You're Being Loyal to Software That Doesn't Know You Exist

I catch myself doing it. You probably do too.

You pick a model. You get good at it. You learn its quirks, you build your prompts around it, you start defending it in comment sections like it's your football team. And then somebody asks which AI is best and you answer instantly, with confidence, based on a test you ran in March.

Here's what I've been noticing lately, and it's changed how I work: the gap between the top models is closing fast. Not closed. Closing. I've been running the same prompts through Fable 5, through Kimi 3, through Claude, through GPT-5.6 Soul — and the results are landing closer together every month. Six months ago the spread was obvious. Now I have to look hard.

Which means the question stopped being which one is best. It's now which one for this, and what's it costing me.

Because here's the part nobody puts in the benchmark charts. Fable 5 is a beautifully built model. It's also a token furnace. In my own head-to-head tests it ran four to five times the cost of the alternative for output that was better — but not five-times better. Sometimes not better at all. On a dashboard build, the cheap one matched it. On social copy, the cheap one beat it.

That's not a knock on Fable. That's just what it looks like when you actually check.

And checking is the move I want to hand you. Stop running your work through one model and calling it done. Run it through a second one and ask it what's wrong with the first one's answer. Have Claude review what GPT built. Have GPT critique what Fable wrote. Two models disagreeing about your work is worth more than one model agreeing with you — and the disagreement is where you find the thing you would have shipped broken.

That's it. That's the whole shift. You're not choosing a side. You're building a panel.

The people getting real leverage out of this stuff right now aren't the ones who picked correctly. They're the ones who stopped picking.

Two ways to get in the room with me:

The Claude Code Workshop runs August 10. Over 2,000 people have come through it, and the results speak louder than I can. It's not too late to get in — but it sells out every single time we run it, and I'm not saying that as a line. I'm saying it because people email me after it closes and I can't do anything for them.

👉 Grab your seat — go.vcinc.ai

And if you want to build with me in person: AI Unleashed Intensive, Cancún, September 30 – October 2.

I went last year. It wrecked me in the best possible way — actually transformational, not the word people use when they mean "nice." All-inclusive, great hotel, better food. This year's lineup includes Molly Mahoney, Austin Armstrong, and a few more names you'll recognize. I'm personally running nightly hackathons plus a full demo day, so you leave with something you built, not a notebook full of notes you'll never open.

Early bird pricing is live right now and it climbs the longer you sit on it.

👉 Get your ticket →

Come build with me.

🛠️ The Last HighLevel Setup Video You'll Ever Need

You're paying for HighLevel right now and using none of it.

Be honest. The account's been open three weeks. Maybe three months. You log in, stare at a left-hand menu with forty things on it, close the tab, and go back to running your business the hard way — while the invoice hits your card every single month.

That's not your fault. It's what happens when every tutorial you find was made for an agency running 40 sub-accounts, and you've got one business. A handyman shop. A consultancy. A brokerage. A med spa. You don't need a white-label empire. You need a page that doesn't embarrass you, a calendar that fills itself, and follow-up that happens whether or not you remember.

So I stopped explaining pieces and built the whole thing. Bare domain to live, revenue-ready business. Twice. Once for home services, once for a consultancy — about three hours each, mistakes left in.

• Here's the part nobody tells you: the order is the product. Every person who fails at this fails because they built the website first. Follow the sequence below and it's one afternoon. Break it and you'll spend three days undoing yourself.

  • Domain and professional email come before you ever open HighLevel. $10 at Cloudflare. $8.40/mo for Google Workspace on the Starter plan — the one Google buries so you'll buy the expensive one. Running your business off a free personal address caps your sending, blocks your security records, and tells every prospect you're a hobbyist. Then stack free aliases — sales@, team@, info@ — under Users → Add Alternate Emails and activate them in Gmail. Four addresses. One inbox. One bill.

  • Three security records, or your email is dead on arrival. DKIM, SPF, DMARC. The inbox providers rewrote the rules in 2024 and this is the toll. Ten minutes now, or a month of "why is nobody opening my emails" later. Your choice, but there are only two.

  • Connecting your domain wrong will take your existing website down. Read that again. If your domain already points at another site builder and you drop the root into HighLevel, you overwrite those records and your site disappears. Use a subdomain — booking. or go. — and nothing breaks. Fresh domain? Use the root, add the www, move on.

  • Build two sending lanes. Never one. Create replies.yourdomain.com and no-reply.yourdomain.com, then open Domain Configuration and route by traffic: calendar notices, receipts, workflow sends, portal mail → no-reply. Conversations, campaigns, bulk → replies. Two lanes, two reputations. Burn one, the other keeps delivering. Everybody sends everything down one pipe and then wonders why their appointment reminders killed their newsletter.

