Greg Isenberg has 725,000 subscribers and he still spends his nights installing random GitHub repos to see what they do.
Same.
This week he dropped a solo episode walking through 8 repos and AI skills he tested after hours. One makes a launch video from a link. Another mines Reddit for customer pain you can charge to fix.
I pulled the transcript, checked every repo on the list, and wrote up what's actually worth your weekend. If you're hunting for GitHub repos to make money, this is the most usable list I've seen in months. Most are free to download. You pay for the model running them, and that's it.
Watch the full episode here, then steal the angles below.
1. Brag: a launch video from one command
Funded startups pay $100,000 to $300,000 for launch videos. Greg drops that almost offhand in the episode, and he's right, you see these videos all over X.
Brag (latent-spaces/brag, around 14,000 stars) is a skill that reads your project, scripts a 20-second launch video with music and motion, and writes the share copy to go with it. The newer version takes a plain website link.
Greg's favorite detail: one of the demos is a fake Series A announcement from 2016. You choose the tone the way you'd choose a vibe.
The money angle is the part I'd steal. Scroll Product Hunt, find launches with no video, make one, send it over. "Hey, I made you a launch video. What do you think?" Worst case, you started a conversation with a founder. Best case, they want more of that work.
Install with npx skills add latent-spaces/brag. The skill is free. The video model costs are yours.
2. Agent Reach: customer research without 40 open tabs
Agent Reach gives your AI agent eyes on 16 platforms (X, Reddit, YouTube, GitHub, LinkedIn, even RSS) with zero API fees. One CLI, pip install agent-reach, done.
Greg's use case is the one that got me: pick software people already pay for, then ask what customers are still doing by hand.
His example is a cleaning company owner who pays for scheduling software and still exports the schedule to a spreadsheet every Friday to work out payroll. That gap is a small tool someone will pay for.
The searches he'd run: "I export it to" and "does anyone know a workaround". Then message one of those people and ask them to show you what they're doing. You still need that conversation before you assume there's a customer.
If workflow gaps like that are new territory for you, Twilead's guide to business automations maps out where businesses bleed hours every week. Cross it with what Agent Reach digs up and you have a build list.
3. VoiceStudio: a local ElevenLabs on your own machine
VoiceStudio (18,000+ stars) does voice cloning, dubbing and transcription in 646 languages, and all of it runs on your own computer. There's no subscription, and your audio never leaves your machine.
Greg's starter move: record a clean sample of your voice, feed it a few lines, and listen to what comes back before you get ambitious.
The business he sees in it: every software company has tutorial videos, and most of them only exist in English. If a product already has Spanish-speaking customers, those English-only onboarding videos are quietly creating support tickets.
So dub the three most-asked tutorials, have a fluent speaker check the result, and charge a monthly fee to keep them updated as the product changes. Products ship changes constantly, which means the videos rot on a schedule. That's recurring revenue.
Two warnings straight from the episode: your hardware decides how fast it runs, and check the model licenses before commercial work. Only clone voices you have permission to use.
4. PPT Master: decks from notes, in your own template
I can hear the eye roll. PowerPoint. Greg said the same thing, then made a decent case.
PPT Master by Hugo He turns documents or meeting notes into a real, editable deck. Native shapes instead of slide images, and it can follow your company's own .pptx template. When a client asks you to change one price, you open the file and edit it.
His workflow: come out of a sales call, hand it your notes with the agreed scope and price, review what it builds, and send the proposal while the conversation is still warm.
One rule from the episode is worth the whole section: write the deck for the person who wasn't on the call. Your contact will forward it to whoever approves the budget, and that person needs to understand it cold.
There's service work here too. Ask a sales team "which slide do you keep rebuilding yourself?" and give them one good version built from approved numbers. (That becomes ongoing work once you know the product.)
5. OpenRig: your coding agents, organized into a team
If you run Claude Code and Codex side by side, you've probably been the messenger between them. Copy a result from one window, paste it into the other, carry the review feedback back.
OpenRig turns that into a system. You define an agent team in YAML, boot it with one command, and it runs Claude Code and Codex together in tmux as one persistent setup. One agent builds, another checks the work, and there's a real way to pass tasks between them.
Setup needs Mac or Linux with Node 20+ and tmux, and it changes some agent settings, so read those before running anything. Then npm install -g @openrig/cli.
Greg's test job is a good one: a quote calculator for a website. One agent builds it, the other checks it against the business's actual pricing.
