Activity
Mon
Wed
Fri
Sun
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
What is this?
Less
More
17 contributions to AI AUTOMATION INSIDERS
Do not set up an AI agent until you are good in the terminal.
Somebody brand new to Claude Code asked me last week whether to build an agent. Here's the order I'd do it in. First, get really good inside the terminal. That's where AI is the most powerful. Give it access to all of your tools in there. Second, n8n. It's the platform we use for automation, and it comes included with the Insiders. Third, an agent like Hermes. Very useful, but only once you know how to give it skills, tools and scheduled tasks. Now, why both? Because they do different jobs. An agent is always on. It runs in the background. You text it when you need something. It runs scheduled tasks. What it's not great at: an email comes in, do this, respond, send a reply. Anything trigger-based, we use n8n. And Claude Code is now very good at building inside n8n. Give it the n8n skills plus a connector that lets it talk to n8n. It builds the workflows. It debugs them. You never log in to n8n. You just tell it the automation flow you need, and it does it for you.
0 likes • 9d
The order matters more than the tools. Terminal first because that is where you learn what the AI actually did versus what it said it did. My rule after months of this: any script that writes anywhere reads itself back before it reports success, because a clean exit is not a result. Agents built before that habit just fail faster and quieter.
New Video: 3 jobs I took away from Claude Code and gave to Codex
Hello nerds. Astra is not AGI. But I did cancel one of my three Claude Code subscriptions, and the Codex app is now open on my screen all day. Three jobs moved over. 1. Anything in a browser. Astra knows where to click. It records my demo videos, fixes the user experience on pages Claude Code built, and maps the internal API of apps that don't have one. Skool included. 2. Writing code. Fable plans and reviews, Codex's cheaper models write the code in parallel, and Astra signs off on the risky calls. Way fewer Fable tokens burned. 3. Images. Through an API every thumbnail cost me about $5. Inside the Codex app it's included in your plan. It generates, reviews its own work and redoes the bad ones for $0. It's a glitch, and I doubt it stays open. Don't cancel Claude Code. The heavy lifting still happens there. Full video: https://www.youtube.com/watch?v=CKDm0lUB_Mo&list=PL0Yyhj9Mcbg8-kTfPi3s9hnIIDk5nl8m1 FREE RESOURCES FROM THE VIDEO -Codex Bridge (yours to install, you need a paid ChatGPT plan with Codex to run it): https://leadgenjay.com/skills/codex-bridge -CLI-Anything, the GitHub repo to hand to Codex: https://github.com/HKUDS/CLI-Anything Go watch it, then tell me which job you're moving first.
0 likes • 9d
Number one is the sleeper. This week I spent two days mapping Skool's internal API from the terminal: notifications, comment trees, DMs, all readable and writable from a script with no browser in the loop. The part nobody warns you about is that a 200 is not proof anything happened, so every write reads itself back before it counts. The browser jobs are where the leverage is hiding.
I threw out 2,024 of my August numbers on purpose
I ran my August numbers and threw out 2,024 of them on purpose. That is how many messages my automations sent last month. Every one of them delivered. None of them are in the number I am about to show you. Jay has been saying it all week in here. The slop problem is not that the AI is bad. It is that the output looks like everything else the AI made. I found the sales version of that in my own CRM. 1,289 messages I typed myself in August. 1,105 of them emails, one person at a time, with real research behind each one. 439 texts off my own phone. 62 a day. Every day. 28 days straight. The published benchmark for personalized outreach is 15 to 17 a day. Meanwhile the phone side was dead. I had been dialing so much from one number that I burned it. No alert, no dashboard. The call rate fell off a cliff, so I dialed my own cell from it. Spam Likely. Right there on my screen. 67 people answered out of 1,278 dials. One in nineteen. 41 closed. Best unit month I have ever had. 34 live calls the whole month. Do the math on where the rest came from. Here is the uncomfortable read for a room full of builders. The automated 2,024 ran flawlessly and I cannot point at a single close from them. The 1,289 I typed by hand carried the month. I am not saying stop automating. I automate more than most people in here. I am saying I know exactly which half of my stack made the money, and it was the half that does not scale. That is the thing worth solving next, and I do not have it solved yet. 👇 If you have an automation that runs clean and still does not convert, put the number below and what you think is actually missing. I read every one of these. Ian Ryan Kirk
0 likes • 16d
@Jay Feldman customization is what gets replies Speaking to them about their exact offer
0 likes • 10d
@Tony Iverson Automation for sorting, human for anything that already replied. That is the split I landed on too, and the dossier per lead is the part most builders skip because it does not scale on paper. It scales on close rate instead. How many touches does a qualified reply get from you before it goes quiet?
