Our vibe coded projects that actually work
Episode
33 min
Read time
2 min
Topics
Design & UX, Artificial Intelligence, Software Development
AI-Generated Summary
Key Takeaways
- ✓Vibe Coding Success Criterion: The "toothbrush test" filters viable projects — only build something you will use at least twice daily. The majority of vibe-coded projects fail not because the technology cannot build them, but because the builder never identifies a genuinely recurring personal need before starting. Narrowing scope to one specific friction point dramatically increases the chance of daily adoption.
- ✓Tool Comparison — Claude Code vs. Codex: Claude Code produces functional, problem-solving code with clear step-by-step guidance but generates visually plain interfaces by default, requiring manual design direction. Codex produces aesthetically polished outputs immediately but confuses non-developers with unclear next steps and failed Expo Go version compatibility during mobile app setup, making Claude Code the stronger choice for functionality-first builders.
- ✓Browser Extensions as Vibe Coding Entry Point: Chrome extensions are the lowest-risk starting project for beginners — they are structurally simple, load instantly without deployment, and have a narrow failure surface. Pierce built a working web clipper tab-filing system as a browser extension faster than any other project, reducing open browser tabs and RAM usage immediately upon completion.
- ✓Security Vulnerabilities Are Universal and Unresolved: Both Claude Code and Codex flagged 15 high-severity and 15 medium-severity security vulnerabilities in Field's habit tracker, yet both tools advised ignoring them. Vibe-coded apps should not be shared publicly or deployed to open internet without developer review, as AI tools systematically underreport risk to non-technical builders who lack context to evaluate warnings independently.
- ✓Data Durability Requires Explicit Backup Instructions: Kastronakis discovered his meeting notes app had no backup system and proactively prompts Claude weekly to audit data integrity. The practical fix — asking the AI to export all stored data as plain text files — takes one prompt and prevents total data loss from local cache failures, a risk present in any locally hosted vibe-coded application.
What It Covers
Vergecast hosts David Pierce, Jake Kastronakis, and Hayden Field share vibe-coded projects they built and actually use daily — a meeting notes app, a custom Gmail label interface, and a personalized habit tracker — then receive a four-week challenge to build personal websites using AI coding tools.
Key Questions Answered
- •Vibe Coding Success Criterion: The "toothbrush test" filters viable projects — only build something you will use at least twice daily. The majority of vibe-coded projects fail not because the technology cannot build them, but because the builder never identifies a genuinely recurring personal need before starting. Narrowing scope to one specific friction point dramatically increases the chance of daily adoption.
- •Tool Comparison — Claude Code vs. Codex: Claude Code produces functional, problem-solving code with clear step-by-step guidance but generates visually plain interfaces by default, requiring manual design direction. Codex produces aesthetically polished outputs immediately but confuses non-developers with unclear next steps and failed Expo Go version compatibility during mobile app setup, making Claude Code the stronger choice for functionality-first builders.
- •Browser Extensions as Vibe Coding Entry Point: Chrome extensions are the lowest-risk starting project for beginners — they are structurally simple, load instantly without deployment, and have a narrow failure surface. Pierce built a working web clipper tab-filing system as a browser extension faster than any other project, reducing open browser tabs and RAM usage immediately upon completion.
- •Security Vulnerabilities Are Universal and Unresolved: Both Claude Code and Codex flagged 15 high-severity and 15 medium-severity security vulnerabilities in Field's habit tracker, yet both tools advised ignoring them. Vibe-coded apps should not be shared publicly or deployed to open internet without developer review, as AI tools systematically underreport risk to non-technical builders who lack context to evaluate warnings independently.
- •Data Durability Requires Explicit Backup Instructions: Kastronakis discovered his meeting notes app had no backup system and proactively prompts Claude weekly to audit data integrity. The practical fix — asking the AI to export all stored data as plain text files — takes one prompt and prevents total data loss from local cache failures, a risk present in any locally hosted vibe-coded application.
Notable Moment
Kastronakis built a fully functional custom Gmail client over a single weekend to solve one problem: surfacing personal labels as top-level tabs. The unintended consequence was accidentally replicating the core feature set of Superhuman, a product that charges users thirty dollars monthly.
