GREG ISENBERG

Get Your Complete Financial OS at https://startup-ideas-pod.link/brex_SIP I welcome Ras Mic back to the pod to explain the phrase "software factory." Mic shares his screen and walks through the exact system that he runs today. His factory has four steps: isolate, build, prove, and ship. He keeps the whole system in five or six markdown files, so it works with any model and any harness. By the end of this episode, you can boot up your own factory, run many agents in parallel, and trust the code that comes back. Create your own Software Factory: https://startup-ideas-pod.link/ras-software-factory Timestamps 00:00 – Intro 02:17 – Software Factory Definition 03:44 – Why the Software Factory Matters 05:23 – Step 1: Isolate With Git Work Trees 11:34 – Step 2: Build With the Code Structure Skill 14:48 – Step 3: Prove With Evidence-Driven Testing 22:25 – Step 4: Ship With Grep Loop and Greptile 26:52 – The Physical Factory Analogy 29:21 – A Software Factory Is Markdown Files 30:02 – Closing Thoughts Key Points • A software factory is a workflow of skills and domain knowledge, so it runs with any model and any harness. • Isolate: every feature starts in a fresh git work tree branched from origin main, so each agent keeps its own station. • Build: a code structure skill makes the agent write service layer code that a human developer can read. • Prove: the agent records a before state and an after state as video, screenshots, or numbers. • Ship: Greptile scores the PR, and the agent loops back to build until it earns five out of five. • Mic runs up to 15 features in parallel and reviews the visual proof instead of the raw code. Numbered Section Summaries 1. What a Software Factory Means Ras defines the term as a workflow, a set of skills, and domain knowledge, packed into markdown files. The word "factory" points to structure, speed, and a conveyor belt that keeps quality high. He is clear that a software factory is a method rather than a product you buy. 2. Why the Factory Matters Now Model intelligence keeps rising, and Ras names GPT-6 Astra as his workhorse and the model with the lowest hallucination rate. A factory lets you use that intelligence in a systematic way, instead of typing back and forth with a chat window. He offers his own skills for free, and he asks people to understand the process first and then adapt it. 3. The Agents.md File as the Master Instruction The agents.md file is a single markdown document that goes into the agent chat ahead of every message. Ras says most people fill it with facts the agent already reads from the code base. He fills his with a workflow instead, because the workflow is the part the agent lacks. 4. Step 1: Isolate A skill called "new feature" starts each task in a fresh git work tree branched from origin main. A work tree acts like a copy of the app, and the agent merges it back later. This keeps two agents apart, so one agent leaves the other agent's files alone. Michael shows four terminal tabs running four features on the same app. 5. Step 2: Build The code structure skill tells the agent to write in a service layer architecture. Models get the job done, and they often get it done in a sloppy way, so the skill supplies the guideline. The result stays readable for a hired developer, for Michael, and for any fresh agent that opens the code base later. 6. Step 3: Prove Two skills handle proof: evidence-driven testing and before and after. The agent records the broken state, does the work, and records the working state, then embeds both in the PR. When the after shot shows a gap, the agent sends itself back to the build step on its own. For work with an invisible surface, such as performance, the agent supplies numbers: one page went from 815 milliseconds to about 61. 7. Step 4: Ship The grep loop skill sends the PR to Greptile, a third-party code review agent, and Greptile returns feedback plus a confidence score. A three out of five sends the agent back through build, prove, and ship. Ras steps in at five out of five and clicks merge. He mentions Code Rabbit and Macroscope as alternatives. 8. The Factory Analogy and the Case for Code Review I map the four steps onto a real factory: a custom order gets its own station, the assembly line builds it, quality control tests it, and shipping sends it out the door. Ras agrees, and he says he may rename his skills to match. 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/ 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

I talk with Ras Mic about GPT-6 Astra. We skip the game demos and the 3D toys, and we focus on use cases to earn money or improve products. I share 9 Astra prompts that I posted publicly, and Greg Brockman reposted. Ras then shows his hardware project: he moved from a speaker idea to a parts list, a Blender layout, and merged code in about 30 minutes. The takeaway is simple: use this model for the ideas that felt too large for you last year. Timestamps 00:00 – Intro 01:53 – Astra Overview 04:14 – 9 Astra Prompts 11:48 – Jarvis Speaker Idea 16:21 – Think Bigger with Astra 18:29 – Vibe Coding to Vibe Manufacturing 21:16 – Closing Thoughts Key Points • Astra costs more per task, and it uses fewer steps, so the value per dollar stays high. • A performance audit moved one of Ras’s apps from 800 ms to 20–30 ms. • A security audit on his live payments app found real risks in production. • Ras went from a speaker idea to a $561 parts order and a merged pull request in about 30 minutes. • Ras’s point: intelligence keeps climbing, and bravery stays flat. Ask for bigger things. • The shift that vibe coding brought to software now reaches physical products. Numbered Section Summaries 1. Why This Review Skips the Demos Ras opens with the reason for this format. The viral Astra clips show games and 3D scenes, and those stay fun to watch. We spend the episode on work that pays: better code, better products, and new businesses. 