📖 A Robot Showed Up At My House And I Have No Idea How Robots Work

Let me be very clear about my qualifications here: I think robots look cool and I know they can walk and talk. That's the list. That's everything I bring to the table.

And yet there's one standing in my office right now, because VH Robotics was kind enough to loan me one to experiment with.

Here's the part that still gets me. I didn't earn that robot by being a robotics guy. I earned it by spending months building things with Claude Code — publicly, badly at first, in front of everybody — until someone looked at what I was making and decided I was worth handing hardware to. The experimenting was the résumé.

I'm going to say something that sounds soft for a marketing newsletter, so bear with me. Claude Code and AI have made me happy. Not "more productive." Happy. I'd lost the joy of building somewhere in the years of running five businesses, and I got it back. I'm making things again — things I have no business being able to make — and it's the most fun I've had working in a decade.

Then it opens doors you couldn't have written down in advance. I never had "get sponsored a robot" on a goal list, because how would you even set that goal? It wasn't reachable from where I was standing. It became reachable because I picked up a tool that let me build past my own skill level.

So I'm going to learn robotics the way I learn everything now — by breaking it on camera. Videos over the next several weeks. The wins, the failures, the parts where I stare at an error message for forty minutes. Fitting, given we just covered a $399 robot duck whose whole design philosophy is fall over, get back up, try the next thing. That's the entire method. That's been the method the whole time.

Which brings me to why you're reading this.

Everything I just described — the builds, the agents, the robot showing up — runs on two skills. And we teach both of them this month.

🤖 If you want AI doing work on your behalf, start with Cowork. This is the beginner room. You'll build agents that handle your inbox, your research, your recurring busywork, while you do something better with your morning. September 8–9, 5:00–6:30 PM EDT. Registration closes September 8 at 5 PM EDT — which is the same day it starts, so don't sit on it.
👉 Grab your seat: https://go.vcinc.ai/claude-cowork-b

💻 If you want to build the way I'm building, that's Claude Code. Four nights of actually shipping — this is where you go from using AI to making things with it. September 21–24, 5:00–7:00 PM EST.
👉 Grab your seat: https://go.vcinc.ai/claudecode

Both have sold out every single time we've run them, because we over-deliver and our community says so out loud. And VCI is past 10,000 members now, which means these fill faster than they used to.

If you're not sure which one: start with Cowork. Get agents working for you first. Then come build with us in September's Code cohort once you've felt what that's like.

Twelve months ago I couldn't have told you what an agent was. This week a robotics company mailed me a robot.

Go get your seat.

🛠️ Ask AI Just Left the Platform — And Went Prospecting Without Me

Every AI feature in every CRM has had the same ceiling: it can only touch what lives inside the CRM. Ask AI just broke that ceiling. I gave it one paragraph, took my hands off the keyboard, and watched it open Chrome, hunt down local HVAC companies with broken websites, score every one of them, then log into my Gmail and write the cold emails.

Not "suggest" the emails. Write them. Sitting in my drafts folder, signed, ready to send.

The unlock is a Chrome extension called Ask AI LeadConnector, and it turns HighLevel's assistant from a thing that answers questions into a thing that does the work on the open internet.

Here's what it actually does once it's on:

• Setup takes about ninety seconds. Grab the Ask AI LeadConnector extension from the Chrome Web Store, hit Add to Chrome, confirm it's toggled on under Manage Extensions, then make sure Ask AI is enabled on your agency or sub-account. That's the whole install. No API keys, no Zapier in the middle, no developer.

• You prompt it in plain English and it builds its own plan. I asked for HVAC companies in my area with weak websites, no chat or voice widget, and thin Google reviews compared to their competitors — plus a site speed check. It broke that into its own list of tasks and started working through them one by one, with a little arrow moving across the screen doing the clicking.

• It comes back with a scored audit, not a list of names. Every prospect got a site score out of 100, a review count benchmarked against their local competitors, a widget check, and a written breakdown of what's critically broken. That's a qualification layer most agencies pay a VA to do by hand.

• It writes the outreach inside your actual email client. I told it to open Gmail and draft cold emails in draft mode. It found the addresses, opened Gmail, and wrote four of them — using the audit data as the hook. Subject lines like "Your website scored 55 out of 100" and "Seven reviews won't win you customers — let's change that." It even pulled my credentials in and signed them.

• It doesn't stop at email. I sent it into Canva to mock up sample social graphics for each prospect, then had it build a Google Doc of blog post ideas for every company on the list. It logged into Canva on its own, figured out an interface nobody trained it on, and produced the assets.

Real talk on the limits: it is not smooth. Watching it fumble through Canva was like watching someone learn to drive stick — it opened four windows, backed itself into corners, and took a few runs at it before it got there. It got there, but you're supervising, not napping. And be deliberate about what you're logged into when you hand over the wheel, because it inherits every session sitting open in that browser.

