"AI Startups" are over done (finally) 🔗 **Source:** ["AI Startups" are over done (finally)](https://www.youtube.com/watch?v=L3vToC1jO64) > [!note]+ Document Notes > tags: [#yc, #ai-startups, #product-market-fit] > - #yc > - #ai > - #product > Author: Theo Browne (YouTube creator; channel often known as “t3”) > Source: YouTube Video > Reference: https://www.youtube.com/watch?v=L3vToC1jO64 > date created: {{date:YYYY-MM-DD}} {{time:HH:mm}} > > # 🧩 Title: “AI Startups” Are Overdone (Finally) — YC’s 2025 Vibe Shift Toward Domain-First Companies > <!-- deep-read:sections start --> ## 🧭 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 (Diversity, Equity, Inclusion, and Belonging) Assessment](#-deib-diversity-equity-inclusion-and-belonging-assessment) - 🔗 [✅ Logical validity](#-logical-validity) - 🔗 [🛡️ DARVO Analysis](#️-darvo-analysis) - 🔗 [🎯 Conclusion](#-conclusion) ### 📚 References: Include a list of sources consulted and cited in the article. --- ## 📝 Summary: - 🔹 The speaker-investor argues that **AI-for-X** startups peaked and that **domain-first** companies now present better opportunities. He cites Y Combinator’s **Summer 2025** batch as evidence of a **vibe shift** toward founders who deeply know a specific industry (e.g., **car washes**) and only then add AI where it fits ([0:38](https://www.youtube.com/watch?v=L3vToC1jO64&t=38), [24:59](https://www.youtube.com/watch?v=L3vToC1jO64&t=1499), [26:39](https://www.youtube.com/watch?v=L3vToC1jO64&t=1599)). - 🔹 He contrasts “**Copilot/Cursor for X**” clones with firms built by **actual practitioners** solving hard, boring, money-making problems; he highlights **Nautilus** (car wash SaaS) and **co-create** (pro video workflows) as emblematic wins ([16:42](https://www.youtube.com/watch?v=L3vToC1jO64&t=1002), [27:26](https://www.youtube.com/watch?v=L3vToC1jO64&t=1646), [34:04](https://www.youtube.com/watch?v=L3vToC1jO64&t=2044)). - 🔹 YC’s selection math changed: **more batches**, **fewer slots**, **far more applicants**, yielding a **~0.14% acceptance rate** and (purportedly) higher quality picks ([11:05](https://www.youtube.com/watch?v=L3vToC1jO64&t=665), [12:56](https://www.youtube.com/watch?v=L3vToC1jO64&t=776), [13:14](https://www.youtube.com/watch?v=L3vToC1jO64&t=794)). - 🔹 He prefers founders who **live the user’s job**; cites Gary Tan’s advice to **“go undercover”** to build empathy and ship **10x** improvements ([22:17](https://www.youtube.com/watch?v=L3vToC1jO64&t=1337), [23:59](https://www.youtube.com/watch?v=L3vToC1jO64&t=1439)). - 🔹 Takeaway: **Know “Thing” > Know AI.** Deep domain expertise + pragmatic AI beats AI maximalism. --- ## 📖 Main Content: - 🔹 **Thesis.** The 2025 YC cohort rewards **operators** over **AI trend-surfers**. Products win when founded by people who know the domain’s workflows, constraints, and revenue levers. AI should assist those workflows, not overwrite them ([20:01](https://www.youtube.com/watch?v=L3vToC1jO64&t=1201), [25:36](https://www.youtube.com/watch?v=L3vToC1jO64&t=1536), [26:39](https://www.youtube.com/watch?v=L3vToC1jO64&t=1599)). - 🔹 **Why the turn.** Application volume skyrocketed while seats shrank. Selection pressure favored teams with **credible user empathy**, **evidence of pull**, and **niche-market mastery** ([11:05](https://www.youtube.com/watch?v=L3vToC1jO64&t=665), [12:56](https://www.youtube.com/watch?v=L3vToC1jO64&t=776)). - 🔹 **Exemplars.