No Code MBA

Thank you to Rork for sponsoring this video. Try Rork here: https://rork.com/?ref=nocodemba In this video, I'm showing you how to use the Rork Max app on your iPhone — not your computer — to build real iOS apps. We're building a mood tracking app from scratch and I'll show you something that genuinely surprised me: building native iOS widgets with a vibe coding tool. I had no idea that was even possible. By the end of this video, you'll feel confident to build your own iOS apps using Rork. ━━━━━━━━━━━━━━━━━━━━━━ GETTING STARTED WITH RORK MAX: ━━━━━━━━━━━━━━━━━━━━━━ Setup is straightforward. Download the Rork Max app, log in with your Rork account and Apple ID, then connect your iPhone via USB. Once installed, you can start building apps directly from your phone — literally from anywhere. You can build iPhone apps in Swift or web apps. There's a plan mode toggle, the ability to paste images or take photos with your camera, and you can choose from multiple AI models or just let Rork pick automatically. One cool thing — you can multitask while it's building. You don't have to stay on the builder page. You can also invite collaborators so you're not building alone. ━━━━━━━━━━━━━━━━━━━━━━ WHAT WE BUILT: ━━━━━━━━━━━━━━━━━━━━━━ With a single prompt ("create a mood tracking app"), Rork came up with everything on its own — app name "Hue," a pigment and ink aesthetic, a Today page, journal, patterns view, seven-day pulse, streaks, and even a matching app icon. All data persists locally with Swift Data. The design quality right off the bat is really polished. Whatever Rork Max has in the backend — whether it's templates, examples, or just strong models — the result looks great on the first pass. It does take longer than other AI app builders, but the tradeoff is a better first result with less back and forth. ━━━━━━━━━━━━━━━━━━━━━━ THE WIDGET (THIS IS THE COOL PART): ━━━━━━━━━━━━━━━━━━━━━━ After the initial build, I asked Rork to build a home screen widget showing your streak. It built four widget sizes — small, medium, large, and lock screen — all connected to the app. I installed the app directly on my phone, searched for "Moodley" in the widget picker, and added it to my home screen. It actually works and is connected to the real app data. This is a native iOS feature built entirely through a vibe coding tool — that's what makes Rork Max special. ━━━━━━━━━━━━━━━━━━━━━━ WHAT MAKES RORK MAX DIFFERENT: ━━━━━━━━━━━━━━━━━━━━━━ Because Rork builds in Swift for iOS, it can implement native Apple features that other AI app builders can't touch: - Home screen and lock screen widgets - AR and LiDAR functionality - 3D games - NFC capabilities - Full Apple ecosystem integration You can also install apps directly on your device without going through the App Store — great for personal use or testing. And when you're ready, you can submit to the App Store or TestFlight in just two clicks through the web interface. ━━━━━━━━━━━━━━━━━━━━━━ PRICING: ━━━━━━━━━━━━━━━━━━━━━━ Rork Max starts at $200/month with 1,000 credits. You can add more credits if needed. Compared to hiring a developer to build a custom iOS app (thousands of dollars), it's actually really reasonable for what you get. ━━━━━━━━━━━━━━━━━━━━━━ MY TAKE: ━━━━━━━━━━━━━━━━━━━━━━ The fact that you can build a real, installable iOS app with native features like widgets — all from your iPhone — is genuinely impressive. The design quality is strong, the first results are polished, and the ability to go from idea to installed app on your phone in minutes is something I haven't seen from other tools. Even if you never put anything in the App Store, being able to create custom apps just for yourself on your own iPhone is pretty cool on its own. ⏰ TIMESTAMPS: 0:00 - Introduction & what we're building 1:47 - Rork Max app walkthrough on iPhone 4:02 - Pricing ($200/month for 1,000 credits) 6:14 - Testing the app and design review 8:01 - App Store submission in two clicks 10:21 - Final thoughts 🔗 Try Rork Max: Link in description #rork #iosapp #vibe coding #mobileapp #swift #sponsored

