What it is: A 24/7 AI trading agent built in Claude Code using Opus 4.7 + the new routines (scheduler) feature, trading via Alpaca paper account with a goal of beating the S&P.

Background context:

• Prior video: Nate ran Opus 4.6 in an OpenClaw agent for 30 days, beat S&P by ~8% on a $10K paper account.
• Opus 4.7 shows ~4% jump on Agentic Financial Analysis benchmark vs 4.6 — motivation for upgrade.
• This build ditches OpenClaw/Hermes entirely — pure Claude Code.

How it works:

• Scheduler: Claude Code *routines* fire on cron-like triggers — pre-market, market open, midday, close.
• Model: Opus 4.7 (pitched as built for "full throttle agentic work, judgment over ambiguity, self-verifying outputs").
• Skills: Custom skills built as-needed — research, decision-making, trade placement, logging.
• Research: Perplexity API (web_fetch/web_search would also work).
• Execution: Alpaca API endpoints (not MCP) to place trades on paper account.
• Journaling: Agent writes context to files between runs so it "learns" across sessions.
• Reporting: End-of-day summary pushed to ClickUp daily.

Setup steps covered:

• Alpaca: sign up → Trading API → paper account ($100K default) → regenerate API keys, save both key AND secret (secret only shown once).
• ClickUp: webhook/API for notifications (interchangeable with Slack/Telegram/WhatsApp).
• Perplexity: API key for research calls.

Tools & links:

• Alpaca — brokerage API, free paper trading with $100K, also has MCP server option
• Perplexity API — research layer
• ClickUp — notification destination
• Skool — AI Automation Society Plus — paid community
• Skool — AI Automation Society (free) — free resources
• Uppit AI — Nate's agency
• Claude Code routines — covered in a separate Nate video (linked in-video)

Why it matters for you: Concrete recipe for wiring Claude Code routines + Opus 4.7 into a long-running agentic workflow with external APIs (Alpaca/Perplexity/ClickUp) — the pattern generalizes far beyond trading to any scheduled autonomous agent you want to stand up.

What it is: A skeptical breakdown of the Opus 4.7 release — investigating whether it's a genuine model upgrade or Anthropic quietly nerfing 4.6 to make 4.7 feel better by comparison.

Act 1 — What happened to Opus 4.6:

• AMD senior director analyzed ~7,000 Claude Code sessions
• Thinking depth collapsed 73% (2,200 → 600 characters of reasoning)
• Model stopped reading files before editing — failure rate jumped from ~6% to 33.7%
• Users had to interrupt the model 12x more often to prevent it going off the rails
• Reports of hallucinated git commit hashes, fake package names, fabricated API versions
• Word "simplest" appeared ~3x more often in output (optimizing for least effort)
• Premature task abandonment — model stops mid-task
• Power users burning through $200/mo Max plans in an hour

What Anthropic actually changed (not the model weights):

• Feb 9: Adaptive thinking introduced — model dynamically decides thinking budget per turn, sometimes allocating zero reasoning tokens
• Boris Cherny (Claude Code creator) confirmed: turns with zero reasoning = fabricated outputs; turns with deep reasoning = correct
• Silently dropped effort default to medium — weights/training unchanged, just runtime parameters dialed down
• As of April 15, Pro and Max users were still defaulting to medium effort
• In the new Claude desktop app: Opus 4.7 doesn't expose the effort slider (low/medium/high/max) that Sonnet and 4.6 had

The core claim: Opus 4.6 didn't get dumber — Anthropic turned off its thinking and reduced effort without telling anyone, then launched 4.7 addressing exactly those complaints.

Tools & links (from description):

• AI Automation Society Plus (paid) — Nate's course/community
• AI Automation Society (free) — free resources
• Nate's YT Podcast application
• Uppit AI — Nate's agency

Why it matters for you: Less a tutorial and more a heads-up — if you're relying on Claude Code for serious engineering work, check your effort/thinking settings explicitly rather than trusting defaults, and treat model version bumps with skepticism when the underlying weights may not have changed.

What it is: A mini-case on Taylor Sheridan (Yellowstone creator) using vertical integration to stack income streams on top of a single creative product.

