The Startup Ideas Podcast

In this episode I sit down with Amir to get tactical about running local AI models as part of a daily workflow. We center on GLM 5.2 from ZAI, how it stacks up against frontier models like Opus 4.8, and how a fusion approach lets you sequence a heavy thinking model with a lighter execution model for the best output at the lowest cost. Amir walks through setup in Cursor and Codex via OpenRouter, shares real token-cost math, and demos GLM 5.2 refining a live app. By the end you will know how to start today, where local models shine, and how model chaining keeps spend in check. Timestamps 00:00 – Intro 02:09 – GLM 5.2 and Z AI 04:01 – Specs: 1M context and Terminal Bench 2.1 05:22 – Making sense of benchmark scores 06:42 – Setup in Cursor or Codex with OpenRouter 10:18 – Local model upside: buy a machine, run tasks 11:42 – Token cost: 44 cents versus $2.38 13:36 – Future-proofing with an upfront hardware bet & The Uber subsidy analogy 16:49 – Model chaining and the vision workaround 19:23 – Token maxing vs routing tasks to the right model 20:54 – Answering the "cost is irrelevant" crowd 21:59 – Closing thoughts Key Points GLM 5.2 ships with a 1M-token context window and scores 81 on Terminal Bench 2.1, landing about four points behind Opus 4.8. A fusion approach (a term OpenRouter coined) sequences models: plan with Opus, execute with GLM 5.2, review with Composer 2.5 or Codex 5.5. Running GLM 5.2 in the cloud through OpenRouter costs roughly 44 cents for a task that runs about $2.38 on Opus 4.8 — close to a 5X saving. You can start today with credit-based access: load $20 in OpenRouter and route tasks to the right model. For images, Amir uses Opus 4.8 to read screenshots and describe them, then hands the layout to GLM 5.2 to act on. Teams are shifting from token-maxing to output-maxing, making model governance and chaining the smart play The #1 tool to find startup ideas/trends - https://www.ideabrowser.com LCA helps Fortune 500s and fast-growing startups build their future - from Warner Music to Fortnite to Dropbox. We turn 'what if' into reality with AI, apps, and next-gen products https://latecheckout.agency/ The Vibe Marketer - Resources for people into vibe marketing/marketing with AI: https://www.thevibemarketer.com/ FIND ME ON SOCIAL X/Twitter: https://twitter.com/gregisenberg Instagram: https://instagram.com/gregisenberg/ LinkedIn: https://www.linkedin.com/in/gisenberg/ FIND AMIR ON SOCIAL Humblytics: https://humblytics.com/?via=community X/Twitter: https://x.com/amirmxt Youtube: https://www.youtube.com/@amirmxt

In this episode I sit down with George Lampropoulos, a 19-year-old founder who turns AI-built mobile apps into real revenue. George walks through his framework for reaching $10K a month—roughly $333 a day—starting with a simple, sellable idea you actually care about and ending with a distribution plan anyone can run. He shares the numbers behind WrestleAI (100K-plus downloads and close to $200K in revenue) and explains why a sharp "gotcha feature" and a clean Instagram funnel do most of the heavy lifting. We also dig into closing influencers, hiring a VA, running paid ads, and reading the metrics that decide whether you grow. If you want a practical, founder-tested playbook for building apps with AI, this one delivers. George’s $10K/mo app playbook: https://startup-ideas-pod.link/George-app-playbook Timestamps 00:00 – Intro 01:54 – George's track record with WrestleAI 02:36 – How AI unlocks fresh app ideas 06:29 – Reverse-engineering a viral idea from your feed 16:16 – Designing the UI/UX of the app 17:49 – The gotcha feature that sells the app 21:25 – Onboarding that converts 23:04 – Actionable Plan to $10k/mo 28:55 – Outreach as a numbers game 33:35 – Paid ads clearly explained 36:20 – Reading metrics: conversion, ARPU, retention 38:30 – TLDR: a great product earns inbound creators 39:51 – Answering the vibe-coding skeptics 39:51 – Scaling with vibe-coded app 43:35 – Why now is the app-building boom 46:05 – Closing Thoughts Key Points I learn why a simple, sellable idea you're passionate about beats pure distribution every time George breaks down the "gotcha feature"—one feature so clear that five seconds explains the whole app We cover a clean Instagram page that doubles as a sales funnel and as social proof for recruiting creators George shares his influencer playbook: lead with relationships, close on a call, and aim for a $2 CPM I get his paid-ads starter method—5 to 15 creatives, $100 a day, then keep the winners. George explains the metrics that matter early: conversion rate, a $2 ARPU target, and retention The #1 tool to find startup ideas/trends - https://www.ideabrowser.com LCA helps Fortune 500s and fast-growing startups build their future - from Warner Music to Fortnite to Dropbox. We turn 'what if' into reality with AI, apps, and next-gen products https://latecheckout.agency/ The Vibe Marketer - Resources for people into vibe marketing/marketing with AI: https://www.thevibemarketer.com/ FIND ME ON SOCIAL X/Twitter: https://twitter.com/gregisenberg Instagram: https://instagram.com/gregisenberg/ LinkedIn: https://www.linkedin.com/in/gisenberg/ FIND GEORGE ON SOCIAL X/Twitter: https://x.com/GeorgeLampro20