  • Then set your headers to match the lane. Three-dot menu → Set Headers → from-address becomes [email protected], matching the server it's leaving from. When the from-domain matches the sending domain, the inbox providers stop squinting. One field. It moves delivery more than anything else on this page. And add your real inbox under Reply & Forward Settings — press Enter until it turns into a gray chip or it silently doesn't save.

  • Take the two free domains everyone leaves on the table. A branded link domain (link.yourdomain.com) so every link you send carries your name, not a software vendor's. And a portal domain (portal.yourdomain.com) for a community. Yes, even if you're a plumber. There's a woman running a pickle shop off a 30,000-person community. Build it now. Use it when you're ready.

  • Offer first. Website last. This is where 90% of people blow it. Nobody's short on leads because their site is ugly. They're short on leads because there's nothing on the page worth booking. Write the offer with AI, get the copy finished, then hand it to Build with AI so it uses your words instead of hallucinating its own. Duplicate the tab and let three pages cook while you build calendars in another.

  • Turn on schema markup on every page. Page settings → SEO → Schema Markup → generate. The AI reads your page and writes the structured data for rich results and AI citations. Free ranking asset. Serve multiple towns? One page per town. That's how local businesses actually win search, and it's why your competitor outranks you with a worse offer.

  • Then put a talking robot on your homepage. Build a Voice AI agent, make a chat widget set to embed inline with the visualizer, copy the code, drop it into a Code element on the page. Your customer clicks, allows the mic, and talks to your business out loud — and it books straight onto your calendar. Add the text bot in the corner with all three handover triggers on: asks for a human, runs out of info, getting frustrated. A bot that admits defeat closes better than one that keeps bluffing.

  Where it fights back: the AI added background images and entire sections I never asked for. It defaulted every button to "scroll to element" instead of "open popup." It swapped my form for a generic one twice. Cloudflare's auto-record handshake failed and I typed records in by hand. One voice agent per assigned number, so multiple agents means multiple numbers. Walk every button before you publish. It's still faster than doing it manually — it just isn't magic, and anyone selling it to you as magic is selling you something else.

  🔥 Pro Tip: Most people will run this on their own business, feel good for a week, and stop there. The actual play is running it for the three business owners already in your phone. Every one of them is sitting on the exact problem you had twenty minutes ago — a domain they bought, software they're paying for, and no clue what order any of it goes in. They will not solve it alone and they'd rather write a check than try. Sell them the sequence: "Business Launch Setup — domain, professional email, full deliverability config, live booking page, follow-up automations, and an AI agent that answers your customers. Live in 5 days. $2,500." You've now watched it built twice; your third run takes under three hours because the snapshot carries the pipelines and automations for you. Three a month is $7,500 for roughly nine hours of work — and each one converts into a $297/mo maintenance client who never fires you, because you hold their domain records and their automations. That's not a side hustle. That's a moat. —

  Here's the truth: the distance between "I own a domain" and "I own a business that books appointments while I sleep" is now a single afternoon — and it's the same afternoon whether you're a handyman or a consultant. Only the offer changes. Everything else is the same nine moves in the same order.

  So the account gathering dust isn't a HighLevel problem. It's a sequence problem, and you now have the sequence.

  The Funnel Bricks snapshot carrying the lead pipelines, the follow-up automations, and the onboarding funnel with every custom GPT link inside it is free in the Skool community. Start there. Don't start from blank.

 🤖 I Handed Claude My GoHighLevel Login and Left the Room

Thirty minutes later I had a live home services funnel, a working form with two dropdowns, an automation that tags the lead and emails me, and a mobile-optimized page I never touched.

I wasn't in the room. I was doing other work.

Here's the setup that made it possible. The GoHighLevel MCP is good — 60+ actions, contacts, conversations, opportunities, payments. But it has a hard ceiling: it cannot build funnels, websites, or workflows. Those are the exact things you spend your day doing. So the connector handles the boring half and leaves the expensive half to you.

That gap is where most people stop and complain. I closed it instead.

The move: build a skill that teaches Claude Cowork how to physically operate the app — then hand it browser and computer control and let it click.

• You write the skill before you write the task. Not a prompt — a persistent instruction set. I told it where the AI funnel builder lives (left menu → Sites), warned it that the AI automation builder chokes on anything complex, and made it include its own QA step. That last part is the difference between a demo and a deliverable.