The thing worth sitting with: the moat around agency prospecting was never talent, it was hours. Somebody had to open twenty-five tabs, eyeball twenty-five sites, count twenty-five review profiles, and write twenty-five emails that didn't sound like a template. That was the whole job. That job just got handed to an extension that runs while you're on a call — and the agencies still doing it by hand are now competing on speed against people who aren't doing it at all.

 🤖I Cancelled $1,296 of Subscriptions and Claude Code Did the Installing

Open your bank statement and add up the software you pay for monthly but use maybe four times a month. Adobe Acrobat at $24.99. Loom Business at $18 a user. A whiteboard tool. A link shortener that caps you at five links on the free plan. None of it is a lot on its own. Together it's $1,296 a year leaving your account for tools that already have free, open-source twins sitting on GitHub with 80,000+ stars and a license that says do whatever you want with this.

The reason you've never installed them is that self-hosting used to mean knowing what a dependency tree is. That's the part that changed. You're not doing the install. Claude Code is.

Two things make this work, and neither one requires you to write a line of code:

• Docker is the sealed box that makes any of this possible. Software normally needs a pile of other things to run — a database, the right language version, config files in the right folders. Miss one and the install falls apart. Docker ships the app and everything it needs inside one container: it runs identically on any machine, and if you delete it, it leaves nothing behind. Free version, Mac or Windows, accept every default. That's the only prerequisite.

• Claude Code reads the repo and does the rest. You make a folder, open it in VS Code, paste the GitHub link, and say "install this here using Docker, explain each step before you run it." It writes the docker-compose file, validates it, pulls the image, runs the container, and hands you a localhost link. When one of them needed a login and I had no mail server attached, Claude wrote me a workaround file with the exact command to pull my own verification code out of the logs. It anticipated the wall before I hit it.

Here's what came out the other side:

• Stirling PDF replaces Adobe Acrobat. 50+ tools — merge, split, OCR, sign, redact, compress, actual text editing. MIT licensed, which I read out loud on camera because I didn't believe it either: you can use it, modify it, merge it into your own product, and sell copies of it. That's $24.99/mo gone, and a business asset you're allowed to resell.

• Cap replaces Loom. Screen recording with shareable links, self-hosted. Loom Business runs $18/user/month, $180 a year, and the free plan caps you at 25 videos and 5 minutes. I've got 3,000+ videos in Loom so I'm not going anywhere — but if you're a two-person shop paying $360 a year to record client walkthroughs, this is the exit.

• Listmonk replaces your newsletter tool's subscriber tiers. Unlimited subscribers, no jumping a pricing bracket every time you cross 10,000. You bring your own sending server and pay pennies per thousand emails instead of a monthly ransom. And to be clear about where this sits: HighLevel stays the CRM. It runs my follow-up, my automations, my pipelines, all five businesses. Listmonk just handles the one job where per-subscriber pricing gets punishing.

• Excalidraw replaces Miro. 130K stars, no account required, runs on your machine, and it's already wired into MCP by a lot of people — meaning Claude can draw your diagrams for you. Roadmaps, flows, system maps, whatever you'd normally pay a seat for.

• Dub replaces Bitly — with a caveat. This is the hard one. It's not a clean one-click Docker app; it wants two free accounts (Tinybird and Upstash) and about $10 a year for a domain if you want it running properly. But it tracks conversions, not just clicks, on your own domain, with an affiliate program built in. Bitly gives you five links a month on the free plan. Five.

Youtube: Kill Your Monthly SaaS Bills: 5 GitHub Repos to Replace Everything (https://www.youtube.com/watch?v=hzJXc1B90A0)

Before you go repo shopping, vet the thing. Is someone still committing to it this month? Does it have real star count, or eleven? Are there hundreds of contributors or exactly one guy who's about to get bored? Is there a company or foundation behind it? One dead repo installed on your machine is worse than the subscription you were trying to kill. And you can just ask Claude directly: "Is this repo safe? Is there anything in here that could harm me or my computer?" It'll run the check before it runs the install.

The best small trick in the whole build: tell Claude Code to create a desktop icon for each one. It writes a shortcut that starts Docker, spins up the container, and opens the app — so using your self-hosted Loom replacement is one double-click, exactly like the paid version you cancelled.

Now the honest part, because I'd rather you hear it from me than find out in month three: self-hosting is not free. You're trading a monthly fee for a machine, backups, and updates that somebody has to own. If that somebody is you, on your own laptop, for your own PDFs — great, take the $1,296 and go. If it's a ten-person team and nobody's monitoring it, you will spend more in chaos than you ever spent on the subscription.