OpenAI just released GPT Image 2.5, its most powerful image model yet, claiming it's faster, sharper, and smarter. So I put it head to head against GPT Image 2 with real prompts, comparing image quality, generation time, and cost on every single test. The quality improvement is real but modest. The speed and cost difference is not close. → Stay up to date with all things AI with our daily newsletter: https://nocode.mba/newsletter?refer=youtube-video/wsHXCmcPUR0 Related videos: → 10 AI trends changing everything: https://youtu.be/pMkMuejz9ZA → Master 90% of Codex in 10 minutes: https://youtu.be/Xly3hFrFulE -------------------- One thing to know before you start: via the API there are two versions of the model, and picking the wrong one will cost you either quality or money depending on your use case. FLARE VS SUNBURST • GPT Image 2.5 Flare - the smaller model optimized for speed. Image quality is comparable to the old GPT Image 2, but faster and cheaper than anything else here. • GPT Image 2.5 Sunburst - the base model, and where the real quality gains are. This is what I test against GPT Image 2 throughout the video. -------------------- THE SIDE-BY-SIDE TESTS 1. SIMPLE PROMPT: A BLUE CERAMIC CUP The headline numbers: 14.6 seconds on the new model versus 39.7 seconds on the old one, and 1.3 cents versus 5.3 cents per image. Generation time cut by well over half, cost cut by roughly 4x. Quality is better and more detailed, though not dramatically so on a prompt this simple. 2. EDITING AN EXISTING IMAGE "Add a green leaf beside the cup." Same pattern on speed and cost, and the leaf itself simply looks better on the new model. Worth testing separately, since editing is a different job from generating from scratch. 3. A DETAILED PROMPT WITH TEXT IN IT Here the gap widens: noticeably better lighting, more detail, and a more realistic result. Then an edit that adds a strawberry while keeping every existing object, letter, and the camera angle unchanged. 4. PHOTOREALISTIC GOLDEN RETRIEVER A specific scene (red front door, small potted fern) that comes out sharper with better lighting on 2.5. Then the real test: have the dog hold a newspaper with the headline "Good Dog Daily." Not only does the headline render correctly, but even the small body text below it looks like real text, which is exactly where weaker image models fall apart. -------------------- KEY TAKEAWAYS • Cost and speed are the real upgrade here, not a night-and-day quality jump • Roughly 4x cheaper and nearly 3x faster, consistently across every test • Choose Sunburst for quality, Flare if you want old-model quality for even less • Text rendering is now genuinely good, down to the small print • The gap shows up most on detailed prompts, less so on simple ones -------------------- CHAPTERS 0:00 OpenAI's new GPT Image 2.5 model 0:39 Flare vs Sunburst: which one to use 1:23 Test 1: the blue cup (14.6s vs 39.7s) 2:24 Editing an image with the new model 3:18 Test 2: a detailed prompt with text 4:08 Test 3: golden retriever + the text test -------------------- Do you agree the new images look better? Leave a comment, I'd be curious if you see the same difference I do. #openai #gptimage #aiimages #imagegeneration #chatgpt #aitools #ai #nocode

