Hallway Chat, #28 - Mastering scale, Google I/O and judging "what's worth making a startup"

We return from a break to discuss the effects of an avalanche of app making due to Claude and Codex, including “Camp,” an experiment by Nabeel in native multiplayer AI-assisted group work beyond shareable outputs. We cover: In order to be a founder leading AI transformation do you need to lead by example? Conductor’s viral “prompt feature requests” workflow, reality of one-shot apps versus iterative prompting, how teams may use less open-source, Gemini's comparative strengths, and what does it mean when the engineering pod optimal size has moved from six to two. We end by discussing Granola’s MCP and why data moats are fragile, favoring best interfaces and customer-centric access. 00:00 Divergent Paths: Two types of Engineering post Claude 4.5 00:00 Introduction: The New Reality of Coding 00:21 Building Camp: Multiplayer Knowledge Work 03:12 Open Source in the Age of Models 08:56 The Recommendation Problem: From Average to Expert 15:29 Why Gemini Works for Personalization 18:32 Submit a Prompt: Conductor's Product Innovation 21:50 Two types of Engineering: Automatic vs Iterative 32:08 Rethinking Team Structure: From Six to Two 35:01 Can you AI transform without living it yourself? 40:12 Data Moats and the MCP Shift 44:05 Making Context Ubiquitous

Why are dentists so into vibe coding? Where's the Steve Jobs of this era? In this episode of Hallway Chat, Fraser and Nabeel discuss the viral AI assistant Poke, within the broader context of the rise of niche AI applications. They also talk about the importance of specialization in AI applications, AI’s role in advancing scientific discovery and societal progress, and the future of AI in creative fields like music and art. Plus, a compelling use case for AI-enabled accessibility.

In this episode of Hallway Chat, Nabeel and Fraser unpack the emerging power of tool use, reasoning traces, and code execution in modern models. They reflect on how startups can stay ahead in a market that feels like it changes every 90 days, and what product choices separate the winners from the also-rans. The conversation spans Conductor’s clever planning/execution design, Notion’s everywhere-at-once API strategy, and the subtle ways models persuade us—sometimes too effectively. Along the way, they explore how founders should handle VCs circling during competitive raises, why misinformation concerns may be shifting, and why Fraser still turns to Gemini for certain jobs despite its quirks. It’s a candid look at the red-ocean dynamics of AI, the importance of owning your narrative, and how the future of work may mean managing a team of agents as naturally as we once managed people.

Nabeel and Fraser sit down with Andrew Mason, founder of Descript, for a deep dive into evolution "open world agents" in AI products. The trio uses the recent Airtable re-founding as a lens to look at open world agent architecture, what makes a great AI-enhanced user experience, and how Descript is embracing agent-first workflows as well. From the positioning of chat windows to the philosophy of “rewiring” users for AI, this conversation explores the strategic decisions shaping the next generation of software tools—and what it really takes to build a product users trust, understand, and love.

Craft in the age of AI. Nabeel and Fraser start with a dive into what makes the Boundary backpack so great. They explore how AI is changing the way we do customer research, the importance of obsessive attention to detail, and why great founders still need great taste. Along the way, they debate what it takes to recreate the creative magic of “1930s Paris” for modern product builders.