Grab the repos here:

• Stirling PDF (Adobe Acrobat) 👉 https://github.com/Stirling-Tools/Stirling-PDF

• Cap (Loom) 👉 https://github.com/CapSoftware/Cap

• Listmonk (newsletter tiers) 👉 https://github.com/knadh/listmonk

• Excalidraw (Miro) 👉 https://github.com/excalidraw/excalidraw

• Dub (Bitly) 👉 https://github.com/dubinc/dub

Which is the real shift here, and it's bigger than five repos. The barrier to running professional-grade software you own was never the software — it's been free and public this whole time. The barrier was the install. That barrier is now a paragraph of plain English typed into a folder. So the question stopped being "can I run this?" and became "do I want to own it or rent it?" — and for the first time in your business, that's actually a choice you get to make.

 🧠 Open-Weight Models Just Took the Lead, and Nvidia Is Betting the Company on It

Three years ago the open-source models were a toy and everybody knew it. This week the token usage on one major AI gateway flipped from 71% closed-weight to 62% open-weight — and the largest chip company on earth is reportedly buying the place where all those open models live.

If you're paying per token to run anything in your business, this is the story that changes your cost structure.

Nvidia is reportedly buying Hugging Face — and the reason why is the whole plot.

Worth flagging clearly: this comes from The Information, and neither Nvidia nor Hugging Face has publicly confirmed it. But if it's real, look at what's driving it. OpenAI just showed off its first inference chip, codenamed Jalapeño, hitting up to 104x performance on open-weight models — a chip they'd otherwise have bought from Nvidia. Meta's building their own silicon. Google's had TPUs for years. Everybody big enough to make their own chips is making their own chips, and every one of them is a customer Nvidia is slowly losing.

So Nvidia's counter-move is to own the other side of the board. Hugging Face is where open models live and where people rent GPUs to run them. If open-weight keeps taking share, every business that switches goes straight to an Nvidia-owned company for compute. They're not defending the closed-model business. They're betting against it.

The usage data says the bet isn't crazy — with one caveat nobody's reading carefully.

Vercel's AI Gateway showed open models going from 28.4% of token volume two months ago to 62% today. Huge swing. But flip that same dashboard from tokens to requests and it's still 62% closed, 38% open. Translation: most people are still sending most of their actual jobs to the frontier models. Open models are just chattier — they burn far more tokens per task. The migration is real, it's accelerating, and it's earlier than the headline number makes it look.

Claude's memory now spans Chat and Cowork — which quietly makes Cowork a different product.

This is the one that changes your Tuesday. Anthropic shipped memory that works everywhere: what Claude learns about you in a regular chat is now the same memory Cowork uses, and the reverse. You can think a project through in chat, jump to Cowork, and it already knows the context — including when Cowork runs a task in the cloud while you're asleep. By default it won't store personal or sensitive categories like health, politics, or religion, and you can edit what it keeps. Anthropic also announced a browser built into Cowork, rolling out to Pro, Max, and Teams on desktop over the coming weeks, and Claude in Chrome is now generally available.

The reason that matters: an assistant that forgets you every morning is a chatbot. An assistant that remembers what you told it last Thursday is staff.

Rapid fire — the rest that matters for your business:

• GLM 5.3 Flash quietly obsoleted a chunk of the market. Launched in stealth as "Ox Alpha," it scores 63.4 on the coding benchmark — beating Opus 4.8 — at a fraction of the cost. Every model in the box above it on the price/intelligence chart is now both dumber and more expensive.

• Qwen 3.8 Flash landed at 125B parameters — big enough that you'll want cloud for it, but it generated a full test build for just over a penny.

• GPT-5.6 Soul is 20% cheaper for the next three months. If you're running anything at volume, that's a temporary margin window, not a permanent one.

• ChatGPT can now sign into websites on web and mobile using your password manager, without ever seeing the passwords. It'll go renew your license at the DMV site while you do something useful.

• ChatGPT tasks now trigger on events, not just schedules — a new Slack message, a specific Gmail, a GitHub change. Automation without a workflow builder.

• Apple's M5 Ultra is a home AI machine. 4.5x the peak GPU compute of the M3 Ultra, with configurations up to 512GB of unified memory that's usable as VRAM. It starts around $11,000, so nobody's impulse-buying it — but running frontier-adjacent models on your desk is now a purchase decision, not a fantasy.

• Perplexity shipped a local agent that runs small models on your own machine and only escalates to the cloud with your permission. Needs serious hardware (they name-check the ~$4,000 DGX Spark), but the direction is unmistakable.

• Gemini Omni 1.1 Flash took the top spot in the text-to-video arena at 10 cents per 720p output, with start/end frame control and 4K upscaling. Google also dropped a new transcription model and added flight tracking plus hotel booking directly into AI mode in search.

• Skild AI's S1 teaches robots 10-minute tasks from a single video with no fine-tuning — wrist cameras in, task replication out.