Episode Transcript
Hello, and welcome to the Vergecast, flagship podcast of the Vercel REST API. I'm your friend David Pierce, and today on the show, we have a challenge. The way this is gonna work, we're gonna we're gonna do a bunch of these in a row, but the way this is gonna work is we've gathered a bunch of people. We're going to give them all a challenge, and then we are going to go off for several weeks and report back on what we found. Today, we are starting with two of my colleagues, Jake Kastronakis and Hayden Field, and we are going to do a vibe coding challenge. The first thing we're gonna do is kind of lay the land, talk about where we all are in our vibe coding experiences, and then we're gonna all go see who can do the best in vibe coding. It's a lot of fun. I'm very excited about this project in part because it's gonna get me to, like, actually get some work done for a change for the podcast. It's gonna be awesome. But first, here's everything else happening on The Verge today. This is ninety seconds on The Verge for Monday, 06/29/2026. Mythos is back, kind of. The anthropic model that was once deemed too dangerous to to release publicly and actually threw even more chaos into the already chaotic relationship between anthropic and the US government is now being released to a Trump administration approved list of organizations. Who those organizations are, who they will be, whether you'll ever get to use Fable five again, hard to say. Nobody seems to quite understand the AI regulatory environment right now. And given that the Supreme Court also ruled today that President Trump has the authority to fire two FCC commissioners, which changes the way that agencies work, whatever happens here, it will start with the White House. Meanwhile, Comcast is splitting into two companies. In the latest of a seemingly endless supply of media mergers and unmergers, now there will be Comcast, the Internet cable and mobile carrier, and there will be NBC Universal, which is the studios, theme parks, and streaming networks, including Peacock. None of this, by the way, applies to Versant, which Comcast spun off in 2024. The media business just continues to be chaos. And a fun side note of this, by the way, is that I have a new disclosure to do here. I guess it's gonna be something like Comcast is an investor in PMX, the parent company of The Verge. I'm gonna have to get used to that one. And finally, quick PSA for you. WhatsApp is rolling out username support later this year so that you can find people and be be found without having to share phone numbers. The feature's not live yet, but you might be able to go to settings, account, username, and reserve yours. There are, like, 3,000,000,000 people on WhatsApp, so you should probably move fast. You …
Get the full transcript (7,234 words) + summary by email — free
One-time email with the complete transcript and AI summary of this episode. No account needed.
One email, no spam. We’ll also show you what SignalCast does.
You just read a 3-minute summary of a 30-minute episode.
Get The Vergecast summarized like this every Monday — plus up to 2 more podcasts, free.
Pick Your Podcasts — FreeKeep Reading
More from The Vergecast
Should you let a robot mow your lawn?
Aug 13 · 33 min
How I Built This
Advice Line: New Offerings, Bigger Markets
May 14
More from The Vergecast
Pixel 11 and Pixel Watch 5: Our first impressions | The Vergecast Livestream
Aug 12 · 57 min
This Week in Startups
5,000+ Tech Workers Laid Off This Week. It's Just The Beginning. | E2286
May 9
Books, tools, and gear mentioned in this episode
SignalCast may earn commission on purchases via these links. As an Amazon Associate, SignalCast earns from qualifying purchases.
Tools
- Claude CodeRecommended
by Anthropic
“Claude Code produces functional, problem-solving code with clear step-by-step guidance but generates visually plain interfaces by default, requiring manual design direction. Codex produces aesthetically polished outputs immediately but confuses non-developers with unclear next steps and failed Expo Go version compatibility during mobile app setup, making Claude Code the stronger choice for functionality-first builders.”
by OpenAI
“Claude Code produces functional, problem-solving code with clear step-by-step guidance but generates visually plain interfaces by default, requiring manual design direction. Codex produces aesthetically polished outputs immediately but confuses non-developers with unclear next steps and failed Expo Go version compatibility during mobile app setup, making Claude Code the stronger choice for functionality-first builders.”
by Expo
“Codex produces aesthetically polished outputs immediately but confuses non-developers with unclear next steps and failed Expo Go version compatibility during mobile app setup, making Claude Code the stronger choice for functionality-first builders.”
Products
by Superhuman
“Kastronakis built a fully functional custom Gmail client over a single weekend to solve one problem: surfacing personal labels as top-level tabs. The unintended consequence was accidentally replicating the core feature set of Superhuman, a product that charges users thirty dollars monthly.”
More from The Vergecast
We summarize every new episode. Want them in your inbox?
Should you let a robot mow your lawn?
Pixel 11 and Pixel Watch 5: Our first impressions | The Vergecast Livestream
Lessons from the very first chatbot
Pencils down: We share our vibe-coded websites
What's behind the Google AI shakeup
Similar Episodes
Related episodes from other podcasts
How I Built This
May 14
Advice Line: New Offerings, Bigger Markets
This Week in Startups
May 9
5,000+ Tech Workers Laid Off This Week. It's Just The Beginning. | E2286
Odd Lots
Apr 17
Planet Money Turned Everyday Annoyances Into an Economics Book
This Week in Startups
Mar 25
$2.5B Chip Heist, The Future of American AI, and Purpose-Built Robots | This Week in AI Ep 6
My First Million
Mar 13
This guy names billion dollar brands for a living, here’s his exact 3-step formula.
Explore Related Topics
This podcast is featured in Best Tech Podcasts (2026) — ranked and reviewed with AI summaries.
Read this week's AI & Machine Learning Podcast Insights — cross-podcast analysis updated weekly.
You're clearly into The Vergecast.
Every Monday, we deliver AI summaries of the latest episodes from The Vergecast and 192+ other podcasts. Free for one show.
Start My Monday DigestNo credit card · Unsubscribe anytime