Our Vibe Coded Projects That Actually Work the Vergecast

Our Vibe Coded Projects That Actually Work | the Vergecast ![rw-book-cover](https://i.ytimg.com/vi/4fUMfB-YLGo/sddefault.jpg) 🔗 **Source:** [Our Vibe Coded Projects That Actually Work | the Vergecast](https://www.youtube.com/watch?v=4fUMfB-YLGo) > [!note]+ Document Notes > tags: [vibe-coding, ai-coding, software-democratization, developer-tools, prototyping, ai-and-work] > author: David Pierce > source: The Verge (The Vergecast — YouTube) > reference: https://www.youtube.com/watch?v=4fUMfB-YLGo > date created: 2026-06-29 09:00 > > # 🏷️ Title: Our vibe coded projects that actually work | The Vergecast >  ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [Episode](https://www.youtube.com/watch?v=4fUMfB-YLGo) · [The Verge](https://www.theverge.com) - Tools named: Claude / Claude Code, OpenAI Codex, Xcode, Gmail API, Google Calendar, Todoist, GitHub, browser extensions ## 📝 Summary - **What happened.** Three Verge staffers — host David Pierce, executive editor Jake Kastrenakes, AI reporter Hayden Field — show-and-tell small personal apps they built by "vibe coding" (describing software in plain language, letting an AI agent write it), then set a four-week challenge to vibe code a personal website. - **Why it matters.** A candid, non-engineer field report on AI-assisted building in mid-2026: good enough that non-coders ship daily-use tools, still fragile, insecure, and hard to share. Directly relevant to who gets to build software. - **Stakeholders.** Non-engineer builders, AI-tool vendors (Anthropic, OpenAI), platform owners whose gaps get patched (Gmail, calendar/to-do apps), and the open-source community that may absorb demand for a "trusted core app." - **Risks/tradeoffs.** Vulnerabilities builders can't evaluate, no safe sharing path, throwaway output ("token waste"), and a personalization paradox: the better an app fits you, the less it fits anyone else. - **Bottom line.** Building is now the easy part; finding a useful idea, making output safe, and making it shareable are the unsolved problems. ## 📖 Main Content Core argument: vibe coding crossed a usability threshold between "last year" (it "didn't work") and "this winter" (it suddenly did). Tech advanced faster than the builders' ability to find worthwhile problems. Jake's thesis: finding an idea useful to *you specifically* is now harder than building it. Each demos a "toothbrush test" app (used twice daily): Jake's per-meeting notes app that auto-rolls action items forward, plus a homemade Gmail-label email client; Hayden's ADHD habit tracker as web *and* native mobile app; David's unified calendar+to-do view and a browser web-clipper. All solve hyper-specific personal annoyances the market hasn't fit. Shared pain points: apps are **insecure** (Hayden's ~15 high + 15 medium vulnerabilities; agents said don't worry), **fragile** (David's planner hijacks every OS keyboard shortcut; Jake's flaky weekly localhost login), and **unshareable** (hard-coded to one workflow). They tolerate bugs they'd call "reprehensible" in others' software. Tooling: Codex had better design taste but confusing guidance; Claude Code solved problems Hayden couldn't conceptualize but had "unsophisticated design sense by default." **Transcription caveats (flagged, not real):** "flagship podcast of the Vercel REST API" ≈ Vox Media Podcast Network; "Comcast spun off in 2024" garbles Comcast's spinoff (Versant); "Mythos"/"Fable 5" garble Anthropic model names in a satirical segment; "Cloud Code" = Claude Code; "live coding" = vibe coding; "X spoke" = Xcode; "Neelay" = Nilay Patel. ### 🗒️ Notes #### ⏱️ [00:00:00](https://www.youtube.com/watch?v=4fUMfB-YLGo?t=0) Cold open + news - Group challenge; regroup in four weeks. Satirical news segment (Anthropic release politics, Comcast split, WhatsApp usernames) with ASR-garbled proper nouns — not core to topic. #### ⏱️ [00:05:00](https://www.youtube.com/watch?v=4fUMfB-YLGo?t=300) Last year vs now - Last year "constantly flopping"; by winter "it had changed a lot." Thesis: finding a useful idea is harder