The Sheridan stack:

• Writes the show (script revenue)
• Owns the filming locations — Four Sixes Ranch & Bosque Ranch (reportedly ~$50K/week to Paramount just for property access)
• Owns the livestock used on set — $25/cow, $2,000/horse per appearance
• Runs "Cowboy Camp" — reportedly $214K to train actors to ride/rope
• Result: Paramount pays him coming and going on every episode

The pattern (vertical integration):

• Don't just sell the output — own the inputs and the venue
• Each layer you own converts a cost center into a revenue line
• Studios/customers keep paying the same person at every step of the chain

How a side hustler copies it:

• Rent a commercial unit → buy the building → sublease to other tenants
• Rent a chair in a salon → buy the salon → rent chairs to other stylists
• Use a product in your service → white-label it → sell it to customers and competitors
• Freelance with a tool → build the tool → license it to other freelancers

Tools & links:

• Four Sixes Ranch — Sheridan's Texas ranch, filming location for 1883/Yellowstone
• Bosque Ranch — Sheridan's Weatherford, TX production property

Why it matters for you: The side-hustle trap is trading time for cash on someone else's platform; the escape is owning the next link in the chain so the same customer pays you twice (or five times) for the same transaction.

What it is: Matt's personal video editing workflow for his AI news channel.

How it works:

• Live editing via Stream Deck — switches camera/screen in real-time while recording
• Records 1-2 hour raw, trims to 20-30 min finals
• Recut auto-removes silence gaps in one click (65min → 27min demo)
• DaVinci Resolve + Greg's Presets (paid) for zoom effects via Fusion overlay green box

Tools & links:

• Recut — auto silence removal
• DaVinci Resolve — free video editor
• Greg's Presets — paid DaVinci preset pack
• Stream Deck — hardware for live scene switching

Why it matters for you: It doesn't. This is content creator ops, not web dev. Skip.

What it is: A sub-2-minute AMA snippet where Matt dismisses the "make AI shorts, get rich" pitch. Says ~1% actually make money; winners still need storytelling, visual polish, and heavy iteration.

How it works:

• Viewer asks if an app that auto-generates and posts AI shorts is viable
• Matt: technically possible, not a great idea
• Breakout AI channels (Fruit Love Island, Italian brainrot) involve real story/process work
• Lower effort than filming, but still demands model testing and optimization

Tools & links (name-dropped, no URLs given):

• Veo 3.1 — Google's video gen model
• Kling 3.0 — Chinese video gen model
• Opus Clip — long-form → shorts clipper
• n8n — workflow automation to stitch + publish
• Cursor / Claude Code — to script the pipeline

Why it matters for you: Nothing here for web app builders. No architecture, no SaaS angle, no dev workflow insight — just a creator economy take. Auto-skip future Matt Wolfe AMA shorts.

Featured guest: Charles Duhigg — journalist and author of *The Power of Habit*, which has sold millions of copies and reshaped pop psychology around behavior change. Sam Parr credits the book with helping him kick a 20-beer-a-day drinking habit.

The core framework — the Habit Loop:

• Cue → trigger for the automatic behavior
• Routine → the behavior itself
• Reward → what your brain gets out of it
• ~40-45% of daily actions are habits running on autopilot
• Neural pathways thicken over repetition, which is why white-knuckling willpower fails

Key insight — don't extinguish, replace:

• Dr. Ann Graybiel (MIT) research: habits persist in rats' brains for years even after removal from the trigger environment
• You can't delete a habit, only overwrite the routine while keeping the cue and reward
• Sam's example: craved sugar after quitting alcohol → ate M&M's instead of beer → then swapped M&M's for non-alcoholic beer

Keystone Habits (term Duhigg coined):

• Borrowed from his biologist wife's concept of 'keystone species'
• One small habit that cascades into broader behavior change
• Sam's weight-loss keystone: slept in workout clothes with shoes next to bed — feet hit floor, go into shoes, out the door
• Writer David Epstein uses the same trick
• The cue removes friction; execution becomes autopilot

How Duhigg makes money:

• Book royalties (*The Power of Habit*, *Smarter Faster Better*, *Supercommunicators*)
• Speaking engagements
• Journalism (Pulitzer winner, New Yorker staff writer)

Tools & links:

• Charles Duhigg on X
• Sam Parr on X
• HubSpot sponsor link: clickhubspot.com/jj20

Why it matters for you: If you're engineering-brained, the habit loop is essentially a state machine for behavior — identify the trigger, swap the function body, keep the return value. Useful for installing focus rituals, exercise, or killing doomscroll loops.