A 944-token mistake, $539K from PDFs, and the LinkedIn toggle that takes 2 minutes.

​If having trouble with any links, you can read the online version here​

Hey guys,

Look, let me cut to the chase. I went back through last 2 months edition I've sent and pulled the 5 pieces that people actually replied about, saved, or told me they used.

One of them starts with a guy measuring his own setup file and finding it burned 944 tokens on every single message. He swapped it for something costing 53 and got better output.

That gap, between what we assume something needs and what it actually needs, runs through all 5 picks.

TL;DR

#1 Ross Mike on why your CLAUDE.md file is dead weight and the loop for building skills that work.

#2 Lucca's "rule of one" behind $539,441 in PDF sales.

#3 Nathan Gotch on why Apple dominates AI answers without ranking for its own keyword.

#4 Mikey's 5 principles for agents that survive real usage.

#5 Harry Long's 2-minute LinkedIn setting that turns every connection into a follower.

➡️ Your Complete Implementation Guide

Not got time to watch the videos? Grab the resources (Summary, Insights & Checklist) for each video from this Google doc

👉 ​Access Your Complete Implementation Guide Here​

1 — Stop Writing Instruction Files. Start Teaching Your Agent Instead.

Ross Mike has a 100% hit rate getting agents to do specific tasks, and he does it by refusing to write skills upfront. He walks the agent through the workflow first, like training a new hire.

What you'll discover:

• Why skills only load their name and description into context (53 tokens) while a CLAUDE.md file dumps its full 944 tokens on every turn
• The recursive loop: run the workflow by hand with the agent, get one successful run, then tell it to review what it did and write the skill
• Why he refuses to download other people's skills, and how his report generator pulls from 8 data sources flawlessly after 5 rounds of fixing failures

🎥 Watch here: Click here​

2 — $539,441 From PDFs, Built On One Rule

Lucca's first ebook flopped because he made it comprehensive. His fix was narrowing everything until it hurt, and the framework fits in four words.

What you'll discover:

• The rule of one: one problem, one person, one page, one traffic source (most people break all 4)
• Why "how to increase bat speed in 30 days" beats "how to get better at baseball," and how to use AI to find that specific subtopic inside your niche
• His ad setup: 1 campaign optimized for purchases, 3 to 5 creatives, broad audience, then let Meta find the buyers

🎥 Watch here: Click here​

3 — Apple Doesn't Rank For "Best Laptops." It Still Wins Every AI Answer.

Nathan Gotch ran the numbers after the June 2026 spam update and found Apple with 23 brand mentions across 37 citations, without ranking on page one.

What you'll discover:

• Why AI answers combine training data with retrieval, so third-party validation beats self-serving pages on your own domain
• How to export the citations behind an AI answer and find the 5 to 7 URLs that stay stable across months
• The local split: roughly 90% of effort into Google Business Profile and the local pack, 10% into third-party directories that ChatGPT actually pulls from

🎥 Watch here: Click here​

4 — The 5 Principles That Separate Working Agents From Abandoned Ones

Mikey argues most people stall at the same spot: the code looks great, then the config files, API keys and deployment settings pile up and the project dies.

What you'll discover:

• Principle 1 in practice: describe the outcome ("monitor my inbox, categorize emails, draft responses") before touching any structure
• Why integrations built by hand become technical debt the moment a tool changes its API, and how connecting Gmail or Google Calendar directly sidesteps it
• The 5 mistakes that kill builds early, including starting too big and mistaking impressive-looking code for progress

🎥 Watch here: Click here​

5 — 0 to 10,000 LinkedIn Followers, Starting With One Settings Toggle

Harry Long grew to 10,000 in 3 months with no paid promotion. Step one takes 2 minutes and converts every connection you already have into a follower.

What you'll discover:

• Where the switch lives (settings, then visibility, then followers) and why rejected connection requests still leave you with a new follower
• His posting rhythm: 3 posts a week, different formats and energy, following Alex Hormozi's give-give-give-take ratio of 4 to 1
• Why connection requests sent with no note get accepted more often than ones with a pitch attached

🎥 Watch here: Click here​

Before you go

Nathan's point about brand visibility only pays off if the page you're pushing actually deserves the spot. If you've got something parked on page two, Skyscraper AI pulls apart the current top result, shows you what it's missing, and helps you build a version strong enough to take its place.

Have a look here 👉 Click here​

Pick one of the 5 and do it today. Probably #5, since it takes 2 minutes.

Talk soon,

- Tony

P.S. If you only act on one pick this week, make it #1. Run your next repeatable task with the agent step by step, then have it write the skill. You'll feel the difference on the second run, and every week you delay is another week of paying 944 tokens for something worth 53.