than building it. "Enterprise crisis": tokens spent on things that amount to nothing. #### ⏱️ [00:09:00](https://www.youtube.com/watch?v=4fUMfB-YLGo?t=540) Jake's notes app + email client - Notes app auto-rolls action items forward; first success, "like 30 seconds." Gmail-label email client surfaces labels as top tabs Gmail won't pin ("sucky superhuman"); scary Gmail API scopes; "a weekend." #### ⏱️ [00:15:00](https://www.youtube.com/watch?v=4fUMfB-YLGo?t=900) Hayden's habit tracker - ADHD accountability tool with non-daily cadences off-the-shelf apps lacked. "Stars rain down" reward. ~15 high + 15 medium vulnerabilities on both Claude Code and Codex builds. #### ⏱️ [00:22:00](https://www.youtube.com/watch?v=4fUMfB-YLGo?t=1320) David's planner + clipper; sharing - Calendar+to-do view (Google Calendar + Todoist); web clipper, "least RAM ever." Browser extensions = safest on-ramp. Unsolved: safe/durable/shareable apps; floats an open-source "trusted core." #### ⏱️ [00:28:00](https://www.youtube.com/watch?v=4fUMfB-YLGo?t=1680) Claude Code vs Codex + the task - Codex: better taste, confusing, version breakage. Claude Code: better problem-solving/previews, weak default design. Task: vibe code a personal website; security not judged "at least not yet." ## 📚 References - [The Vergecast](https://www.theverge.com/the-vergecast) · [Claude Code (Anthropic)](https://www.anthropic.com) · [OpenAI Codex](https://openai.com) - [Todoist](https://todoist.com) · [Gmail API](https://developers.google.com/gmail/api) · [Apple Xcode](https://developer.apple.com/xcode/) ## 🗣️ Paraphrase - Three non-engineers each built tiny apps they now use daily; the recurring lesson is the AI will build almost anything you describe, so the bottleneck moved from "can I build it" to "is this worth building." - The apps work *because* they're tuned to one person, which is exactly why they can't be shared: insecure, brittle, idiosyncratic. The panel agrees nobody has solved making personal vibe-coded software safe and durable enough to give away. ## 📰 Newsletter **Vibe coding grew up over the winter (the bugs did not).** Three Verge journalists who can't read a stack trace now ship daily-use apps — a meeting-notes tool, an ADHD habit tracker, a homemade Gmail client — by *describing* them to Claude Code and Codex. For the WordPress crowd this is the democratization story in miniature: the build step is nearly free, but the hard parts (a useful idea, security you can actually evaluate, something safe to share) are exactly what our ecosystem already does well. When a reporter ships 30 vulnerabilities and the AI shrugs "don't worry about those," that's a curriculum gap and a plugin-review opportunity. Worth a Learn WordPress lesson on "you vibe coded it, now harden it" — and a reminder that the "trusted open-source core app" these folks wish existed is kind of what our plugin directory is for. [Watch](https://www.youtube.com/watch?v=4fUMfB-YLGo). ## ✅ Action Items #### DevRel - Advocacy: a "vibe code it, then harden it" talk/demo for non-engineers shipping personal tools. - Education: workshop on reading an AI security report ("high vs medium" when the agent says "don't worry"). - Developer experience: ship a "trusted core" starter template (notes/to-do/clipper) the panel wishes existed. - Community: feedback channel for non-engineer builders entering via AI tools. #### OSPO - Policy & Governance: stance on AI-generated contributions and provenance. - Security Management: intake path for "vibe coded, never reviewed" submissions; signing/attestation. - Developer Support: office hours for hardening AI-generated apps. - Software Inventory / Community Engagement / Strategic Alignment: N/A — not in source. #### Community - Maintainers: publish a "trusted community core app" reference (the panel's open wish). - Contributors: "first issue" rotation for AI-assisted newcomers. Ecosystem: safe-sharing guidance. - Sponsors: N/A — not in scope. #### WordPress - Training: hardening/sharing vibe-coded apps; AI tool design-taste limits. - Plugins: review-team note on AI-generated submissions + provenance checks. - Playground: "vibe code a block plugin, then audit it" lab. Documentation: examples-first patterns. - Accessibility: AI defaults ("stars rain down") may ignore prefers-reduced-motion. - Core / Performance / Core AI / Core Program Team / Mobile / Photos / TV / Polyglots / Meta / CLI / Hosting / Tide / Openverse / Themes / Design / Support / Marketing: N/A — no hook in this consumer-tooling episode. #### The WPCC - N/A — no funding/fellowship content in source. ### FAIR - N/A — no metadata-independence or governance-capture content in source. ## 🧪 Hypotheses - H1: As AI build cost nears zero, the binding constraint shifts from implementation skill to problem-finding. - H2: Personal usefulness and shareability are inversely correlated for vibe-coded apps. - H3: Non-engineers under-weight security warnings because they can't evaluate them and the agent minimizes them. - H4: A "trusted open-source core" (notes/to-do/email scaffolds) would materially increase safe reuse. - H5: AI coding tools differ measurably on a taste-vs-problem-solving axis at a given snapshot. ## 📊 Methodology - H1/H2: build-diary study — non-engineers log idea-time vs build-time, rate usefulness + willingness-to-share per app. - H3: controlled task — participants self-rate severity of an AI security report vs expert audit. - H4: A/B — scaffold-provided vs scratch cohorts on completion, vulnerability counts, reuse. - H5: blind side-by-side prompts across tools, rated by independent designers (taste) and engineers (correctness). ### ❓ Questions - What's the liability/disclosure model when a non-engineer publicly shares an insecure app? - Could plugin directories offer "audited vibe-coded" tiers? - Does AI-assisted building widen or narrow access for non-coders over time? - What design-system defaults fix "unsophisticated by default" without removing user control? ### 🔮 Future Research - Longitudinal retention of vibe-coded "toothbrush test" apps (do they survive a model upgrade?). - Security-literacy interventions for non-engineer builders. - Standard "safe share" packaging (sandboxing, secrets, backups) for personal apps. - Open-source "trusted core" reference implementations and adoption. ### 💡 Implications - **Practice:** advocacy/docs should target idea-finding and hardening, not just "you can build now." - **Policy:** review bodies need provenance and security stances for AI-generated submissions. - **Security:** real gap — builders ship and dismiss vulnerabilities they can't read. - **Education:** demand for "harden and share" curricula; reading an AI report is a teachable skill. - **Funding:** opportunity for sponsored "trusted core" community projects. ### 🧠 Learning Styles Assessment - Reading level: conversational spoken English; Flesch–Kincaid Grade ~6–8 (short clauses, filler, first-person anecdote). #### 👀 Visual learners - Present: none (audio only). Improve: app architecture sketches, before/after screenshots, a tool-comparison table. #### 🔊 Aural learners - Present: full podcast audio. Improve: 3-minute audio recap of the three builds and the tool verdict. #### 🗣️ Verbal learners - Present: rich analogies ("toothbrush test," "sucky superhuman," "describing a dream"). Improve: glossary (vibe coding, localhost, API scope). #### 🧩 Kinesthetic learners - Present: implicit "go build a browser extension" call. Improve: guided first-project lab with checkpoints. #### 🧠 Logical learners - Present: implicit cost-shift argument (build cheap, ideas scarce). Improve: "should I share this app?" decision tree + vulnerability triage flow. #### 🧑‍🤝‍🧑 Social learners - Present: three-way discussion, listener hotline/email. Improve: community thread to swap personal-app ideas safely. #### 🧘‍♀️ Solitary learners - Present: self-directed "build for yourself" ethos. Improve: reflection prompts on your own twice-daily annoyance. #### 🌿 Naturalistic learners - Present: none. Improve: token/energy waste ("all these tokens for nothing") as a sustainability lens. ### 📊 Accessibility Assessment #### Screen reader semantics - Present: auto-transcript. Improve: human-corrected transcript with speaker labels. #### Keyboard navigation - Present: N/A (media). Improve: note the irony — David's app *captures all keyboard shortcuts*, a real keyboard-trap anti-pattern worth teaching. #### Contrast and readability - Present: N/A for audio. Improve: published show notes should meet 4.5:1. #### Text resizing and reflow - Present: N/A. Improve: ensure transcript page reflows at 200%. #### Media and captions - Present: auto-captions. Improve: corrected captions (ASR garbles brands), downloadable transcript, no autoplay. #### Forms and interactions - Present: N/A. Improve: the "stars rain down" reward should respect prefers-reduced-motion. #### Standards reference - WCAG 2.2 AA: 1.2.2 Captions, 1.2.1 Audio-only (transcript), 2.1.2 No Keyboard Trap (shortcut-capture bug), 2.3.3 Animation from Interactions (reward). ### 🌍 DEIB Assessment #### Cultural breadth - Present: US, English-only, single-newsroom view. Improve: builders outside tech-media and non-US contexts. #### Inclusive language - Present: casual, mostly self-defined jargon. Improve: define API/localhost/Xcode for general audiences. #### Neurodiversity support - Present: strong — Hayden centers an ADHD use case and accountability design without stigma. Improve: generalize "build tools around how your brain works." #### Representation - Present: mixed-gender panel; a senior woman AI reporter as lead builder. Improve: feature non-journalist, non-affluent builders to avoid a "$20/month and free time" skew. #### Psychological safety - Present: high — hosts model vulnerability ("we are all vastly in over our heads"). Improve: pair the "don't worry about vulnerabilities" candor with a safety caveat. ### ✅ Logical Validity #### Strong reasoning - Inductive (strong): three independent builders converge on the same pattern (easy to build, hard to find a worthwhile idea, impossible to safely share). - Causal (plausible): cheap building → flood of low-value apps → "enterprise crisis" of wasted tokens; mechanism stated. - Analogical (clarifying): "toothbrush test" and "describing a dream" bound what counts as a good personal app. #### Fallacies - Hasty generalization: "this is becoming like a enterprise crisis" extrapolates from anecdote + vague "the data." Fix: cite actual studies. - Anecdotal evidence: tool verdicts are n=1 per tool at one moment. Fix: blind, repeated trials. - Normalization/appeal to own authority: "anyone else vibe coding would be like me and just be like, oh, it's no big deal" normalizes ignoring security. Fix: separate "what people do" from "what's safe." #### Tone of voice - Self-deprecating, playful, candid ("chief shenanigans reporter," "Get the stars out of my face"); earnest about limits. ### 🛡️ DARVO Analysis - N/A — no accusation/defense dynamic. No denial, attack-on-accuser, or victim/offender reversal; the discussion is cooperative self-reporting, so DARVO doesn't apply. ## 🎯 Conclusion - Bottom line: vibe coding now reliably produces useful personal software for non-engineers, relocating the hard problems to idea quality, security literacy, and shareability — none solved. - Next steps for readers: pick one twice-daily annoyance and build the smallest fix (start with a browser extension); always ask the agent for text-file backups; treat dismissed security warnings as a to-do; watch the inverse relationship between "perfect for me" and "shareable" before handing your app to anyone. ## 🔄 Refactor Appendix — Our vibe coded projects that actually work | The Vergecast > Density: 2–5 Literature atoms (H3), 0–3 Permanent candidates (H4). Resonance filter: if you can't say in one line why it matters, drop it. ### 📚 Literature Atoms (H3) #### [H1] Finding a useful idea now beats building it > [!quote] ID:rw_h1 > "finding an idea that is useful to you specifically is harder than building it now, which is is nice." **Context & use**: - ~[00:07](https://www.youtube.com/watch?v=4fUMfB-YLGo?t=420), Jake Kastrenakes. - Central thesis: AI moved the bottleneck from implementation to problem-finding. **Tags**: #xref #articleseed **Highlight**: [view highlight](readwise://)  #### [H2] Vibe coders ship lots of useless throwaway output > [!quote] ID:rw_h2 > "you're going to make a lot of stuff that in the moment you're like this I'm cooking. Like this this is amazing. And then you use it and you're like I have no use for this. Like this is total garbage." **Context & use**: - ~[00:06](https://www.youtube.com/watch?v=4fUMfB-YLGo?t=360), Jake. - The "token waste / enterprise crisis" pattern, observed from inside. **Tags**: #xref **Highlight**: [view highlight](readwise://)  #### [H3] Simpler problems make the best vibe-coded apps > [!quote] ID:rw_h3 > "With vibe coding, I feel like the simpler the better. Like it's usually solving a problem that like is really easy, but you just can't get it together." **Context & use**: - ~[00:10](https://www.youtube.com/watch?v=4fUMfB-YLGo?t=600), Hayden Field. - Matches David's "toothbrush test"; scope discipline predicts success. **Tags**: #xref #aphorisms **Highlight**: [view highlight](readwise://)  #### [H4] Non-engineers ship and then dismiss real vulnerabilities > [!quote] ID:rw_h4 > "it said that I had like 15 high security vulnerabilities and 15 medium security vulnerabilities. So I'm not a developer, so I don't know what they are." **Context & use**: - ~[00:18](https://www.youtube.com/watch?v=4fUMfB-YLGo?t=1080), Hayden; agents said "don't worry about those." - The security-literacy gap DevRel/OSPO can address. **Tags**: #xref #articleseed **Highlight**: [view highlight](readwise://)  #### [H5] Browser extensions are the safest beginner on-ramp > [!quote] ID:rw_h5 > "For anybody starting with live coding, by the way, I really recommend starting with browser extensions, because they're very simple to make." **Context & use**: - ~[00:24](https://www.youtube.com/watch?v=4fUMfB-YLGo?t=1440), David Pierce ("live coding" = vibe coding, ASR slip). - Concrete, low-blast-radius first project for teaching. **Tags**: #xref #share **Highlight**: [view highlight](readwise://)  ### 🧠 Permanent Candidates (H4) #### [P1] The build-cost collapse moves the bottleneck to idea quality **Claim** (Courtney's voice): When AI makes implementation nearly free, the scarce resource becomes finding a problem worth solving, not the skill to solve it. **Evidence pointer**: [[#H1]] **Scope & conditions**: Holds for small personal/utility software; breaks where correctness, scale, or security dominate cost. **Links**: - [[Software democratization]] — because extends - [[Idea-finding as the new literacy]] — because applies-in DevRel education - [[AI token waste in enterprises]] — because contrasts on cost-locus  #### [P2] Personal fit and shareability trade off in vibe-coded software **Claim** (Courtney's voice): The tighter a vibe-coded app fits one person's workflow, the less safely transferable it is to anyone else. **Evidence pointer**: [[#H4]] **Scope & conditions**: Holds while apps are hard-coded and unaudited; weakens if a trusted, configurable open-source core exists. **Links**: - [[Open-source trusted core apps]] — because applies-in safe reuse - [[Security literacy for non-engineers]] — because evidence-for  ### ✏️ Fleeting Queue - [F1] Draft a Learn WordPress lesson: "You vibe coded it — now harden and share it." — #fleeting #inbox - [F2] Keyboard-trap teaching example: David's planner captures all OS shortcuts (WCAG 2.1.2). — #fleeting #inbox - [F3] Could a community "trusted core" notes/to-do/email scaffold answer the panel's open wish? — #fleeting #inbox ### → Things 3 (max 2 actions) - [ ] Outline a "vibe code, then harden" non-engineer workshop for DevRel — #things - [ ] Write up the build-cost-collapse → idea-scarcity thesis as an article seed — #things  📄 **Full text:** [[Full Document Contents/Articles/Our Vibe Coded Projects That Actually Work the Vergecast.md|Our Vibe Coded Projects That Actually Work | the Vergecast by youtube.com]] ## Highlights - you're like this I'm cooking like this this is amazing and then you use it and you're like I have no use for this like this is total garbage. Um so I I made a bunch of tools that