What it is: Claude Code now supports "routines" — prompts that run on Anthropic's web infrastructure on a schedule, via API call, or triggered by GitHub events, without your laptop open.

How to set them up:

• Create routines from the Claude Code terminal (`/schedule`), the web at claude.ai/code, or the desktop app
• Each routine is a one-shot prompt — no interactive Q&A, so your prompt needs to be self-contained
• Configure: name, prompt, model, repository, cloud environment, and cadence (minimum hourly)
• Routines require a GitHub repository to sync to — Claude clones the repo in the cloud, reads your `CLAUDE.md`, scripts, and skills, then destroys the clone after execution

The gotchas Nate hit:

• Secrets don't transfer — your `.env` is in `.gitignore`, so the cloud clone has no API keys. You must add them as environment variables in the cloud environment settings
• Access levels matter — default is "trusted" mode, but many automations need "full" access. Check permissions before wondering why things silently fail
• Connectors — you can wire up Slack, Gmail, etc. via the connectors panel, or just hit APIs directly with your keys in env vars
• Automations don't just "migrate over" — expect to troubleshoot prompt structure and permissions

Tools & links:

• Skool (paid courses + support) — Nate's paid AI automation community
• Skool (free resources) — Free tier of his community
• Uppit AI — Nate's agency / work-with-me page
• Claude Code web — Where you can create and manage routines in-browser
• ClickUp — used as the test target for his first routine (standard ClickUp API)

Key setup flow:

1. Create a new remote task in Claude Code (desktop, terminal, or web)

2. Write a self-contained prompt (no user interaction possible)

3. Point it at a GitHub repo so Claude has your project context

4. Create a cloud environment and load your API keys as env vars

5. Set access level to "full" if your automation needs network/write access

6. Set cadence (hourly, daily, weekdays)

7. Test and iterate — first runs will likely need prompt tweaks

Why it matters for you: This is the exact tutorial you need to turn Claude Code into a hands-off automation runner — Nate shows the real setup steps, the env var workaround, and the permission traps that'll bite you if you just click through.

What it is: Sell advertising space to local service businesses (plumbers, landscapers, roofers) on shared postcards, then mail them to targeted homeowners via USPS Every Door Direct Mail (EDDM).

How it works:

• Create a shared postcard (front and back) with ad slots for multiple local businesses
• Only one business per category — no competing ads on the same card
• Use USPS EDDM to mail to targeted zip codes (sort by higher income / single-family homes)
• Cost: ~25¢/home postage + ~23¢/home printing = ~48¢/home total
• 10,000-piece run costs ~$4,800 (postage + printing on thick 14-point card stock)
• Sell 3x4" ad slots at ~$600 each; front-side ads cover your costs (break-even)
• Back-side ad sales are pure profit — roughly 50% margin overall
• Close deals via Facebook Messenger or text — 95% of sales happen there, no calls

Finding customers (the Facebook trick):

• Uses a specific Facebook approach to get local business owners to DM him first
• No cold calling, no paid ads, no Zoom pitches
• One plumber paid $1,100 after just five text messages

Startup costs:

• Started for under $100 (first two cards sent without even having an LLC)
• Skip the logo, website, and branding — validate revenue first
• Can technically start for free if you hustle the sales before printing

Tools & links:

• USPS EDDM Tool — select zip codes, filter by income/household type, mail without a mailing list
• Thick 14-point card stock postcards — order from online printers shipped to your door

The business model compared to newsletters:

• It's a two-sided marketplace like a newsletter, but you only need to fill the demand side (advertisers)
• The supply side (eyeballs) is solved by EDDM — you can legally mail anyone for 40¢/piece
• No audience-building, no email list, no SEO grind

Why it matters for you: Dead-simple side hustle with near-zero startup cost, no technical skills required, and a clear path from first card to $50K/year — the kind of shiny object that actually has a repeatable playbook.

What it is: A short motivational punch arguing that performative "not caring" is a defense mechanism that keeps you broke, and that focused obsession is the actual antidote.

How it works:

• The core thesis: The internet sold you "anti-hustle culture" but the opposite of toxic hustle isn't apathy — it's directed obsession toward something you actually like
• The "I wasn't really trying" trap: Saying you weren't trying is just pre-loading an excuse for failure so your ego stays intact
• The 30-Day Obsession Challenge:
• Pick ONE thing — a skill, an income stream, a strategy
• Show up daily for 30 days straight
• No splitting focus, no hedging, no "I'll try"
• The point: compressed intensity forces real signal on whether something works

Key mindset shift:

• Broke mindset = protect ego by feigning indifference
• Rich mindset = go all-in, risk looking stupid, iterate fast

Tools & links:

• No specific tools or repos mentioned in this video

Why it matters for you: If you're dabbling in a side hustle 2 hours a week and wondering why nothing's moving, this is the framework — pick the one lever (cold outreach, content, a product) and hammer it for 30 days before deciding it doesn't work.

What it is: Alex walks through the 8 AI tools/subscriptions he still pays for after testing dozens, explaining what each does in his workflow.

The stack he kept:

• GitHub Copilot — His longest-running AI tool. Inline completions, chat, agent mode. Values reliability over flash. Access to Gemini, Claude, GPT, and local models from within the IDE.
• ChatGPT (Plus) — General-purpose go-to. Research, brainstorming, image work. Dropped the $200/mo Pro tier — wasn't worth it for him.
• Claude (Max plan) — Longer context, nuanced code review, better writing. Uses it alongside ChatGPT, not as a replacement. Pastes outputs between them to cross-check.
• ChatLLM Teams (Abacus AI) — Single interface for multiple models (GPT, Claude, Gemini, Grok, DeepSeek R1, Qwen 3.6, etc.) plus image gen (GPT Image 1.5, Flux, Cadream) and video gen (Sora 2, Kling, Veo, WAN 2.5). $10/mo. Solves the "too many subscriptions" problem.
• Deep Agent (Abacus AI) — Full-stack app builder with APIs and payments. He offloads presentations and reports to it.
• Recall — Knowledge base for saving trusted content (mostly YouTube videos, articles). AI answers grounded in your own curated sources.

Tools & links:

• ChatLLM Teams — Multi-model AI hub, $10/mo
• Deep Agent — Full-stack agent builder
• Recall — Personal knowledge base (code AZISK25 for 25% off, expires June 1 2026)

Transparency note: Three of the tools sponsored the video (indicated on-screen). He says he uses all of them regardless.

Why it matters for you: No hardware content in this one — it's all cloud/SaaS. If you're waiting for his next local inference or Mac Mini benchmarking video, skip this. The ChatLLM consolidation angle might save you money though.