New AI system: Ghost Ink (get out of Promotions)
Alright, new module is live. Ghost Ink is what I use to get my marketing emails out of Promotions and into the Primary tab. It writes a hidden passage of personal copy that sits behind your actual email, so the providers read the message as a real note from a person instead of a promo blast. I'm running 42 to 55% open rates on it. I thought about selling this as a standalone product. People are charging $10,000 right now to implement it for other companies. You're getting it free, and nothing stops you turning around and selling it as an offer. Works with GoHighLevel, Kit and Listmonk. On something else like Mailchimp or ActiveCampaign? Download the files and have your Claude Code reconfigure it for your tool. One warning. There are real traps in this. Reuse the same passage, run one unrelated to the email, or hide it the wrong way, and you'll hurt your deliverability instead of helping it. Use the whole system, not just the prompt. Watch the lesson: https://www.skool.com/ai-automation-insiders/classroom/36507c93?md=8850c45a92a14839b86e3a964d87d825 Grab the system from AI Systems in your portal. If you're on GoHighLevel, install Fully Auto GoHighLevel first. https://leadgenjay.com/systems/ghost-ink A Ghost Ink UI inside Inbox Insiders is coming for anyone who'd rather not do this in Claude Code. This is version one. Tell me what breaks.
New AI system: Ghost Ink (get out of Promotions)
0 likes • 16d
@Jason Jackson seen any results 👀
I built an AI coach to review my own sales calls. It found something I didn't want to see.
You cannot see yourself on a call. You feel like you're doing the right things. The frame feels solid. The questions feel sharp. The close feels close. And then they go quiet and you don't know why. I built a coach to fix that. Not because I'm a developer. I'm not. Because I was losing deals I should have closed and I couldn't figure out where they were going. I built Call Reflekt Coach. After every call, I drop in the transcript. It breaks down exactly where I lost the frame. Where I pitched too early. Where I let the prospect lead me somewhere I shouldn't have gone. Then it writes a coaching note and a follow-up email. It is like having a coach in the room who is never tired and never polite about what went wrong. Here is what I found in the first month: I was giving away authority in the first four minutes of almost every call. Small things. Letting them redirect. Answering questions before I'd established the frame. Sounding helpful when I should have sounded certain. The coach caught it every time. If you are closing leads from cold email and you're not reviewing your calls with AI, you are leaving money on the table. You just don't know exactly where it's going yet. That is the part worth fixing. Ian Kirk
0 likes • Aug 1
@Asad Patel maybe you're interested in actually licensing my product. It actually has all of the vector knowledge base of some of the most popular top-rated sales coaching in the space.
1 like • Aug 6
@Mujahid S it uses Fathom to get the transcript through an API connection then it processes it through its knowledge based on course materials and coaching from different leading sales experts. Essentially its knowledge base is probably worth over $30,000 in coaching and course materials.
1-10 of 17
Ian Kirk
3
2 points to level up
@ianryankirk
💰 I Help Install AI Systems Into Your Business 📈 10x Productivity & Revenue 🎯 Get clients on autopilot ☎️ Call ↓ Book a clarity call

Online now
Joined Jun 29, 2025
Carlsbad CA
Powered by