** - **Nautilus**: vertical SaaS for **car washes**—memberships, POS, ops, marketing—built by someone who **worked in car washes since 16**. They sell through **industry-native channels** (podcasts, conferences), show strong adoption, and directly **move revenue** ([27:26](https://www.youtube.com/watch?v=L3vToC1jO64&t=1646), [29:08](https://www.youtube.com/watch?v=L3vToC1jO64&t=1748)). - **co-create**: pro video workflow tooling focused on **asset orchestration** (timecode, string-outs, local processing) rather than “auto-edit” hype. It targets **professional editors**, not “aspiring” creators ([34:04](https://www.youtube.com/watch?v=L3vToC1jO64&t=2044), [35:01](https://www.youtube.com/watch?v=L3vToC1jO64&t=2101)). - **Bitrig**: native **SwiftUI** app builder by **SwiftUI’s creators**. Shows why **deep platform knowledge** can beat framework-agnostic AI coding toys ([39:15](https://www.youtube.com/watch?v=L3vToC1jO64&t=2355), [40:15](https://www.youtube.com/watch?v=L3vToC1jO64&t=2415)). - 🔹 **Anti-patterns.** - “**Copilot/Cursor for X**” where founders lack **domain literacy** (e.g., PM tooling by non-PMs; video-editing tools by non-editors) ([18:41](https://www.youtube.com/watch?v=L3vToC1jO64&t=1121), [21:59](https://www.youtube.com/watch?v=L3vToC1jO64&t=1319)). - **Aspiring-user markets** (“people who gave up”) rarely compound; serve **professionals** with real pains and budgets ([37:01](https://www.youtube.com/watch?v=L3vToC1jO64&t=2221), [37:36](https://www.youtube.com/watch?v=L3vToC1jO64&t=2256)). - 🔹 **Investor lens.** Prefer teams with: - Proof they **do the job** or **sat with doers**. - Channels into the niche’s **media and events**. - A path to **price on value** (e.g., doubling ARPU) instead of generic SaaS MRR ([29:08](https://www.youtube.com/watch?v=L3vToC1jO64&t=1748), [32:25](https://www.youtube.com/watch?v=L3vToC1jO64&t=1945)). ### 🗒️ Notes - 🔹 **Sponsor aside & CI gripe**: Faster CI using gaming CPUs; observability matters ([2:19](https://www.youtube.com/watch?v=L3vToC1jO64&t=139)–[3:38](https://www.youtube.com/watch?v=L3vToC1jO64&t=218)). - 🔹 **VC vs YC 101**: Liquidity, risk, equity lockups, and YC as a bootcamp + capital network ([3:40](https://www.youtube.com/watch?v=L3vToC1jO64&t=220)–[6:21](https://www.youtube.com/watch?v=L3vToC1jO64&t=381)). - 🔹 **Returns** (claimed screenshot): Non-YC portfolios top 10% ~3.5x since 2017; YC-focused top 10% ~15x; bottom quartile YC-focused ~3.3x; unicorn incidence ~6.5%/batch; typical demo-day valuation ~$20–25M ([7:13](https://www.youtube.com/watch?v=L3vToC1jO64&t=433)–[10:12](https://www.youtube.com/watch?v=L3vToC1jO64&t=612)). - 🔹 **Admissions shift**: 4 batches/year; 120–140 companies/batch; ~100k apps → ~0.14% pick rate ([11:05](https://www.youtube.com/watch?v=L3vToC1jO64&t=665)–[13:21](https://www.youtube.com/watch?v=L3vToC1jO64&t=801)). - 🔹 **Copilot vs Cursor ethos**: Copilot felt additive for devs; Cursor-style “takeover” can displace the fun, create backlash if builders don’t understand the craft ([16:42](https://www.youtube.com/watch?v=L3vToC1jO64&t=1002)–[20:22](https://www.youtube.com/watch?v=L3vToC1jO64&t=1222)). - 🔹 **User empathy**: Gary Tan on ethnography; do the job, then code the tool ([22:17](https://www.youtube.com/watch?v=L3vToC1jO64&t=1337)–[24:04](https://www.youtube.com/watch?v=L3vToC1jO64&t=1444)). - 🔹 **Case: Nautilus**: founder worked in car washes; vertical suite; industry-native GTM; revenue thesis on memberships and ops automation ([27:26](https://www.youtube.com/watch?v=L3vToC1jO64&t=1646)–[33:07](https://www.youtube.com/watch?v=L3vToC1jO64&t=1987)). - 🔹 **Case: co-create**: pro editors, timecode, string-outs, local compute; solves orchestration, not “auto-editing” ([34:04](https://www.youtube.com/watch?v=L3vToC1jO64&t=2044)–[36:19](https://www.youtube.com/watch?v=L3vToC1jO64&t=2179)). - 🔹 **Case: Bitrig**: SwiftUI-native, built by SwiftUI creators; serious native