How I Use Claude Code to Make INSANE Instagram Carousels
A step-by-step demo of turning a Pinterest screenshot into a branded, reusable carousel system using a design system MD file, Higgsfield MCP, and a slash-command skill.
May 28thHow one creator handed Claude a game plan, walked away, and came back to a working agentic OS.
UltraCode's three-layer agent architecture — orchestrator, specialist agents, and checker agents — lets you hand Claude a structured plan, leave the room, and return to a working application, shifting the bottleneck from prompt skill to human patience.
UltraCode is a three-layer multi-agent system where an orchestrator spawns specialist agents, each paired with a dedicated checker that validates work before it reports back — enabling long unattended coding sessions. The practical workflow is: brainstorm scope in Claude chat, drop a design system file into the project folder, use Plan Mode with Sonnet to generate a structured .md game plan, then switch to Opus 4.8, enable UltraCode, and execute the plan file. A first pass produces a functional but flat dashboard; a second natural-language refinement pass adds bento-box layout and an embedded terminal. The real tradeoff is cost: Opus 4.8 plus UltraCode is the most token-intensive path available.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →
Hook, product demo preview, host credentials (art director for Apple, PlayStation, Nissan; 6-figure AI agency).

Three-layer architecture explained: orchestrator, specialist agents, checker agents. On-screen diagram shown.

Use Claude chat to brainstorm AIOS scope; Claude pulls from conversation memory to suggest personalized modules.

Download Antigravity IDE, create aios folder on desktop, open Claude in terminal.

Shift+Tab enters Plan Mode. Uses Sonnet 4.6 for cheap planning before switching to Opus 4.8 for execution.

Copies existing HTML design reference and design.md rules file into project folder for automatic brand inheritance.

Claude asks interactive scope questions: areas to track, interface type, update frequency. Selects web dashboard, always-on live mode, YouTube + LinkedIn + Instagram.

Instructs Claude to save the structured plan as a .md file before coding begins.

Clear chat, /model switches to Opus 4.8. Type effort to toggle UltraCode on.

Sends execute the plan in plan.md and steps away while UltraCode spawns sub-agents autonomously.

Returns to a working dashboard: content intel left, YouTube metrics center, active projects and skills launcher right. Honest critique of flat visual hierarchy.

Shows Skool community where the AIOS-builder prompt lives at $9/month.

Requests bento-box grid, varied backgrounds, embedded terminal in plain English. UltraCode runs a second pass.

Honest callout: Opus 4.8 plus UltraCode is the most token-intensive path. Worth it for big projects; track usage carefully.

Refined bento-box dashboard with draggable panels, competitor tracking, YouTube stats, embedded skills terminal.
The real unlock in multi-agent coding is not the agent itself — it is the structured plan you hand it before you let it run.
“We are going to use two of its newest and most insane features to build an AIOS with as little human involvement as possible.”
“It actually spawns up another layer of sub agents to check and verify the work of the actual agents performing the task.”
“While this was working, I actually took my dog to the vet for his annual checkup — and Quad basically built all of this for me.”
“It is the most cost intensive way to use it. It's going to use the most tokens. It's going to cost the most money.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
A former Apple and PlayStation art director takes Claude Opus 4.8's new UltraCode feature on its first real test: build a full agentic operating system with almost zero human involvement while he takes his dog to the vet. He comes back to a working dashboard.
Three-tier system where orchestrator assigns tasks to parallel specialists, each monitored by a dedicated checker that validates output before it reports back.
Five-step workflow from blank folder to working agentic OS dashboard, using cheap planning first and expensive execution only after scope is locked.
“If you wanna create an AIOS for yourself, I just built this super long prompt. All you have to do is copy and paste this into Claude, answer a couple questions, and you're gonna have your own dashboard ready to go.”
Mid-video CTA at natural content pause after first build reveal, repeated at close. Points to Skool community at $9/month — extremely low friction.
00:00
00:16
00:29
00:33
00:45
00:55
01:14
01:23
01:35
01:41
01:52
02:09
02:20
02:35
02:37
02:50
03:06
03:12
03:26
03:42
03:54
04:06
04:18
04:31
04:45
04:50
05:04
05:14
05:25
05:46
05:52
06:06
06:19
06:35
06:55
07:03
07:22
07:38
07:54
08:12
08:18
08:28
08:32
08:43
08:51
08:59
09:06
09:12
09:19
09:21
09:37
09:48
09:59
10:11
10:21
10:32
10:43
10:54
11:04
11:08
11:21
11:31
11:42
11:56
12:01
12:13
12:21
12:37
12:48
12:59
13:05
13:21
13:31
13:43
13:59
14:06
14:18
14:30
14:43
14:54A step-by-step demo of turning a Pinterest screenshot into a branded, reusable carousel system using a design system MD file, Higgsfield MCP, and a slash-command skill.
May 28thA 15-minute walkthrough of the exact four-level Claude system that grew one creator from 1,200 to 14,847 LinkedIn followers — ending with a fully autonomous post-and-DM machine.
May 22ndHow the worker-plus-evaluator loop actually works, why most devs will write it wrong, and the good-condition pattern that makes it finish for real.
May 14thA 9-minute demo showing how native cron sessions turn any Claude Code skill into a self-healing, self-improving background agent.
March 7thA 12-minute walkthrough of three agentic workflows built around one principle: a second verification pass is what turns a fast answer into one you can trust.
June 1stA 21-minute walkthrough of spinning up three competing AI agents that debate your ideas and write their reasoning to a shared file.
January 23rd