• It directs the other AI. Claude opened HighLevel's Build with AI, wrote its own prompt for it, and read the output back. AI operating AI. That's the whole trick — you're not replacing the platform's builder, you're giving it a supervisor who never gets bored.

• It caught its own mistake. The page came back with a form that was coded into the layout instead of being a real HighLevel form — it would have collected nothing. Claude noticed, navigated to Sites → Forms, built an actual form, added the service dropdowns, renamed it, and dropped it into the page. Nobody told it to. That's the QA instruction earning its keep.

• It wrote SEO and AEO copy on its own. The business was in Denver, so it worked Denver HVAC, plumbing, and home repair into the page without being asked.

• It tested the thing it built. Filled the form with fake names and phone numbers, submitted it, confirmed the automation fired. I got a "new lead" email from a business that didn't exist forty minutes earlier.

• It checked mobile. Went back through and optimized the layout on a smaller screen. Not perfect. Better than most people ship.

The honest limits: you have to log into the browser yourself and land on the right account first — browser control doesn't authenticate for you. You'll need to tell it where buttons live (left menu, top bar, blue button on the right) because it's reading screenshots, not code. It took about 30 minutes for one funnel plus one automation. And the site it built was solid, not stunning. The fix is the same fix every time: go back and sharpen the skill. Every run makes the next one better, and the skill is the asset — not the output.

The bigger pattern is this: an MCP is no longer the boundary of what an AI can do inside an app. It's just the fast lane. Anything with a login and a screen is now automatable — certifications, podcast bookings, forms, portals, any tool that never bothered to build a connector.

👉 The skill is yours. But read Rule Two first.

I field-tested this on 2026-07-17 and found something that should worry you: Funnel AI builds fake forms. It renders your lead capture as a Custom Code element — raw HTML that looks right and does absolutely nothing. No submit handler. No contact created. Never shows up in Sites → Forms. I filled it out on a live preview, hit submit, and verified zero contacts twice. If you've shipped a Funnel AI page for a client, go test the form right now.

Workflow AI has the same disease. It writes correct structure and then invents references to forms, tags, and merge fields that don't exist — a workflow that looks perfect on the canvas and is completely inert.

The doc has all of it: the exact replacement sequence for the fake form, the three hallucinated-reference fixes, why {{form.service_needed}} should be {{contact.service_needed}}, and the one that cost me 20 minutes — multi-line prompts hang the builder. Same content as flat prose built a 10-section page in five.

Get the GHL AI Builder SOP →

 🧠 The Week the Benchmarks and the Users Stopped Agreeing

Claude Opus 5 shipped, scored near the top of the charts, and got publicly dragged by half the people using it. That gap — between what a model measures and what it feels like to work with — is the actual story this week, and it changes how you should be picking tools.

Opus 5: cheaper, faster to your limits, and wildly divisive. On benchmarks it lands close to Fable 5 across terminal coding, knowledge work, novel problem solving, agentic search, and computer use — at meaningfully lower cost, which means you hit rate limits far slower. On paper that's the value play of the quarter. In practice, the loudest practitioners online went the other way hard: verbose, scattered, treats every minor comment like a critical bug, doesn't answer the question you actually asked. Multiple people publicly rolled back to 4.8. Worth saying plainly — these are vibes from heavy users on X, not a controlled study, and plenty of people are getting great results. But there's one thing nobody's disputing: Opus 5 is unusually good at generating 3D graphics. Matt Schumer one-shot a Call of Duty-style game that looks borderline AAA, shared the prompt on GitHub, and set off a wave of people building their own. One tester ran the same prompt for 11 straight hours overnight and woke up to a world with real mountains, trees, and water. The gameplay was broken — controls reversed, attacks that didn't land — but the environment was legitimately beautiful. Run the identical prompt through GPT-5.6 Soul and you get a more playable game with noticeably worse visuals. Same prompt. Different strengths. That's the lesson.

Jack Dorsey built a Slack where you invite the agents in. It's called Buzz — decentralized, open source, free right now. You install Claude Code, Codex, and Goose as agents, assign each one a different model, then tag them in a channel like coworkers. In one test, three agents each built a webpage, then got asked to critique each other's work — and started replying to each other's feedback in-thread. Then you tell one to go fix what another flagged. Human channels, agent channels, and mixed channels, all in the same workspace. This is the first version of multi-agent collaboration that looks like an actual office instead of a demo.

Nano Banana is now inside Google Earth, and it's free. Pull up any location in 3D, angle the camera, hit Create Image, and describe what you want. Google's own examples: Pompeii rebuilt as it stood in 78 AD, a neighborhood rendered as it'd look post-development, professional-grade real estate visuals. Technically you could always screenshot Earth and feed it to Nano Banana yourself — but the friction is gone now, and free tools with zero friction are the ones that actually get used.