bar; hedging vs React Native tools ([39:15](https://www.youtube.com/watch?v=L3vToC1jO64&t=2355)–[41:08](https://www.youtube.com/watch?v=L3vToC1jO64&t=2468)). - 🔹 **Dunning–Kruger riff**: YC partner selection risks when partial expertise skews picks, esp. in devtools ([47:29](https://www.youtube.com/watch?v=L3vToC1jO64&t=2849)–[50:05](https://www.youtube.com/watch?v=L3vToC1jO64&t=3005)). --- ## 📖 References: - 🔹 Video: “AI Startups” are over done (finally) — https://www.youtube.com/watch?v=L3vToC1jO64 - 🔹 Y Combinator (general): https://www.ycombinator.com - 🔹 Gary Tan (general): https://x.com/garrytan - 🔹 Learn WordPress (for newsletter context): https://learn.wordpress.org - 🔹 WordPress News: https://wordpress.org/news/ --- ## 🗣️ Paraphrase: - 🔹 **On why Copilot worked for devs** ([16:42](https://www.youtube.com/watch?v=L3vToC1jO64&t=1002)–[17:34](https://www.youtube.com/watch?v=L3vToC1jO64&t=1054)): *Original idea*: Copilot added speed without stealing the craft. It slotted into editors, autocomplete-style, and didn’t try to replace planning or reading code. *Paraphrase*: Copilot succeeded because it **augmented** developer flow. It lived inside tools we already used, trimmed drudgery, and left the thinking and building to us. - 🔹 **On domain-first startup building** ([24:59](https://www.youtube.com/watch?v=L3vToC1jO64&t=1499)–[26:39](https://www.youtube.com/watch?v=L3vToC1jO64&t=1599)): *Original idea*: Many founders knew AI, not the job. Better to master the job and add AI later. *Paraphrase*: The winning play is **work-as-research**. Do the job, absorb the pain points, then use AI like epoxy—only where it bonds and strengthens. --- ## 📰 Newsletter: - 🔹 **Learning web dev:** Start with **MDN** and **Learn WordPress**. Build something small, shippable, and opinionated: https://developer.mozilla.org • https://learn.wordpress.org - 🔹 **Recent updates:** Track **WordPress Core** roadmaps and PHP compatibility notes: https://make.wordpress.org/core - 🔹 **Community highlights:** New contributor sessions are live across Make Teams. Bring a friend; pair programming welcome: https://make.wordpress.org - 🔹 **New plugins to watch:** Browse with a security mindset; read changelogs and recent support threads before install: https://wordpress.org/plugins/ - 🔹 **Big idea:** YC’s 2025 shift favors **domain-first** products. That energy translates to WordPress: solve the gritty workflows (hosting ops, contributor onboarding, nonprofit membership sites) and sprinkle AI where it **saves minutes**, not where it **owns decisions**. --- ## ✅ Action Items: - 🔹 **Reusable prompt for Courtney’s workflow (advocacy + OSPO path):** “Analyze how this article aligns with **open-source advocacy**, **contributor engagement**, and **education** within the WordPress ecosystem. Identify: 1) Which **domain problems** it surfaces that map to WordPress contributor pain (onboarding, mentorship, moderation, funding). 2) Where **evidence-based practices** support **transparent governance** (Code of Conduct, clear decision logs, budgets). 3) Practical **teaching artifacts** I can produce (lesson plans, workshop labs, Git/GitHub exercises, PHP/JS walkthroughs). 4) Steps that advance an **OSPO** function: policy templates, risk registers, contributor safety, metrics. 