2. Price, Speed, and Two Audits Astra sits at OpenAI's top price and matches Fable. A Codex subscription covers most of that cost today, and the model finishes tasks in fewer steps. Ras recommends two prompts for any live app: a full performance review, and a security audit. 3. Reference Images and Computer Use Ras gets his best UI results when he gives Astra a reference image, and he used the old Microsoft XP network icon for one design. He also rates the computer-use ability highly. He pointed Codex at the Canada Revenue Agency site, and it found pages that Google search missed. 4. Nine Prompts I Posted I share the prompt list from my post that Greg Brockman reposted. The first three serve solo operators: a bill renegotiator that talks to provider chat support, a service-business teardown that turns an agency workflow into a $500–$5,000 per month software product, and a marketplace watcher that texts me every underpriced listing in my city. 5. Agents for Operations and Growth The next six prompts run a company. A weekly operator dashboard reads my docs, Stripe exports, and calls, then names the three highest-leverage actions. An agent-opportunity audit ranks tasks by human hours saved within 30 days. A browser operator returns an SOP, a nightly QA agent opens my app on a real phone, a competitor agent reports monthly, and a browser game captures leads for the core product. 6. Ras’s AI Speaker Project Was likes the HomePod hardware and wants his own agent inside it. Jailbreaking stayed closed to him, so he asked Astra to design a build. Astra connected to Blender, modeled the speaker, specified a Raspberry Pi build, and set a $350–$450 budget before tax. 7. From Blender Layout to a Merged PR Astra returned Amazon.ca links totaling $561 before tax, and Ras bought the parts. It then rendered the wiring layout in Blender, named Chinese suppliers for the shell, wrote the loader code for his agent repo, and opened a pull request. Ras merged it, and the whole path took about half an hour. 8. Vibe Manufacturing and Subsidized Intelligence I compare this moment to 2024, when Lovable, Bolt, Replit, and v0 turned everyone into a vibe coder. I expect thousands of one-million-dollar-per-year businesses from the physical version of that shift. Ras closes with the economics: his agent counted more than $24,000 of model use in one week against a pair of $200 subscriptions. 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/ 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

I run this episode solo. I explain local AI in plain terms: the model runs on hardware I control, and a cloud model runs somewhere else. I map the four pieces of the local AI landscape — the model, the warehouse, the software, and the workflow — and I define the words that beginners meet first: parameters, tokens, context window, quantization, and GGUF. I walk through the Google open model stack (Gemma 4, Google AI Edge, LiteRT-LM, AI Edge Gallery), compare the other open model families, and show three ways to run a model today. I close with a first workflow you can copy and three startup ideas that use local AI as the wedge. And a special thank you to Google for supporting the podcast. Get the full guide to running local AI: https://startup-ideas-pod.link/local-ai Timestamps 00:00 – Intro 01:35 – The Open Model the Landscape 03:09 – Vocab Decoder 06:48 – Google Gemma Clearly Explained 10:29 – Other Open Model Families 14:20 – Path 1: Run Gemma in LM Studio 18:17 – Path 2: Ollama 20:15 – Path 3: Google AI Edge 21:07 – Hardware Cheat Sheet 21:52 – First Workflow to Build 22:47 – Workflows Before Fine-Tuning 25:06 – Local vs Cloud vs Hybrid Eval 26:33 – Framework for Local AI Startup Ideas 27:22 – Startup Idea 1: Home Health QA Reviewer 29:24 – Startup Idea 2: Offline Field Report Copilot 32:10 – Startup Idea 3: Pre-Send Reviewer for Professional Services 34:47 – Build Your Local AI Lab 37:55 – Closing Thoughts Key Points • Ask whether the model is good enough for the job, and the business opportunities become clear. • Local AI has four pieces: the model, the warehouse (Hugging Face), the software (LM Studio or Ollama), and the workflow you build around them. • Gemma 4 E4B is my practical starting point; E2B fits phones and older machines. • Hybrid architecture wins: local does the private first pass, cloud does the heavy reasoning, and a human approves anything important. • Start with one repeated workflow — one folder, one model, one output — and run it 10 times. • I see a 24-month window to build local-AI-native software for verticals that still run early-2000s tools. Numbered Section Summaries 1. The Four Pieces of the Local AI Landscape I break the space into the model (the brain file, such as Gemma, Llama, or Mistral), the warehouse (Hugging Face), the software that runs the model (LM Studio or Ollama), and the workflow (the product around all of it). Underneath the tools sit Llama.cpp and MLX, and for shipping on-device apps in the Google ecosystem you reach Google AI Edge and LiteRT-LM. On Hugging Face I read a model card slowly and look for six things: purpose, size, license, hardware, supported inputs, and quantized files. 