Anthropic says Fable 5.1 is the best AI model available right now, and the hype has been loud. So Will, founder of Dyad and an AI coding expert, joins me to cut through it. He built an original benchmark you won't find anywhere else: instead of abstract coding puzzles, it has models build three real business apps (a CRM, a help desk, and a SaaS admin portal) with authentication, user roles, and permissions. The results are not what the headlines suggest. Check out Will's channel here: https://www.youtube.com/@dyad-sh Stay up to date with all things AI with our daily newsletter → https://nocode.mba/newsletter?refer=youtube-video-k-oH85RJcHE The short version: Fable 5.1 is a genuinely strong model, but for most everyday work you probably don't need it. The interesting question isn't which model is best, it's which model is right for the task in front of you. ━━━━━━━━━━━━━━━━━━━━━━ WHAT WE COVER: ━━━━━━━━━━━━━━━━━━━━━━ 1️⃣ WHAT'S ACTUALLY NEW Cheaper cache rates, with Anthropic claiming up to 45% savings on agentic work. True in Will's testing, but it's a slightly less expensive version of an already very expensive model. Plus the anecdote making the rounds: an investment firm's software bug that went unfound for years, which no other model could catch and Fable 5.1 did. 2️⃣ SECURITY SCANS ACTUALLY WORK NOW Fable 5 constantly refused security requests, a persistent complaint from Dyad users and the reason I'd been using GPT 5.6 Sol for security work instead. Anthropic says 5.1 blocks legitimate defensive work far less often. 3️⃣ THE CLAUDISMS ARE GONE Side by side on the same prompt, Fable 5 produces the strange phrasing, the "two honest caveats" structures, and em dashes everywhere. Fable 5.1 just reads like a person wrote it, with no em dashes by default, which neither of us has ever seen from an AI model. Big implications if you write content or SEO articles with AI. 4️⃣ CURSOR BENCH, READ CAREFULLY Grok 4.6 tops Cursor's own benchmark, which is unsurprising given Cursor's data trained it. Fable 5.1 beats Opus 5 on quality-per-dollar despite double the per-token cost, because it needs far fewer tokens. The missing variable nobody discloses: what the tasks actually are. 5️⃣ DYAD'S NEW BENCHMARK (BUILT FOR THIS VIDEO) Three real apps, transparent tasks, with video of a user clicking through each result, including whether a view-only invited user is correctly locked out of records. This is exactly where cheaper models tend to break: authentication and permissions. 6️⃣ THE COST RESULTS All the top models land in roughly the same tier on quality. The gap is price: Fable 5.1 cost $47 to build all three apps, versus $24 for Opus 5 and $20 for GPT 5.6 Sol, and 5.1 took noticeably longer while burning more tokens. All run at medium reasoning effort for a fair comparison. 7️⃣ THE VERDICT AND MODEL CHOICE Reach for 5.1 on genuinely hard problems, or if you have a Claude subscription with usage to spare. Paying API costs? Cheaper models give you very similar results. Also: why no single model won every app, why power users still want to pick their own model, and the delegation pattern Will is exploring where a strong model oversees cheaper models doing the simple work. 8️⃣ DESIGN, WHERE FABLE STILL LEADS Fable 5.1 sits well clear of the field on head-to-head web design ELO (1765, with open-source Qwen 3.8 at 1688). But we get into AI design tells, like the two-tone headline giveaway, and why feeding it two or three designs you love beats cloning one site. ━━━━━━━━━━━━━━━━━━━━━━ KEY TAKEAWAYS: ━━━━━━━━━━━━━━━━━━━━━━ 💡 Quality is converging at the top; price is where models still differ wildly 💡 Cost per task beats cost per token — a smarter model can be cheaper by using fewer of them 💡 Benchmarks built by a model's own company deserve extra scrutiny 💡 No model won every app, which is the case for keeping model choice in your tools 💡 Fable 5.1's writing quality may matter more than its coding scores if you publish content ⏰ TIMESTAMPS: 0:00 Is Fable 5.1 actually worth it? 03:23 The Claudisms are gone 05:53 Cursor Bench — why Grok 4.6 looks so good 09:26 Dyad's new benchmark: three real-world apps 12:05 The cost results — $47 vs $20 for the same work 15:07 The verdict 17:39 Should tools let you pick your own model? 19:08 Design — where Fable 5.1 still crushes #fable5 #anthropic #claude #aicoding #vibecoding #benchmarks #dyad #aimodels