5) How to ensure **inclusivity** (accessibility, DEIB) in programs and docs, with measurable outcomes.” - 🔹 **Program ideas:** Build a “**Domain-First Builders**” workshop at WordCamps: teams shadow a real site maintainer, document friction, then prototype a plugin or workflow fix. - 🔹 **Funding:** Draft a **micro-grants** rubric for maintenance work; publish outcomes and costs for community trust. - 🔹 **Metrics:** Track onboarding-to-first-PR time, retention at 30/90 days, issue turnaround, and incident rates (moderation/safety). --- ## 🧪 Hypotheses: - 🔹 H1: **Domain-immersion** (shadowing real users) leads to higher **feature adoption** than AI-first ideation in WordPress tooling. - 🔹 H2: Training contributors in **evidence-based triage** (logs, timecode-like metadata for issues) reduces **rework**. - 🔹 H3: **Vertical WordPress solutions** (e.g., membership for car washes, clinics, unions) outperform generic plugins on **LTV** and **NPS**. ### 📊 Methodology: - 🔹 **Design:** A/B cohorts of contributor teams: AI-first vs domain-immersion-first. - 🔹 **Data sources:** GitHub issues/PRs, plugin telemetry (opt-in), support forum tags, surveys. - 🔹 **Analysis:** Diff-in-diff on adoption, retention, and mean time to resolve; mixed-effects models per team. ### ❓ Questions: - 🔹 Where does **AI** remove toil without muting expert judgment? - 🔹 Which **verticals** in WordPress show “Nautilus-style” revenue lift (subscriptions, reminders, ops)? - 🔹 How do we **teach empathy** to plugin authors at scale? ### 🔮 Future Research: - 🔹 Compare **vertical WP stacks** vs **generalist stacks** in total cost of ownership and churn. - 🔹 Study **community moderation** during hype cycles to prevent “aspiring-user traps.” ### 💡 Implications: - 🔹 WordPress wins when builders **live the use case**. AI then becomes **glue**, not the product. - 🔹 OSPO practices (policies, metrics) stabilize contributor ecosystems and funding pitches. --- ## 🧠 Learning Styles Assessment: - 🔹 **Readability (estimate):** Conversational, short sentences, frequent asides → ~Grade 9–10 reading level. - 🔹 **Visual (Spatial):** Minimal visuals in the transcript. ➜ Improve with **timelines**, **system diagrams** (e.g., car wash rev loop). - 🔹 **Aural:** Video format supports auditory learners. ➜ Add podcast cut + chaptered audio. - 🔹 **Verbal (Linguistic):** Rich analogies (Copilot vs Cursor). ➜ Provide **glossary** (ARR, post-money). - 🔹 **Physical (Kinesthetic):** None. ➜ Add **hands-on labs** (shadow a user; map workflow → plugin spec). - 🔹 **Logical (Mathematical):** Some ROI math; few citations. ➜ Add **tables** of returns, acceptance rates. - 🔹 **Social (Interpersonal):** Community stories; YC anecdotes. ➜ Facilitate **group critiques** of startup theses. - 🔹 **Solitary (Intrapersonal):** Offers personal reflection moments. ➜ Provide **journaling prompts** (“Where do I lack domain empathy?”). - 🔹 **Naturalistic:** Not present. ➜ Use **ecosystem** metaphors for open-source maintenance. --- ## 📊 Accessibility Assessment: - 🔹 **Screen readers:** Ensure chapter titles and links have descriptive labels (e.g., “YC acceptance math at 12:56”). - 🔹 **Keyboard navigation:** Provide timestamp hotkeys or a chapter sidebar. - 🔹 **Contrast/Readability:** High-contrast captions; large subtitle fonts. - 🔹 **Text resizing:** Publish transcript with responsive CSS; avoid text baked into images. - 🔹 **Alt text:** Add alt text for any charts (returns, acceptance rates). - 🔹 **Captions/Transcripts:** Keep accurate, punctuated captions; supply downloadable transcript. --- ## 🌍 DEIB (Diversity, Equity, Inclusion, and Belonging) Assessment: - 🔹 **Cultural relevance:** Focus is US/SF-centric startups. ➜ Include global founder stories and markets. - 🔹 **Inclusive language:** Generally direct; occasional profanity. ➜ Offer a profanity-free transcript. - 🔹 **Neurodiversity support:** Dense riffs; many tangents. ➜ Add **chapter markers**, **summary cards**, **checklists**. - 🔹 **Representation:** Case studies skew to male-dominated spaces. ➜ Highlight