2. The Vocabulary That Matters I define parameters as the internal weights, where more parameters give more capacity and cost more memory: 2B and 4B for edge devices and fast workflows, 12B as a middle ground, and 26B or 31B for workstation territory. I define tokens, the context window, quantization (Q4 for easier running, Q8 for more quality), and GGUF as the common local file format. I suggest you start on a phone or a spare 2021 laptop and keep your money for later. 3. The Google Open Model Stack Gemma is Google's open model family, and Gemma 4 targets efficient local and on-device use across E2B, E4B, 12B, and 26B/31B. The specialized models deserve attention: EmbeddingGemma for search by meaning, FunctionGemma for tool use and structured function calling, PaliGemma for vision, ShieldGemma for safety, and Gemma Scope for interpretability. Around the models sit Google AI Edge, LiteRT-LM, AI Edge Gallery, and Gemini plus Google Cloud for frontier-level reasoning. 4. Three Ways to Run Gemma Today Path one is LM Studio: download the app, search for Gemma 4, pick E4B or E2B, grab the quantized GGUF, and paste real customer notes into a chat to feel the value. Then start the LM Studio local server so your scripts and prototypes call the model through localhost. Path two is Ollama with a local API on port 11434, and path three is Google AI Edge with LiteRT-LM for Android, iOS, web, desktop, and edge apps. I also give a RAM cheat sheet: 8 GB stays small, 16 GB runs useful experiments, 32 GB opens larger workflows, and a strong GPU makes the bigger models realistic. 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/

On this solo episode, I review five free, open source GitHub repos that help you build products, make money, or save time: Peter Yang's No AI Slop Skill, the CRM by TryComp AI, Video Use by browser use, SkillSpector by NVIDIA, and Phone Harness. For each repo I explain what it does, why it matters, how to install it, and the first small workflow to try. I close with a simple three-step method: install the repo, make one small workflow work, then decide to productize it or keep it as your own leverage. Timestamps 00:00 – Intro 01:44 – Repo 1: No AI Slop 05:20 – Repo 2: Agentic-first CRM 10:52 – Repo 3: Video Use 15:15 – Repo 4: SkillSpector 18:49 – Repo 5: Phone Harness 22:25 – Closing Thoughts Links to repos: petergyang/no-ai-slop — https://github.com/petergyang/no-ai-slop trycompai/crm — https://github.com/trycompai/crm browser-use/video-use — https://github.com/browser-use/video-use NVIDIA SkillSpector — https://github.com/NVIDIA/SkillSpector phone-harness — https://github.com/ShawnPana/phone-harness Numbered Section Summaries 1. Why I Watch GitHub Now I treat GitHub as an early view of the agentic era. Many SaaS companies, agencies, workflows, and startup ideas start there first. I picked the repos from the ones that got attention over the last 30 days, and I explain each one in plain English. 2. No AI Slop Skill: Human Ideas First, AI Cleanup Second Peter Yang's skill strips the patterns that make writing smell like AI, and it keeps your actual voice. I care about this because founders write constantly: tweets, landing pages, cold emails, launch posts, onboarding copy, and investor updates. Readers who sense AI copy trust you less, so this repo makes your communication more believable. 3. The CRM Built For Agents The CRM by TryComp AI treats your CRM as a workspace for an agent instead of a filing cabinet for you. The agent researches people, enriches company data, schedules follow-ups, and keeps the notes current. Most businesses lose money because opportunities scatter across old emails, so a system that maintains the relationship graph for you is real revenue. 4. One Small Pipeline To Start With I would start with a single pipeline of warm leads that deserve a follow-up: everyone who replied, booked a call, asked about pricing, or promised to circle back. The install needs bun and docker, a git clone, the env example, and a few bun commands. It runs locally on localhost 3000 with the API on localhost 3001, and it uses Google or Microsoft OAuth for sign-in, email, and calendar. 5. Video Use: Turn Your Edit Style Into A System Video Use by browser use lets coding agents edit your raw footage. The agent removes filler words, cuts dead space, adds subtitles, color grades, builds overlays, renders the video, and then checks the output. Your edit style usually lives in muscle memory, and this repo makes it explicit and repeatable, which shrinks the distance between raw material and published content. 6. SkillSpector: Scan The Tool Before You Trust It NVIDIA's SkillSpector scans agent skills for prompt injection, data exfiltration, supply chain risk, hidden instructions, and MCP related risks. A skill carries instructions, scripts, dependencies, and tool access, so it changes how your agent behaves. As builders assemble their own AI work environments, security becomes a normal founder problem, and I like that NVIDIA gives this category credibility. 