Thank you to Softr for sponsoring this video. Try Softr here: https://softr.io/build-with/no-code-mba?utm_source=seth&utm_medium=influencer&utm_campaign=ai_co-builder&utm_content=nocodemba_june_campaign_yt_video2 Bending Spoons just agreed to buy Airtable for nearly $1.3 billion — and if you watched what happened to Evernote and Harvest after their acquisitions, you understand why people are nervous. But migrating off Airtable immediately might be the wrong move. In this video: when to stay, when leaving actually makes sense, and the easiest path out if you decide to go — including a live Airtable → Softr migration. Stay up to date with all things AI with our daily newsletter → https://nocode.mba/newsletter?refer=youtube-video/IVgmFakWWzo ━━━━━━━━━━━━━━━━━━━━━━ THE THREE SCENARIOS: ━━━━━━━━━━━━━━━━━━━━━━ 1️⃣ YOU'RE HAPPY WITH AIRTABLE — DON'T PANIC The acquisition hasn't closed, so Airtable is still operating independently: no price changes, no feature changes, nothing different today. Migrating has real switching costs — rebuilding automations, moving data, retraining your team. Instead: export your bases as a backup, check what you're paying per seat, and model what a price increase would actually cost you. 2️⃣ YOU WERE ALREADY THINKING OF LEAVING — NOW'S THE TIME If per-collaborator pricing, record or API limits, slow bases, expensive external access, or limited interface customization were already pain points, move before you're forced to. Precedent: after Bending Spoons acquired Harvest, some customers reported price increases of over 1500%. 3️⃣ THE HYBRID PATH (WHY SOFTR IS DIFFERENT) This doesn't have to be all-or-nothing. Keep your data in Airtable and build the Softr front end on top of it now, then move the database later if pricing forces your hand. ━━━━━━━━━━━━━━━━━━━━━━ THE MIGRATION (LIVE IN THE VIDEO): ━━━━━━━━━━━━━━━━━━━━━━ Create → Database → Import data → Airtable → connect your account → import. That's it. It brings over your tables, records, standard fields, linked records, attachments, and most formulas, without changing anything in Airtable. You can also import from CSV or Google Sheets, or connect SQL, HubSpot, and Notion as data sources. ⚠️ Before you fully commit: verify your formulas, automations, and scripts all came across correctly — and talk to the Softr team if your setup is complex. ━━━━━━━━━━━━━━━━━━━━━━ WHY FULLY MIGRATE: ━━━━━━━━━━━━━━━━━━━━━━ No API bottleneck — Airtable caps apps at 5 requests per second per base; Softr databases run natively, so that limit disappears. Self-service plans support up to 1 million records. Real branded portals — Airtable interfaces are built for basic internal views; Softr gives you full-stack branded apps, custom domains, and white-labeling, with AI app building on top. Granular permissions — user groups with access control at the page, block, action, and record level. Cheaper external access — Softr separates app-user access from app-building access, instead of paid collaborator seats or a Portals add-on. Better forms + visible workflows — multi-step forms with conditional logic, and if-then-else workflows you can actually see (not a black box) triggered by data changes, schedules, or webhooks. ━━━━━━━━━━━━━━━━━━━━━━ KEY TAKEAWAYS: ━━━━━━━━━━━━━━━━━━━━━━ 💡 Nothing has changed at Airtable yet — but export a backup today regardless 💡 The Harvest precedent is why "wait and see" shouldn't mean "do nothing" 💡 You can move the interface first and the data later — it isn't all-or-nothing 💡 Fully migrating is what removes Airtable's 5-requests-per-second API ceiling 💡 Test alternatives now so a price increase never forces a rushed migration ⏰ TIMESTAMPS: 0:00 Bending Spoons is buying Airtable 02:46 When leaving Airtable actually makes sense0 5:01 What Softr is 06:40 Migrating Airtable data into Softr 07:59 API limits, permissions + pricing advantages 12:29 Recap — which scenario are you in? 💬 COMMENT BELOW: Which scenario are you in — staying, watching, or already migrating? #airtable #softr #nocode #bendingspoons #migration #clientportal #airtablealternative #nocodetools

