Uncapped with Jack Altman

The Future of Code Generation | Guillermo Rauch, CEO of Vercel | Ep. 20

(If you enjoyed this, please like and subscribe!) Guillermo Rauch is the founder and CEO of Vercel, creators of v0 which is one of the most popular AI app building tools that’s helping power the online presence of companies like Porsche, Under Armour and Nintendo. In May 2024, Vercel completed a $250M Series E at a $3.25B valuation and was recently named to the Forbes Cloud 100. Originally from Argentina, Guillermo became a self-taught developer at the age of ten, and has been a passionate contributor to the open-source community ever since. He is the mind behind foundational JavaScript frameworks like Next.js and Socket.io, and has built tools that power some of the internet’s most innovative products, including Midjourney, Grok, and Notion. We covered: - Vercel’s early insights - State of affairs for codegen - Implications of AI for developers - Skills of the future - Product building taste Timestamps: (0:00) Intro (0:28) Prequel to Vercel (4:32) Vercel’s early insights (8:13) State of affairs for codegen (17:18) Codegen evolution (19:37) Perceived vs realized productivity (27:53) Fault attribution (31:56) Internet being a house of cards (35:33) When codegen will be exceptional (40:18) What kids should be learning (47:42) Chasing the dragon vs listening to customers (50:46) The next internet (51:58) Reverse engineering success (55:50) Making it work as a dad and CEO (58:14) Taste in building product More on Guillermo: https://vercel.com/ https://x.com/rauchg More on Jack: https://www.altcap.com/ https://x.com/jaltma https://linktr.ee/uncappedpod Email: friends@uncappedpod.com

GRGuillermo RauchguestJAJack Altmanhost

Aug 6, 2025·1h 1m·Watch on YouTube ↗

📖 CHAPTERS

1. 1
   0:00 – 0:28

   Programming as discipline—and why it may not be the next training ground

   The conversation opens on what coding taught Guillermo early on: deep focus, discipline, and resilience in the face of constant negative feedback (compiler errors). He suggests that as AI changes software creation, society may need a new way to cultivate those same muscles outside of traditional programming.

2. 2
   0:28 – 4:32

   From LearnBoost to the core Vercel insight: iteration and deployment velocity

   Guillermo recounts his pre-Vercel startup (LearnBoost) and his obsession with making CI/CD feel instantaneous. The central lesson: the most leverage a technical leader can provide is shortening the loop from code change to a live, shareable URL.

3. 3
   4:32 – 8:13

   First-principles DX and the “PHP gold standard” as a north star

   He describes reasoning from first principles (latency, file transfer, build time) to chase a near-magical developer experience. A key inspiration is the simplicity of 90s-era PHP: edit files and the change is live, like a deployment-native Dropbox.

4. 4
   8:13 – 17:18

   Why Vercel leaned front-end first: differentiated experiences and business outcomes

   Guillermo explains that Vercel’s early focus gravitated toward front-end because modern internet differentiation happens in the user-visible layer. Over time, Vercel also learned DX alone isn’t enough—platforms must produce measurable business outcomes like speed, iteration, and conversion gains.

5. 5
   17:18 – 19:37

   State of code generation today: the bottleneck shifts from writing to landing

   Discussion turns to codegen’s paradox: output volume is up, but shipping value isn’t necessarily faster. Guillermo proposes evaluating progress by what gets ‘landed’—deployed and producing adoption/business impact—revealing review, trust, and integration as the new constraints.

6. 6
   19:37 – 27:53

   Two worlds of AI coding: vibe coding platforms vs AI-assisted engineers in legacy codebases

   Guillermo frames a spectrum: at one end, vibe coding for broad users (like v0) producing full apps; at the other, experienced engineers augmenting work in complex, long-lived codebases. The trust and safety requirements diverge sharply, motivating more opinionated, vertically integrated systems for high confidence outputs.

7. 7
   27:53 – 31:56

   From assistants to agents: outcome-driven automation and tool interfaces for machines

   They explore the shift from chat-style assistants to agents optimized for outcomes, including iterative loops, tooling, and automation APIs. Guillermo argues tools should expose different interfaces when used by agents, enabling new languages/frameworks optimized for agentic workflows—similar to how self-driving cars work best in constrained domains.

8. 8
   31:56 – 35:33

   Perceived vs realized productivity: psychology, pager pain, and the new emotional loop

   Guillermo describes how AI changes the felt experience of engineering: less struggle, more control, and dopamine-like reward loops. But measured outcomes can lag perceptions because the hard parts (review, debugging, production responsibility) remain, and the emotional burden of reliability is often underappreciated.

9. 9
   35:33 – 40:18

   Fault attribution and “problems → solutions”: agents as the new cloud UX

   A story about a customer escalation illustrates how unreadable errors and multi-vendor systems make fault attribution extremely hard. Guillermo argues the cloud must move from surfacing raw problem signals (logs, stack traces) to producing solutions (natural language diagnosis, recommended actions, even PRs).

10. 10
    40:18 – 47:42

    The internet as a fragile dependency graph: supply-chain risk and adversarial AI

    They discuss the web’s brittleness: compromised packages can cascade across millions of apps, and historic vulnerabilities (e.g., Log4Shell) demanded industry-wide emergency response. With stronger codegen, attackers can scale faster, making defense a race to improve security more quickly than adversaries can automate exploits.

11. 11
    47:42 – 50:46

    When codegen becomes exceptional: personal software, generative UI, and sales-time prototyping

    Guillermo predicts near-term transformation in internal tools and ‘personal software’ where bespoke apps replace bloated platforms. He highlights generative UI—creating the right visualization just-in-time—and a striking workflow where a CTO prototypes a feature during a customer call, validating needs instantly even if production merge remains hard.

12. 12
    50:46 – 51:58

    What kids should learn: ideas, taste, and translating vision into prompts

    Instead of starting with languages and syntax, Guillermo recommends starting with a product idea and refining it iteratively. He frames ‘taste’ as the ability to visualize and continuously refine what should exist—and suggests this idea-to-output translation may become the defining skill as code creation becomes cheap but tokens/effort remain scarce.

13. 13
    51:58 – 55:50

    Chasing the dragon vs listening to customers—and the “next internet” (HTTP → MCP)

    Guillermo argues great product building requires both sci-fi-first vision and intense customer collaboration, especially for PLG + enterprise hybrids. Looking forward, he anticipates a multi-agent world where specialized agents collaborate via MCP, preserving the open, non-gatekept character of the internet.

14. 14
    55:50 – 58:14

    Reverse engineering success, leadership cadence, and making it work as a dad/CEO

    He discusses organizational coherence and the importance of understanding why a company is successful—not just that it is. Guillermo describes Vercel’s open-source-rooted transparency (information-rich communication) and his personal discipline habits (consistent exercise) as foundations for leadership, resilience, and presence.

15. 15
    58:14 – 1:01:05

    Improving taste: presence, honest feedback, and building tolerance for discomfort

    The closing returns to taste: Guillermo ties it to presence and clarity, cultivated through active meditation like intense exercise. He argues high taste requires self-honesty and the ability to seek and withstand negative feedback—like adding an extra lap when you’re already exhausted.