7. Phone Harness: Agents That Operate A Real Phone Phone Harness connects an agent like Codex or Claude Code to a real iPhone through Mac iPhone mirroring, or to an Android through ADB. The agent sees the screen, taps, types, scrolls, opens apps, and verifies the result, so mobile QA, API-free automations, and repetitive creator workflows become possible. The project is early, and face ID and camera flows stay tricky, but the direction is obvious. 8. Install, Test, Then Decide To Productize My method has three steps: install the repo, build one small workflow that adds value, then decide to productize it or keep it for your own speed. Even people with light technical skills learn a lot from the install itself. I recommend a monthly habit of finding and installing new repos, because it connects the dots in this era. 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/ FIND ME ON SOCIAL X/Twitter: https://twitter.com/gregisenberg Instagram: https://instagram.com/gregisenberg/ LinkedIn: https://www.linkedin.com/in/gisenberg/

In this solo episode I explain a role that I call the marketing engineer. I believe this person becomes one of the most valuable hires in tech in the next 18 to 24 months. I define the job, I show the four eras of marketing that lead to it, and I give the tool stack that makes it work. I use a commercial HVAC software company as a worked example, and I list six systems that a marketing engineer builds. I close with four ways to earn money from this skill and a 30-day plan to learn it. Timestamps: 00:00 – Intro 01:46 – The Evolution of Marketing 04:29 – What is a marketing engineer 07:19 – Build the Growth OS 10:18 – Marketing Engineer Tool stack 13:23 – Live Data Workflow 14:32 – Agent Job Description 16:56 – Example: vertical SaaS for HVAC contractors 18:27 – System 1: Customer Truth 20:20 – System 2 - 4: Founder content, Outbound signal and Creative Testing 23:31 – System 5: AI search visibility and the growth cockpit 24:19 – System 6: Eval Loop 25:06 – Ways to Monetize 29:41 – The 30-day plan 32:24 – Closing Thoughts Key Points • I expect the marketing engineer to command salaries from 250K to more than 1 million dollars. • I build the growth repo first, because it holds the marketing memory of the whole company. • I write a job spec for each agent, in the same way that I write a job description for a person. • I measure qualified replies and pipeline, because business results show the true signal. • I treat taste and judgment as the moat, because agents become a commodity. • I recommend one working system over five half-built ones. Numbered Section Summaries 1. The Four Eras of Marketing I started and sold three venture-backed companies across the web, social, and mobile eras. Each technology shift creates a new type of valuable marketer: the traditional storyteller, the digital acquisition marketer, then the growth hacker. The agentic shift now creates the marketing engineer. 2. My Definition of the Role A marketing engineer turns market signal into pipeline with AI agents, data, code, and taste. The person keeps the old skills of positioning, customer understanding, and distribution. The new part is that the person also builds the system behind the marketing. 3. The Growth Repo Most teams use AI in random chats, and the work disappears each week. I fix this with a GitHub repo or a structured folder that I call Growth OS. It holds folders for customer truth, the content engine, the outbound engine, creative testing, and agent jobs. 4. The Tool Stack I use Grokbot as the layer that stays close to the live internet, because it connects to the X ecosystem. I use Claude and Codex to build the repo, the landing pages, and the internal tools. Hermes-style workflows run scheduled jobs with memory and approval, creative models handle ads and thumbnails, and local AI covers sensitive data. 5. Agent Job Specs Every agent needs a written spec: the data source, the run schedule, the filters, the expected output, the approval step, and the metric. I train agents in the same way that I train a new hire, with small tasks and corrections. Each correction goes back into the repo, so the system compounds. 6. Six Systems, With an HVAC Example I use a vertical SaaS company that sells to commercial HVAC contractors. The six systems are the customer truth file, the founder content engine, the outbound signal engine, the creative testing engine, AI search visibility, and the growth cockpit. Each one turns a sharp customer pain, such as missed follow-up quotes after a service call, into content, outbound, and tests. 7. Four Ways to Earn Money The first path is the role inside a company, which sits directly beside revenue. The second is consulting on a 30, 60, or 90 day embed, at five to thirty thousand dollars per month. The third is a productized service with one tight wedge, and the fourth is software that comes from the pain that repeats across clients. 8. The 30-Day Plan Week one is an audit of one real company and a market map. Week two is the growth repo and the first "what the market is telling us" file. Week three is one working machine, and week four is results and a written case study. 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/ FIND ME ON SOCIAL X/Twitter: https://twitter.com/gregisenberg Instagram: https://instagram.com/gregisenberg/ LinkedIn: https://www.linkedin.com/in/gisenberg/