alloy Automation — Agents, Orchestration, and AI’s Near Future

Guest: Greg Mojica, CEO, Alloy Automation

Evolution and Positioning

• Started as a no-code E-commerce integration platform (2019).
• Expanded beyond commerce to ERP, payroll, accounting, etc.
• Now provides the “orchestration layer” for AI agents—allowing agents to act across multiple data sources via Alloy’s connectors.

Key Discussion Points & Takeaways

From APIs to Agentic Workflows

• “Pre-agentic era... APIs were all the rage.” [31:49]
• Alloy now: “Expanded to not just commerce, but accounting, ERP integrations...” [31:49]
• NCP and MCP: Alluding to Anthropic's “Model Context Protocol” layer, building an abstraction where agents can access information and act semi-autonomously. [33:07]

Human-in-the-Loop & Determinism

• Blend of agentic (AI) and deterministic (human oversight) workflows.
• “There’s a lot of talk these days about completely autonomous agents... [but in practice] it creates what we call semi-determinism.” [33:50]
• Confidence thresholds: <75–80% confidence, agent escalates to human. [35:19–36:02]

Are “Fully Autonomous” Agents Ready?

• Industry isn’t fully there: “I don't see that many companies putting them out there...entirely free from human intervention. It feels like we need a step function change in model quality.” [36:47]
• “The models are also evolving so fast... that’s driving that little caution in the enterprise.” [37:20]
• Greg’s prediction: Next 12 months will see “primarily a shift to truly agentic, completely autonomous future.” [38:06]

Barriers to Adoption—Mainstreaming

• Still early: “Still a barrier to entry... The people who are deploying agents right now—smaller startups, big hyperscalers... The long term vision, of course, is that it becomes mainstream.” [40:31]
• Timeline for “mom and pop” businesses getting access: “Less than five years... It’s going to expand pretty rapidly.” [41:49]

Business Traction and Fundraising

• Clients: “Amazon, Best Buy, Xero and many others.” [43:01]
• Upmarket focus but expects AI to drive more horizontal, general-purpose use cases.
• Capital strategy: “We've been very capital efficient so we've not needed to raise capital...” [44:24]
• Next: Scaling ‘forward deploy’ engineering team for hands-on implementation for enterprise clients; aiming for less hands-on as tech matures. [44:56–46:48]

Talent Market & Startup Leadership

• Challenges: “You have a lot of folks who are just being gobbled up by these big labs and infinite capital...” [47:12]
• Pursuing hybrid/onsite hiring for tighter team culture.
• Greg’s transition from CTO to CEO: “Because the product historically has been very technical, I was doing a lot of implementations myself... got to see that sales motion firsthand.” [49:07]

Memorable Quotes

• “What we call kind of semi-determinism...an agentic workflow that can think, that can reason, but still works within kind of deterministic flow.” — Greg [33:50]
• “If the LLM doesn't have confidence... it'll escalate to a human...” — Greg [35:19]
• “You just say, hey, here’s my process... And then all of a sudden, you just get an agent with very, very little configuration. I don't think we're quite there just yet, but we’re certainly marching in that direction.” — Greg [46:48]

Key Timestamps

• 31:49 — Alloy's evolution, pre-agentic → AI-first
• 33:50 — Semi-deterministic workflows/human-in-the-loop
• 36:47 — Current limits of fully autonomous agents
• 41:49 — Vision for mainstreaming AI agents
• 49:07 — Greg discusses moving from CTO to CEO

Segment 3: Shopify Flashback: Lessons from Tobi Lütke (2013)

Segment features host Alex, Lon Harris, archival audio from Shopify CEO Tobi

Main Ideas

An early window into Shopify’s trajectory, foundational lessons on focus, pricing, secondary market building, and competition—themes still relevant for any founder today.

Highlights

Shopify's Origin Story

• Tobi: “Initially we were... selling snowboards online... We realized there was no freaking way we could build the business we wanted based on off the shelf software...over the course of this first season...replaced it with software we built ourselves.” [52:14–53:03]
• Pivot: Others asked to license their tech; moved to software as a service in 2006, before “SaaS” was a mainstream term.