I thought would be useful for work. Different kinds of news digests different kinds of things that would you know pull together Twitter feeds. things that would I I just one time I told Claude I was like hey find every single you know open US government ([View Highlight](https://read.readwise.io/read/01kwazh6v2p4awym0yh5jacp6m)) - Note: The throwaway pattern: "I'm cooking... this is amazing. And then you use it and you're like I have no use for this... total garbage." Building feels productive; usefulness is the filter. - Tags: prototyping, atom/literature/h2 - harder than building it now, which is is nice. This is I mean this is like increasingly supported by the data by the way, right? That like you look out into the vibe coding world and everybody is building tons of stuff that amounts to nothing. And this is becoming like an enterprise crisis because everybody's like, "We're using all these tokens for nothing. What is the point of any of this money that we're spending?" So you've you've encountered the same problem that every Fortune 500 company is happening in spades right now. >> Except I only paid like $20 a month. So it was not it was not so bad. Um ([View Highlight](https://read.readwise.io/read/01kwazh0y0xp7e903yrbgr4yxm)) - Note: Thesis + the crisis: idea-finding is now harder than building, and "everybody is building tons of stuff that amounts to nothing" — the same waste hitting Fortune 500s. - Tags: atom/literature/h1, vibe-coding - that. Uh I want to know like does does Jake's description of what he vibe coded do anything for you? >> Yeah, I mean I don't know with vibe coding I feel like the simpler the better. Like it's usually solving a problem that like is really easy but you just can't get it together. That's how mine was at least what I like built. And so this seemed similar. It's like, yeah, he could, you know, take the notes and look back at them, but he's not. So, he's like figuring out a small tiny fix that's going to make him look. So, I feel like, yeah, I mean, it seems like ([View Highlight](https://read.readwise.io/read/01kwazhf5513j7e1ew9499jmkk)) - Note: Zettel: "with vibe coding the simpler the better — usually solving a problem that is really easy but you just can't get it together." The sweet spot is small personal friction, not ambitious builds. - Tags: vibe-coding, atom/zettel/z1 - which is like working out and create a project or some sort cuz I'm starting to try to like learn how to embroider and stuff like that. >> And then only every other day does it pop up to say like, "Oh, take your iron." And that's another habit that appears. So basically, I made it perfectly for myself. And I think this is actually something that other people would want to use, but I will say you guys, it said that I had like 15 high security vulnerabilities and 15 medium security vulnerabilities. >> So I'm not a developer, so I don't know ([View Highlight](https://read.readwise.io/read/01kwazhp4jrryc8zj826bapb8c)) - Note: The security-literacy gap: "15 high security vulnerabilities and 15 medium... I'm not a developer, so I don't know what they are." Democratized building outruns security understanding — a live WP plugin-review/Learn concern. - Tags: ai-and-work, atom/literature/h1, developer-tools - extensions because they're very simple to make. They're really easy to load. Um, you can obviously do lots of like security disaster things, but like the the there are fewer horrible ways for it to go wrong when you're just trying to like do a simple thing with a web page on your computer. Highly recommend messing around with Chrome extensions to solve some of your browser problems. The email client you described, Jake, all of a sudden like has my has my brain spinning. I'm like, this is this is a thing that I want. I have gone back into ([View Highlight](https://read.readwise.io/read/01kwazhw9d0s1rf4xnza6kvq38)) - Note: Practical on-ramp: start with browser/Chrome extensions — "fewer horrible ways for it to go wrong when you're just trying to do a simple thing with a web page." - Tags: atom/literature/h3, developer-tools