In this solo episode, I walk through the implications of the ban of Claude Fable 5 — the most powerful model on the planet and the one I planned to build with — after the US government sent Anthropic a letter. I make the case for local AI by walking through the benefits: intelligence that lives on your own hardware, stays private, runs free after the hardware cost, and keeps working through bans, outages, and price hikes. I lay out the exact order I'd learn it in — runtimes, model-to-hardware matching, quantization, and agents — and I name the specific tools and models I reach for. Then I hand you five startup ideas that exist precisely because intelligence now sits on your desk. The payoff for you is a clear plan to own a resilient layer of your stack starting this week. Timestamps 00:00 – Intro 01:20 – The Fable 5 Ban 02:31 – Renting Access vs. Owning Intelligence 03:41 – How a Local Model Works 07:19 – The Local Model Stack 08:45 – Match Model to Machine 10:45 – Pick Your Model (Qwen 3, DeepSeek, Gemma, Llama) 13:09 – Quantization Explained 14:36 –The Local Agent Loop 17:45 – Model Routing (The Real Skill) 18:44 – Five Startup Ideas for the Local-AI Era 22:17 – Closing Thoughts Key Points One government letter took Fable 5 offline overnight, which is why I now own a private layer of my stack. Local models already handle roughly 80% of everyday ChatGPT or Claude tasks, fully offline and free after hardware. I'd learn it in order: runtime first (LM Studio or Ollama), then match model size to your RAM. A 12-billion-parameter model on 16 GB of RAM is the sweet spot where most people should live. Quantization (look for Q4) roughly halves the memory a model needs while keeping quality high. Pointing an agent like Hermes at a local model turns your desk into a private, always-on mini data center. The #1 tool to find startup ideas/trends - https://www.ideabrowser.com LCA helps Fortune 500s and fast-growing startups build their future - from Warner Music to Fortnite to Dropbox. We turn 'what if' into reality with AI, apps, and next-gen products https://latecheckout.agency/ The Vibe Marketer - Resources for people into vibe marketing/marketing with AI: https://www.thevibemarketer.com/ FIND ME ON SOCIAL X/Twitter: https://twitter.com/gregisenberg Instagram: https://instagram.com/gregisenberg/ LinkedIn: https://www.linkedin.com/in/gisenberg/