Here's what the pattern actually tells you. Four years ago the question was whether open models would ever catch up. This week they're inside single digits of frontier performance on real benchmarks, they're winning on token volume, they run on a desk if you buy the right desk, and the biggest chip company alive is repositioning its entire strategy around the assumption that they keep winning. The cost of putting real intelligence into your business isn't dropping gradually. It's collapsing — and the operators who build the routing habits now are going to be running the same systems next year at a third of the price, while everybody else is still paying frontier rates to have a model count characters in a headline.

Hugging Face Built a A $399 Robot Duck That Learns Tricks You Teach It

At first glance Microduck looks less like a serious robot and more like something that escaped from a very ambitious toy factory. It's 25 centimeters tall. It's shaped like a duck. It waddles. It roller-skates.

Then you look at the spec sheet and realize somebody looked at a tiny duck and thought yeah, give it sensors.

• 780 grams of absurdly over-equipped bird. Fifteen degrees of freedom, a front-facing camera, an 8x8 depth lidar, two IMUs, microphones, a speaker, NFC, Wi-Fi, and Bluetooth. Out of the box it walks, sits, crouches, kicks, and grabs objects with an articulated beak. Knock it over and it doesn't lie there waiting for rescue — it has a trained recovery behavior that gets it off the floor and back on its feet.

• Then it gets ridiculous. Snap on the roller attachments and the thing skates. Drive it with a game controller. Let it run autonomous behaviors. Tag objects with NFC and watch it react to them. Or get several Microducks in a room and hold races. Or robot football. Yes, that's a real listed use case.

• It's open source and built on reinforcement learning — which is the actual story. You're not stuck with whatever behaviors the manufacturer decided to ship. You train a new one inside a physics simulation, transfer the policy to the real duck, watch how it performs, adjust, and go again. The duck falls over in the sim so it doesn't have to fall over in your living room. Much.

• And then you share the policy. This is the part worth sitting with. Somebody teaches their duck a new movement. Somebody else improves it. Suddenly there's a whole community shipping tricks for the same $399 robot — the exact same pattern that made the GitHub repos in the AI Builds Corner worth installing, except now it's happening to physical hardware.

• $399, four colors, delivery before Christmas 2026. Cream, graphite, lavender, and sky. Pre-orders are open now.

What I like about it is what it isn't. It's not trying to be the giant humanoid that does everything and costs more than a car. It's a small approachable platform you can play with today, experiment with tomorrow, and maybe teach something nobody's taught it yet.

It waddles, it falls over, it gets back up, it skates, and it grabs things with its face. Not bad for a robot duck. Not bad at all.

💲Deals at a Glance

🚀 Claude Code Workshop — 4 Days, 8 Hours, $67: Build apps, AI agents, and automations from scratch — hands-on, start to finish. 👉 Register Now

🎓 NEW — The Automated Marketer Skool Community (LIVE NOW): The room where the building doesn't stop. Twice-a-week group calls with me, plus courses on vibe coding anything — HighLevel, Claude Code, Lovable, AI, and automations. Introductory pricing for a limited time, with a lifetime option on the table. Two ways in:

• 👉 Join the Free Community — free access, free course, daily community

• 👉 Go VIP — $47/mo intro — hot-seat support, advanced templates, inner-circle access

🏗️ Agency Build-A-Thon — Free: A complete AI agency built from zero on GoHighLevel — domain, white-label SaaS plans, a client-ready website, voice + chat AI, the whole back end — recorded live, every step in order, nothing re-staged. Your start-to-finish blueprint if you sell to clients (or want to). 👉 Watch Free

🚀 Business Build-A-Thon — Free: Two real businesses built online with AI in one afternoon — domain, website, automations, and the AI that answers the phone, all from a blank screen. For launching your own business, not clients'. 👉 Watch Free

🛠️ Done-For-You — Built by Me and My Team at Rapid Active Marketing: Don't want to build it yourself? Skip the learning curve and have us build the whole thing for you — GoHighLevel setups, websites, funnels, automations, and AI voice + chat widgets — done right, by the people who actually teach this. We take on a handful of clients at a time, on purpose. 👉 See What We Build

💰 The Best GoHighLevel Affiliate Offer on the Planet: Everything you need to win with GHL, in one place. 👉 Get Started

🔥 Vibe Coding Incubator — FREE: Hands-on vibe coding, live weekly trainings, app templates, support, and Lovable how-to's. 👉 Join Free

😲 GoHighLevel Masterclass on Udemy — $20: 5,000+ students, 4.7★, 18+ hours of content. 👉 Grab It

10x the context. Half the time.

Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.

Try Wispr Flow free

Remember, there’s always something new coming from the Automated Marketer YouTube Channel—new content drops every Tuesday & Thursday. Subscribe so you never miss the next big idea.

Keep watching, keep learning, and keep moving forward!!

Nuno