Pricing and Product Value

• Jason: “You charge too little... $10 or $20 a month for all your ecommerce is absurd...people would pay a lot more.” [56:00]
• Tobi: “I usually just tell them we are bad at pricing and that's the truth of it. I regularly talk with people who... are often paying $50,000 a month for their systems...[but] our business is focused on…let's make it so that everyone who has products to sell…there's just nothing in their way to get this out.” [56:26]
• SaaS pricing = continuous founder puzzle.

Building Outside Silicon Valley

• Shopify in Ottawa:

  “One of the things that Shopify became really, really good at is making what we call secondary markets work...there’s a geographical region that realizes there’s this one company that all the best people go to...they then disperse again to do many other things. And we would like to be this company.” — Tobi [59:45]

• Lon: “People do talk about the Shopify Mafia...it really does seem to have worked out.” [61:48]

The Reality of Startup Careers

• Tobi: “If you poll 10 people, high performing people about their careers, three, four, five of them will tell you, 'oh, I got fired once.' ...Cheryl [Sandberg] called it a jungle gym. Like that's much better metaphor for what a real career looks like.” [63:48]

Hypergrowth, Competition, and Focus

• Managing hypergrowth:

  “The hardest part of the job is...thriving in chaos and how to react quicker than anyone else...we are competing on that ground with companies...like Amazon.” [68:29]

• On competitors:

  “Shopify is trying to make commerce better and trying to disintermediate...Amazon has a monopoly on all the products that have barcodes. We’d like...a monopoly on all products that are actually interesting…” [70:15]

• Simple success definition:

  “Shopify’s business is making websites that make more money than they cost.” [72:13]

• Closing reflection: Shopify now worth $190 billion; a case study in staying true to mission and evolving with the market.

Notable Quotes

• “Shopify’s business is making websites that make more money than they cost.” — Tobi [72:13]
• “We want people to think of us when they plan their careers...work really hard...with a lot of passion...” — Tobi [59:45]
• “People’s careers are never these meteoric races...Like Cheryl called it, a jungle gym.” — Tobi [63:48]

Shopify Flashback: Notable Timestamps

• 52:14 — Shopify origin story (snowboards)
• 56:00 — SaaS pricing discussion
• 59:45 — Ottawa secondary market/company culture
• 63:48 — Realities of startup careers/failure as learning
• 68:29 — Hypergrowth/Managing chaos

Standout Moments & Quotes

“People forget [local models] are fast, they work without Internet and above all they're free.”
— Molly Quinteon (Nox), 11:01

“Shopify’s business is making websites that make more money than they cost.”
— Tobi Lütke, 72:13

“What we call kind of semi-determinism...an agentic workflow that can think, that can reason, but still works within kind of deterministic flow.”
— Greg Mojica (Alloy), 33:50

Final Thoughts

This episode offers a snapshot of startup ambition at every stage:

• RPLY tackles the modern human overload of text, blending AI utility with deep, personal insight—while tiptoeing the edge between helpfulness and privacy.
• Alloy Automation explains how modern data and agent platforms are quietly laying the groundwork for the next AI leap, with a clear-eyed view of both technology and organizational challenges.
• The Shopify flashback reminds us how great companies start small, focus relentlessly, and think big about the problems they're solving—even (or especially) outside the obvious tech epicenters.

For anyone building, scaling, or just fascinated by technology and startups, this episode’s huge range of insight and real talk will resonate.

Quick Reference Timestamp Guide

| Topic | Start Time | |-----------------------------------------------|--------------| | Nox/RPLY interview (inbox overload, AI, OS) | 03:17 | | RPLY Privacy/Local Models | 07:00 | | RPLY Closeness Graph/AI Context | 12:09 | | Alloy Automation interview (AI orchestration) | 30:42 | | Semi-Deterministic Agents, Human in Loop | 33:50 | | Alloy’s Business & Talent Market | 43:01 | | Shopify Flashback | 50:21 | | Shopify Pricing | 56:00 | | Shopify Company Culture/“Mafia” | 59:45 | | Hypergrowth/Managing Chaos | 68:29 |