The Real Bottleneck in AI Agents Isn't the Tech | Regina Lin, ThirdLayer

Why It Matters

Dex demonstrates how deep browser context can unlock practical AI automation for everyday knowledge workers, potentially reshaping the RPA market and boosting productivity across sales and go‑to‑market functions.

Key Takeaways

• Dex uses custom tool‑calling, not generic browser automation, for efficiency.
• Founders identified real pain after a demo impressed a UiPath bot manager.
• Target users are “browser‑only” knowledge workers, especially go‑to‑market teams.
• Early growth relies on word‑of‑mouth and personal use cases, not paid ads.
• Founder’s piano discipline translates into handling pressure and rapid iteration.

Summary

The interview with Regina Lin, co‑founder of Dex, explores why the real bottleneck for AI agents lies in workflow integration rather than raw model capability. Dex builds a Chrome extension that leverages contextual cues from the browser and executes actions via custom tool‑calling, delivering far higher accuracy and speed than generic automation approaches. Key insights include the founders’ personal frustration with fragmented digital chores, which led to a prototype that could order food and sync calendars. A chance demo to a UiPath bot manager convinced them the technology could replace legacy RPA, prompting a focus on “browser‑only” knowledge workers—particularly go‑to‑market and sales teams whose daily tasks live entirely in web apps. Early user acquisition has been organic, driven by the team’s own heavy usage and word‑of‑mouth on LinkedIn and Instagram, with retention prioritized over aggressive scaling. Lin frequently references her piano background, citing years of disciplined practice, high‑pressure performances, and improvisation when she forgot a concerto ending. She draws a parallel between that resilience and the iterative, feedback‑driven product development at Dex, noting the decision to drop out of Harvard to pursue the venture after a pivotal conference encounter. The broader implication is that contextual browser agents could automate a vast swath of repetitive admin work, freeing knowledge workers to focus on higher‑value activities. Dex’s early traction suggests a sizable market for tools that seamlessly bridge disparate web applications without requiring users to adopt new platforms, positioning the startup to challenge traditional RPA solutions.

Original Description

What if your browser knew your entire workday — and could quietly handle the parts that drain your focus?

In this episode of First Commit, Eliya Elon and Laura Hamilton sit down with Regina Lin, co-founder and CEO of ThirdLayer, the company building Dex — an AI workspace that lives inside Chrome and acts as a true knowledge worker for the browser.

Regina started building Dex as a college side project with her co-founder Kevin while juggling a math degree and years of competitive piano performance — originally just trying to get GPT-4o to order DoorDash at midnight. The insight that it could replace entire categories of manual, browser-bound work crystallized when they snuck into a Boston AI conference with fake lanyards and ran into someone managing hundreds of UiPath bots.

What Dex became is a hybrid agent that pulls context directly from the browser while executing actions through precision tool calling — not screen scraping, not MCP — built for the knowledge workers who spend every minute of their day living in tabs.

Join us as we explore:

-ICP Discovery: How Regina landed on go-to-market professionals as the core user — people whose entire workday is implicit, scattered across tabs, with no to-do list because their to-dos are the tabs.

-Hybrid Agent Architecture: Why Dex combines browser context with custom tool calling for maximum accuracy — and why the "pure browser agent" framing misses what actually makes it fast.

-The Education Problem: Why giving people agents isn't enough — and how human creativity, not technology, turns out to be the real bottleneck to adoption.

-Lessons from Piano: How eight-hour practice days, public masterclass critiques, and improvising a concerto ending she'd forgotten trained Regina for the pressure of building in real time.

[00:00] — First Notes, First Code: Regina's origin story — from competitive piano at age five to vibe-coding late-night DoorDash orders in college.

[03:00] — The Fake Lanyard Moment: How crashing a Boston AI conference with printed name cards produced their first real signal from a UiPath bot manager.

[04:05] — Harvard, Piano, and the Drop: What eight-hour practice days, public masterclasses, and a Harvard math degree have in common with building under pressure.

[06:25] — Improvising the Ending: The concerto Regina forgot and made up on stage — and why nothing in startup life has felt as daunting since.

[07:45] — Grounding Dex: What the product actually does, who it's for, and why the browser is the right command center for knowledge workers.

[09:35] — The Tab Behavior Revelation: How go-to-market people manage their entire workday through open tabs — and what that signals about where their attention actually lives.

[12:00] — From Finance to Market Research to Browser Agent: The YC pivot story and how Regina's investment banking internship crystallized the automation opportunity.

[16:00] — Browser Wrapped: What Dex's data reveals about how people actually organize their work — cyclers, hoarders, and window architects.

[18:40] — The Coworking Shift: Why the mental model moved from full automation to human-in-the-loop — and how that unlocked browser agents for mainstream users.

[21:20] — Married to the Browser?: Why Dex started in the browser, why it's not only the browser, and how custom tool calling outperforms screen-based agents by 100x.

[23:30] — 2026 vs. 2027: The unsolved problems — team context, agent interfaces, observability, and making agents reliable enough for real enterprise use.

[27:05] — Running a Company on Its Own Product: How a four-person team uses Dex to replace a CRM, auto-populate Linear, and analyze months of customer conversations in minutes.

[32:15] — The Creativity Bottleneck: Why human imagination, not model capability, is the real constraint to agent adoption right now.

[35:35] — The YC Pivot: What they were building when they entered the batch, why they walked away from a 50% CMO response rate, and what founder-product fit actually feels like.

[38:05] — Lightning Round: The daily task Regina plans to eliminate in two years, the most underrated consumer AI use case, and the one thing she'll never hand off to AI.

Follow Notable Capital:

Find Us On X: @notablecap

Find Us On LinkedIn: @Notable Capital

Visit Our Website: https://www.notablecap.com/

Subscribe on Spotify: https://open.spotify.com/show/7pOVa1E3o5Bjo0trJGxgO0?si=c0b89ab8f70e4101

Subscribe on Apple Podcasts: https://podcasts.apple.com/us/podcast/inprod/id1678521678