Thank you to Dyad for sponsoring this video. Try Dyad free → https://dyad.sh Most people are leaving serious money on the table when they vibe code. In this video I bring on Will, the founder of Dyad and a former Google engineer, to cover three topics that don't get discussed enough: using open-source and cheaper models to cut your AI costs (one tip alone can save you 10x), how to host your vibe-coded apps without surprise bills, and vibe coding security. At the end, Will honestly explains when NOT to use his own tool — and when Claude Code, Codex, or Lovable is the better choice. Stay up to date with all things AI with our daily newsletter → https://nocode.mba/newsletter?refer=youtube-video/tZQld_jZwOk Everything here applies to any vibe coding tool that lets you pick your model — not just Dyad. ━━━━━━━━━━━━━━━━━━━━━━ WHAT WE COVER: ━━━━━━━━━━━━━━━━━━━━━━ 1️⃣ PICKING MODELS BY PRICE VS PERFORMANCE Will walks through real benchmarks (Cursor Bench, DeepSWE, side-by-side ELO rankings) showing open-weight models within touching distance of the frontier — at up to 100x lower cost. GPT 5.6 Luna after its 5x price cut: ~8 cents a task for 50% of Fable-level performance. 2️⃣ THE EFFORT SETTING MOST PEOPLE IGNORE Thinking harder can 3x your cost for a ~10% quality gain. Match the model AND the effort to the task — restyling buttons doesn't need Fable. The future: routers that let AI pick the right model for you. 3️⃣ TOKEN EFFICIENCY + SUBSCRIPTIONS VS API Cheaper per token isn't always cheaper per task — some models burn more tokens per step. And model choice matters on subscriptions too: cheaper models drain your $20/month plan far slower. 4️⃣ HOSTING WITHOUT SURPRISE BILLS The mental model: front end (Vercel, Netlify, Cloudflare) + backend database (Supabase, Neon). Supabase charges ~$10/month per project once you're off the free tier; Neon's serverless architecture gives you dozens of projects free — Will explains what serverless actually means and when serverful still wins (high traffic, low latency). 5️⃣ VIBE CODING SECURITY Security issues aren't new — even Google teams ship them. AI is now better at catching vulnerabilities in 20 minutes than a human staring at code for 20 hours. Don't skimp on the model for security reviews (Sol or Opus 5 — Fable tends to refuse security questions), and run one before every launch. Plus: why integration tests beat narrow unit tests for vibe coders. 6️⃣ WHEN TO USE DYAD VS EVERYTHING ELSE Dyad is the middle ground: Lovable's easy vibe coding experience with zero lock-in — any model, any cloud, code on your machine, Claude Code in a terminal inside it. Total beginners may prefer Lovable's all-in-one; professional engineers may want Codex. The real rule: never build on a platform you can't export from. ━━━━━━━━━━━━━━━━━━━━━━ KEY TAKEAWAYS: ━━━━━━━━━━━━━━━━━━━━━━ 💡 Matching model + effort to task complexity can cut vibe coding costs 10x 💡 Check cost per task, not cost per token — efficiency varies wildly 💡 Serverless (Neon) can host your side projects for free; serverful wins at scale 💡 Always run a security review with a top model before you publish 💡 Lock-in is the real risk — pick platforms that let you leave ⏰ TIMESTAMPS: 0:00 Three topics nobody discusses enough (with Dyad's founder) 01:56 Model benchmarks — price vs performance 03:55 Effort levels + the 10x savings tip 08:58 Token efficiency, subscriptions vs API costs 10:00 Open-weight models are closing the gap 13:24 Are local models ready yet? 15:38 Hosting — Supabase vs Neon (serverless explained) 21:35 Front end vs backend — Vercel, Netlify + your database 23:21 Vibe coding security — reviews, models + testing 28:40 When to use Dyad vs Claude Code, Codex or Lovable 💬 COMMENT BELOW: Questions for me or Will? Drop them here and we'll get back to you.