Get my Fable 5 prompt pack: https://startup-ideas-pod.link/fable5-prompt-pack In this episode I break down how to get the most out of Fable 5, the most powerful model I've ever used. I move past the benchmarks and go straight into tactical use cases, copy-and-paste prompts, and startup ideas you can build today. I walk through tournaments for copy and landing pages, an interview-before-build workflow that hunts for product-market fit, and ways to point Fable at contracts, churn data, and years of your own notes. I close with three of my favorite startup ideas — a synthetic focus group firm, 48-hour custom software, and a contract refund firm — plus the exact prompts behind each. My goal here stays simple: leave you ready to build and earn with Fable 5 while it remains included in your plan. Timestamps 00:00 – Intro 02:22 – Anthropic Employee Edits a Launch Video With Fable 05:50 – Building an AI Content Engine 07:30 – Best way to configure Fable 5 08:42 – Prompt 1: Copywriting Tournament for Landing Pages 13:18 – Prompt 2: The Interview-Before-Build Prompt 18:34 – Prompt 3: Hire Fable to Kill Your Company 20:18 – Prompt 4: Your One-Page Operating Manual 21:20 – Prompt 5: Find the Gaps Worth Filling 22:06 – Prompt 6: Negotiation Simulator 23:11 – Prompt 7: The 80-Page Second Opinion on Contracts 24:56 – Prompt 8: Make Fable Build Its Own Tools 25:47 – Startup Ideas 31:23 – Closing Thoughts Key Points I show why low effort is the alpha, since Fable Low beats Opus High on routine work. I run tournaments — landing pages and ad copy scored by AI judge panels — to ship far stronger output. I use an interview-before-build prompt so Fable pushes back and writes specs with real product-market-fit odds. I point Fable at big datasets — contracts, churn data, support tickets, years of notes — to surface money and patterns. I share startup ideas Fable 5 makes viable today, including a synthetic focus group firm and a contract refund firm. The #1 tool to find startup ideas/trends - https://www.ideabrowser.com LCA helps Fortune 500s and fast-growing startups build their future - from Warner Music to Fortnite to Dropbox. We turn 'what if' into reality with AI, apps, and next-gen products https://latecheckout.agency/ The Vibe Marketer - Resources for people into vibe marketing/marketing with AI: https://www.thevibemarketer.com/ FIND ME ON SOCIAL X/Twitter: https://twitter.com/gregisenberg Instagram: https://instagram.com/gregisenberg/ LinkedIn: https://www.linkedin.com/in/gisenberg/

S/o Coderabbit for sponsoring today’s vid: https://startup-ideas-pod.link/code-rabbit On this episode I sit down with Professor Ras Mic to break down agentic loops. We define what a loop is, explain why well-known builders like Boris and Peter swear by them, and stay honest about who they truly serve. Mic argues that human-in-the-loop remains the strongest setup today, and he walks through the one loop he runs every day for code review using Cursor, GitHub, and Greptile. By the end you will know when a loop earns its place and when your own hand belongs on the wheel. Timestamps 00:00 – Intro 01:23 – What is a Loop 07:59 – /goal Explained 11:32 – The Slop Machine 12:42 – Code Review as a use case for Agentic Loop 18:19 – Honest Take for Builders 20:42 – The Future of Loops 21:50 – Closing Thoughts Key Points A loop fires once from a human, then the agent generates, reviews its own result, and feeds it back to keep building. Human-in-the-loop keeps you directing, governing, and approving each step while the agent builds. Wide-open loops make heavy assumptions and burn serious tokens; Michael cites Peter's tweet about $1.3 million worth of tokens in one month. Reserve slash goal and similar loops for the $200/month plan, since the $20 and $100 tiers burn through fast. Loops shine in confined, fixed-feedback work: code review, SEO pages, and other binary tasks. Mic’s daily win is a closed code-review loop with Cursor, GitHub, and Greptile that chases a 5/5 score. The #1 tool to find startup ideas/trends - https://www.ideabrowser.com LCA helps Fortune 500s and fast-growing startups build their future - from Warner Music to Fortnite to Dropbox. We turn 'what if' into reality with AI, apps, and next-gen products https://latecheckout.agency/ The Vibe Marketer - Resources for people into vibe marketing/marketing with AI: https://www.thevibemarketer.com/ FIND ME ON SOCIAL X/Twitter: https://twitter.com/gregisenberg Instagram: https://instagram.com/gregisenberg/ LinkedIn: https://www.linkedin.com/in/gisenberg/ FIND MIC ON SOCIAL X/Twitter: https://x.com/Rasmic Youtube: https://www.youtube.com/@rasmic