| Happy Sunday! If you’re reading this, I hope you’re feeling energized, got time to relax this weekend, had a Sunday Funday, and you’re looking forward to getting back in the saddle tomorrow morning. I’m writing to you from my desk on Saturday here in New York City where we have absolutely perfect weather outside. Today’s newsletter will be something you can have open on one screen, and then “do” alongside on another screen with me. When I talk to other colleagues and people within our industry, I love to find out how people are using AI. Most people use it to do brainstorming, research, or ask questions and get a quick answer without having to manually search for something. To that extent, most people use regular Claude, Claude Co Work or ChatGPT. By the end of this newsletter, I want you to feel like you can take one or two things into your week, apply them, and see material differences and improvements in your own workflow and efficiency. This newsletter was inspired by a few emails I got mentioning how helpful the AI episodes of Limited Supply have been. But, before we get into today’s newsletter... Vendor of the WeekPollen — The easiest way to recruit affiliates, creators, and to whitelist. It’s no secret, the new playbook for building upper-funnel awareness quickly is by way of creator marketing through affiliates on TikTok and Instagram. However, the process of searching for the right affiliates, messaging them with ease, negotiating rates, then managing the creators and keeping them informed is all fragmented. Pollen, launching in September, handles the entire process end-to-end. I’ve been a beta user for the last few months and have particularly enjoyed the competitor scanning (you can see who top affiliates are for other brands) and being able to use my Hermes agent to handle all the back-and-forth messaging with creators. Pollen also integrates directly into Discord, so you can easily onboard new creators and make sure they stay informed about new product and company updates. When a creative takes off, you can use Pollen to run those creatives as whitelisted ads on any ads platform (Meta, YouTube or TikTok). Pollen launches next month. 12 brands who sign up on this link will get free lifetime access. Limited SupplyThis past week’s episode of Limited Supply was actually all about how to best setup your AI system from scratch. I guess this thinking has been in my mind all week, and today’s newsletter is the perfect companion to this week’s episode. I sat down with Jacob Posel, one of the co-founders of HQ, to learn about how the biggest brands have set up AI infrastructure internally to leverage real-time leanings/skills/memory/context across the whole company. The episode got a ton of great early feedback, so give it a listen while you’re on a run, in the shower, driving to work, or walking the dog. Listen on Apple Podcasts, Spotify, or YouTube. Just search “Limited Supply” in the search bar! Building Better Workflows with AISo back to what I was saying earlier, most people are still using Claude or CoWork to execute many of their AI tasks. For today’s newsletter, my hope is that 1 or 2 of the things mentioned below get carried into your week and allow you to 1. level up your own AI usage; and 2. allow you to see what else is possible, which may bring more ideas to you of how to leverage it. A few weeks ago I put out a talk about my Hermes setup (see it here) and how I use it day-to-day. Jet Damon, my agent, lives inside a Mac Mini on my desk, and is my daily co-pilot. What I realized with this setup, is everything just stays local to the Mac Mini. So since around mid-June, I added another layer to it, called HQ, which is a memory system that keeps my whole team in sync (in this case, me, Jet, and my assistants). I want you to mirror my setup, because while it’s advanced, it’s not complicated in a way that someone who’s not technical can’t figure this out. Try to replicate everything below, in order, and you’ll also feel like you’re not behind when it comes to AI. The only thing you have to get comfortable with is moving over to Claude Code and Codex (ChatGPT’s version of Claude Code). As someone who didn’t make that move for a long time myself, I agree, the idea of going to “Code” sounds intimidating... I don’t know anything about code, I’m one of few Indians who doesn’t know how to code. BUT, as someone who’s non-technical, I’m here to tell you that you don’t do any coding in Claude Code or Codex. You can, but Claude Code/Codex is basically just a more open-sourced version of Claude or ChatGPT. When you use regular ChatGPT or Claude, it’s somewhat of a black box. Your memory, context, skills, conversations, etc are all just saved in this black box. With Claude Code/Codex, you’re just using a harness, and choosing the model (either Anthropic or OpenAI’s models) and then it’s using files from a source. So all that said, I would start in this order: Get set up with a good memory system.Because it’s AI, we believe the answers should always be instant, and always correct. If I say that my product only can be talked about a certain way, it should never mess that up in the future. Unfortunately, it happens. It’s because the markdown files that save things to your knowledge base like plain-text files. After a certain point, when your systems/agents start up, they are moving so fast, they don’t read the entire file, and they miss context. Also, once your own Claude Code or Codex is trained up on things, don’t you want to make it easy to share those exact context and memory files with the rest of your team so everyone is operating from the same POV and knowledge files? That’s where I use HQ for Work. It’s a vectorized memory system that is able to make memory recall for agents instantly fast (I believe Tobi, the Shopify founder, has a lot to do with creating this). For me, each project I work on (a new brand I’m working on, a brand I’m doing consulting for, or one of my sidequests like Zurp) has its own folder. Each folder holds its own context, memory, way of working, etc. Within Claude Code/Codex, I always select HQ as my source folder, and that way any conversations I have, they get saved to HQ, which gets synced to the rest of my team every 5 minutes, so no one is using outdated systems. It’s like Dropbox sync for your company’s organizational AI infrastructure. Connect all your tools & add APIs.When you open Claude Code or Codex, natively you can connect a bunch of stuff. If you haven’t already, I would highly recommend you do the following at a minimum:
Those should all be direct integrations, and if your call recording software, for example, doesn’t have a direct integration, it likely has an API key. An API (Application Programming Interface aka the ability for an app to universally connect to another piece of software) key makes it easy for your setup with Claude Code/Codex to any software. The ones I’d recommend outside of your call recorders:
Access to all of these allows you to not only pull data and read what is happening inside each of the accounts, but then use that data. For example, if you built a workflow to come up with new angles to test with Claude Code, you can also now plug back into Meta to pull the results of how those angles did, and use that information as helpful data to contribute to the next round of briefs. The last piece of this would be using an MCP connector, which some apps have if they don’t have a direct API available. APIs allow software-to-software connections. MCP allows LLM models-to-software connections... it’s basically like an API, but only LLMs can access it. If you use software like Motion/Runneth or Higgsfield, you can connect to them via MCP, and your Claude Code or Codex can then send routine tasks to Higgsfield to develop creative a certain way, without you logging into Higgsfield. Build your own knowledge files.As you leverage more AI in your work, whether it’s by using Claude Code/Codex on your own computer, or when you add new tools like Runneth to your stack and need to give these tools context files, you will want to have a set of updated and informative knowledge files. Some of the basic ones I use often:
You can simply tell Claude Code/Codex to put this together for you in markdown files (.md files), which are like plain text (.txt) files for AI agents to read. This allows your output across your suite of AI tools (whether it’s your “custom” setup at home or using AI within other SaaS tools) to be consistent and uniform. You’re not having to fully train each AI tool like a fresh puppy with no memory of your product, because now you have knowledge files ready to drop in and immediately bring the tool up to speed. These are essentially like AI-agent-friendly brand book files. Build a contract review skill.This one is too easy to setup, and will save you so much money in the long run. If you’re a brand doing less than $250M per year, it’s likely that you outsource your contract review to a law firm. Those lawyers, rightfully so, can be $200 to $750 per hour, and they bill in 15 minute increments. So when you have a contract to be reviewed, whether it’s for a new SaaS tool, an influencer campaign, a new distribution channel, or for a partnership event, those lawyer hours add up. If you’ve been operating for the last couple of years and have history with your attorneys over email that documents all the changes to your contracts and what they consistently ask to change, you have enough to build a contract review skill. If Claude Code/Codex are plugged into your email, you can simply prompt it to put together this contract review skill (I made this prompt you can steal here). Build your own custom dashboard.Now, everyone on Twitter talks about how AI is just used to build dashboards that people never open again. And maybe that’s 1000% true, but you know what? Building a dashboard is like every consumer app founder’s first idea being a list app, or every entrepreneur’s first idea being a roommate matching app... I don’t make the rules, but the path to becoming great with knowing how to use AI starts with creating dashboards. Yes, building a custom dashboard that you can use is a reason in itself to do it, however, the process of building one is what opens your eyes to what’s possible. Connecting everything, learning you can create any custom metric, and pull it without any mistake or breaking in the data gives you the confidence to then build your own internal tools and growth OS. Automate a workflow & turn it into a skill.As you get to use Claude Code/Codex for a few weeks, you can just ask it, “What am I doing routinely that we can turn into a skill to be more efficient?” It’ll simply tell you. For me, it recognized I would run a simulation with a Sr. Art Director, a Growth Marketer, Copywriter, and Creative Strategist. So we ended up making a skill called “Review Board” that runs the full process when I send “/review-board” back in chat. If you already have something in your day-to-day workflow you want to automate, you can just teach it to Claude Code/Codex, and you’re one step closer to freeing your time by routine tasks. “Skills” are one of the most underrated, and under-talked-about pieces of AI agents. To me, it’s the unlock. Your own skills, the ones you create, and the ones molded by HOW you think, are what separate your own AI tools from the next person’s. Yes we all have access to the same models, but the skills you create and how you use the model is what determines the quality of the output. This is somewhere you should spend more time in. With HQ (memory system mentioned above), these skills automatically are shared across your organization once created. If you start with even one of these, you’re going to start to really feel the power of AI and how it can plug into your workflows across more than just the basic Claude or ChatGPT chat window. That’s all for this week!I hope today’s newsletter inspires you to take a few of these nuggets and implement them. If you have any questions on how or what, reply here to this email or DM me on Twitter and I’d be happy to help. It’s Sunday night, so I hope you plan to get 9 hours of sleep going into the new week. Make sure you prioritize some self care, get sweaty, go for long walks, call friends and laugh, and stay hydrated. Have an amazing upcoming week. I’ll see you next Sunday! | |||||
|