diverse founders/verticals. - 🔹 **Psychological safety:** Critiques are sharp. ➜ Pair critiques with **clear improvement paths**. --- ## ✅ Logical validity: - 🔹 **Positive reasoning** - **Causal restraint**: The claim is **not** “AI is useless,” but “AI-first without domain mastery underperforms.” This nuance avoids overreach ([24:59](https://www.youtube.com/watch?v=L3vToC1jO64&t=1499)). - **Analogical reasoning**: **Copilot** analogy shows assistive vs replacement framing; clarifies adoption dynamics ([16:42](https://www.youtube.com/watch?v=L3vToC1jO64&t=1002)). - **Ethical/Pragmatic reasoning**: “Build for pros, not aspirants” aligns incentives with real budgets ([37:36](https://www.youtube.com/watch?v=L3vToC1jO64&t=2256)). - 🔹 **Potential fallacies / weaknesses** - **Hasty generalization**: From selected YC anecdotes to a general rule for all AI startups. ➜ Remedy: include broader data sets (market-wide survival rates). - **Selection bias**: Highlights favorites (Nautilus, Bitrig) and weak dev tools; may undercount counterexamples. ➜ Remedy: present **base rates** and failures. - **Appeal to anecdote**: Uses personal meetings and chats as major evidence. ➜ Remedy: add **third-party metrics** and external studies. - **Ad hominem tone** (light): “Clueless” dev tool founders. ➜ Remedy: critique systems/assumptions, not people. - **Bandwagon risk**: Pointing to YC acceptance stringency as a proxy for quality everywhere. ➜ Remedy: separate **selection** from **outcome**. **Tone of voice:** Conversational, blunt, humorous, occasionally abrasive; confident investor-operator stance. --- ## 🛡️ DARVO Analysis: - 🔹 **Deny:** No denial of facts; the speaker acknowledges investment positions and uncertainties. - 🔹 **Attack:** The rhetoric occasionally **attacks competence** of some founders/partners (e.g., “don’t know the space”). This critiques ideas/process, not specific accusers. - 🔹 **Reverse Victim/Offender:** Not present. The speaker does not claim victimhood; he positions himself as a candid observer. - 🔹 **Net:** Minimal DARVO. Strong opinions, but not a denial/attack reversal strategy. --- ### 🔍 SEO Evaluation: - 🔹 **Keywords:** Present implicitly (AI startups, Y Combinator, domain expertise). ➜ Add explicit phrases in title/description: “domain-first startups,” “vertical SaaS,” “YC 2025.” - 🔹 **Meta description:** Add a 155-character blurb: “YC’s 2025 shift: domain-first startups beat AI-first clones. Why operators win and where AI actually helps.” - 🔹 **Title tag:** Keep under 60 chars; include “YC 2025” + “AI startups” + “domain-first.” - 🔹 **Headers:** Add chapter headers in the YouTube description with keywords. - 🔹 **Length:** Video adequate; transcript long enough. - 🔹 **Links:** Add external links to case sites and YC resources in description. - 🔹 **URL:** YouTube URL is fixed; use **chapters** for deep linking. - 🔹 **Images:** Custom thumbnail with readable text; include alt in companion post. - 🔹 **Mobile optimization:** Rely on YouTube; ensure description scannability. - 🔹 **Engagement:** Use chaptered timestamps, pinned comment with summary and resources. - 🔹 **Schema:** If mirrored as a blog, use **VideoObject** + **Article** schema. - 🔹 **Social sharing:** Include Open Graph/Twitter cards on companion post. --- ## 🎯 Conclusion: - 🔹 The piece argues that the market is rewarding **operators who know their niche** and use AI **surgically**. YC’s 2025 cohorts reflect that turn. For builders and advocates in open source and WordPress, the lesson is clear: **shadow real users, fix the unglamorous workflow, and only then add AI**. The boring path is the durable one. <!-- deep-read:sections end -->