Rapid-fire, and what it means for your business:

• Meta AI got agentic connectors — hook up your calendar and email, ask what's coming next week, have it draft replies across your whole inbox without sending. Late to this, which is the point: agentic inbox and calendar control is now standard everywhere, not a differentiator.

• Grok added Build mode for making and sharing sites, apps, games, and dashboards — locked behind the $100/mo tier.

• Gemini on macOS added natural-language dictation that looks a lot like Whisper Flow. Free voice-to-text is quietly becoming default OS behavior.

• Google's Omni video model was free for 10 generations through August 4 — if you're reading this the day it landed, you have hours. After that, back to paid.

• MidJourney v8.2 shipped. Officially bolder and edgier. Practically, hard to spot the difference.

• Mirage Avatar X claims the best identity preservation in AI avatars at ~$30/mo, and HeyGen turned NotebookLM's podcast trick into video — two-host shows with studio scenes, multicam cuts, and B-roll from any document. Both impressive. Both accelerating the flood of low-effort video.

• LinkedIn added a "seems like AI slop" button. Read that as a signal, not a feature. The platforms are now actively sorting for human-sounding writing, which is exactly why your emails shouldn't read like everyone else's.

• Gemini Robotics ER2 dropped, and the demos are the real thing — a robot bagging grapes without crushing them, unscrewing a lightbulb intact, tying a trash bag. Meanwhile the US is blocking new Chinese humanoid imports over surveillance concerns, right as China leads the category.

Here's the pattern under all of it: the "best model" question is dead. Opus 5 builds better worlds and loses to GPT-5.6 on making them playable. One model runs your inbox, another runs your builds, a third writes your copy — and in Buzz they're now sitting in the same channel arguing about each other's work.

Stop shopping for a winner. Start building a bench.

I Made Fable 5 and GPT-5.6 Soul Fight Five Times

Everyone online has an opinion about which model is better. Almost nobody has run them side by side on the same prompt and counted the receipts.

So I did. Five tests. Identical prompts. Fresh folders. Fable 5 at max effort, Codex running 5.6 Soul in Ultra mode. Straight API math, no subscription cushioning the numbers. Every prompt structured the same way — goal, task, why, outcome, constraints, verification — and looping until the model verified its own work.

Here's what happened, and I'm genuinely not sure who won.

Here's what happened, and I'm genuinely not sure who won.

• Round 1 — Build a GTA 2 clone. GPT gave me stick figures, decent maps, smooth transitions. Fine. Fable gave me characters with legs, a marked car to steal, a heat level, skid marks, and smoke pouring out of the car after I wrecked it. Then the car exploded and it said WASTED. Fable: $16.86, 39.8 minutes. GPT: about $3, 30 minutes. Fable clearly better. Also five times the price. (Also I am terrible at this game and you all had to watch that.)

• Round 2 — "Most impressive scroll-stopping website you can imagine," one HTML file, no other direction. Fable built a thing with sound that climbs as you scroll, taking you from an atom out to the solar system to the Milky Way. GPT went poetic and practical — clock imagery, a page that visually breaks as you scroll down. Fable: 6M+ tokens, 27 API calls, $23.37 — and it finished faster. Which makes no sense given Round 1, and I'm still annoyed about it.

• Round 3 — Agency dashboard. Nearly identical output. GPT asked for more direction before starting. Fable ran about $27. GPT ran dramatically cheaper for the same result. No contest on value.

• Round 4 — Seven social posts in my voice, publish-ready, not drafts. Fable graded itself a 9.5 and rewrote twice. Solid. But GPT beat it — better formatting, tighter posts, and honestly better than it usually does at writing. Then it dropped #agencylife into an Instagram caption. No agency owner has ever typed those words. Close, and then a hashtag gave it away.

• Round 5 — Analyze my real YouTube channel from a CSV. Both built genuinely good dashboards. Both found the same retention cliff in late June and gave me nearly identical advice. Neither of them knew the cliff was because I was on vacation for two weeks. Both of them made me copy and paste a link to find the report, which is a small crime.

The scoreboard: Fable 5 is still ahead on creative work — not close, in my opinion. GPT is better on data and building, and it costs a fraction as much. And 5.6 Soul Ultra feels comparable to Opus 4.8, not to Fable. The real comparison was probably 4.8 all along.

I'm posting the full prompt set so you can run it yourself. If you get different results, I actually want to hear it — my sample size is five and my GTA skills are worse than that.

💲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