Where it pays: client website work. Launch a site and the small change requests never stop. I've been in that business, and the coordination eats more time than the actual changes do.
6. Free Tools: Corey Haines' engineering-as-marketing skill
This one lives inside Corey Haines' marketingskills repo, a bundle with over 280,000 installs. The free-tools skill helps you work out what you could build and give away to attract the customers you actually want.
Greg's example: a packaging supplier whose prospects are stuck deciding between ordering 2,000 or 5,000 boxes. A calculator that shows what each option costs is genuinely useful. And whoever asks for a quote afterward has already told you what they're buying.
The detail that matters: show the result before you ask for an email. Then offer to turn the estimate into a real quote without making them type everything in again.
I've watched free tools outperform paid ads for more than one founder on the podcast. A calculator that answers a buying question beats a banner every time.
7. Grill Me: the skill that interviews you before you build
Matt Pocock's skills repo is one of the most-starred projects on GitHub right now, and Grill Me is the standout inside it. You give it an idea or a decision, and it interviews you about it in rounds, following up on your answers.
Greg's example hit close to home. Say you want to sell local ads to franchises. Who pays you, the franchise owner or headquarters? Does the person paying even have permission to approve the ad? You can confidently build a whole dashboard before discovering every ad goes through someone at head office.
His instruction to add: track every time you say "probably". "They probably have budget." "They probably gave us access." Those are the assumptions to take to a real customer first, because the skill prepares you and the customer tells you what's true.
It works for smaller calls too. Should we offer an annual plan? Is this feature worth building for one big client? Cheap questions to answer before you open the code editor.
8. Impeccable: design taste for your coding agent
Paul Bakaus built Impeccable to fix the "make it look better" problem. It's design guidance for coding agents: 50,000+ stars, 24 commands like critique, polish, bolder and quieter, plus 59 detector rules for the awkward crops and overlapping elements AI designs love.
Quieter is underrated. Greg said sometimes quieter is exactly what he wants from an AI-generated site, and half the ones I see prove his point.
Start with critique on a page you already built and see what it catches that you missed. Fix the parts you agree with, and focus on one flow like onboarding or checkout instead of repainting the whole site.
The money angle: find a small product you like whose site is hard to buy from. Pricing that's confusing on mobile, a signup form that asks too much. Put together two specific fixes, show the founder what you'd change and why it helps, then quote $1,000 a month for design work.
Greg pre-empted the comment section on this one: $1,000 a month isn't a business. Ten clients later it's $10K a month, and every big business started smaller than that.
The playbook hiding in this list
Greg's two picks to try first: Brag, because you end up with a video you can send to someone tonight, and Agent Reach, because customer pain is where real ideas come from.
But notice the shape of almost every money angle in the episode.
Install a free repo → do something specific for a stranger without asking permission → let the work start the conversation.
That's the launch video for a Product Hunt founder, the dubbed tutorial library, the design fixes with a $1,000 quote attached. None of it needs an audience or anyone's approval to start.
Greg's earlier lists run the same direction. I broke down his 13 businesses for the age of AI last week, and the repo roundup I covered in AI agent GitHub repos pairs well with this one if you want more ammunition.
Services first, product second. Do the job 20 times, learn where it hurts, then build the tool. Most of the founders I interview got started exactly like that.
FAQ
Are the GitHub repos in Greg Isenberg's video free?
All eight are free to download or install. What you pay for is the model running underneath: Brag renders video through AI models, OpenRig runs Claude Code and Codex sessions, and VoiceStudio runs local models, so your only cost there is hardware. Check individual model licenses before commercial use, especially for voice cloning.
Which GitHub repo should I try first to make money?
Greg's pick is Brag, and I agree with him. A 20-second launch video gives you an asset you can send to a founder the same night you install it. Agent Reach is the second install: it surfaces the customer pain that tells you what to build or sell next.
Can you actually make money from GitHub repos?
Not from the repos directly, since they're free. The money is in what they let you ship: launch videos for founders, dubbed onboarding libraries, design audits at $1,000 a month, small tools that kill someone's Friday spreadsheet ritual. Ten $1K clients is $10K a month, and that's a real business.
Do I need to be a developer to use these?
Less than you'd think. Most of these install with one command and run inside a coding agent, which handles the technical part. OpenRig is the exception since it assumes you already live in agent workflows. If you can write an honest description of a problem, you can operate the rest.
