Video summary

The NEW Agentic OS standard for Claude 5 Models is here (Full Breakdown)

Main summary

Key takeaways

Technology

Technological Concepts: “Agentic OS” for Claude 5 (Framework + Dashboard)

The video argues that modern Claude 5–class models are far more capable, but most people’s agent setup (agents + OS/workspace organization) hasn’t caught up.

To address this, the creator introduces an “Agentic OS” concept—specifically a virtual command center dashboard—designed to make agent workflows:

  • Faster
  • Cheaper to run
  • More productive

Product/System Features Shown (Dashboard + Tooling)

Agentic OS Dashboard (Home / Command Center)

A central view that includes:

  • Calendar/time zone summaries
  • Email summaries, including what Claude flags as needing attention
  • Quick links to “micro apps” built by the creator
  • Custom widgets (e.g., a YouTube widget)
  • Routines/scheduled tasks with:
    • what will fire
    • when it will run

“Skills Deck” Widget

  • Adjusts effort level and model per skill run
  • Triggers skills directly from the dashboard

Widget Management

  • Widgets are created via Cloud Code
  • Supports adjustable size/placement and the ability to create new widgets

“Artifacts Ring”

  • Uses prior Claude Code artifacts
  • Searchable by client name
    • Example: opening a stored HTML artifact for a specific client

“Second Brain” Visualization

  • A click-through system that links:
    • skills
    • files
    • workspace structure
  • Intended to help users explain and present systems to others (useful for consulting/clients)

Skill-Triggering + Artifact Generation Workflow

  • Uses Cloud Code to consistently generate artifacts such as:
    • HTML
    • PDFs
    • reports
  • Emphasizes that the dashboard is only ~20–30% of the overall value.
  • The remaining ~70% is how context/workspace is organized underneath.

The Guide/Tutorial Structure: “ARMS Framework” (4-Part OS Organization)

The setup is broken into four bottom-up elements, each with increasing sophistication levels:

  1. Skills
  2. Memory
  3. Routines
  4. Applications

The video also provides a 9-page PDF guide (suggested to be uploaded to Cloud Code) containing prompts and setup steps.

1) Skills (How to Build and Structure Agent Actions)

What Skills Are

  • Skills are described as shortcuts to SOPs
  • Heuristic: if you prompt Claude for the same task twice, turn it into a skill

Level 1: Use Pre-Built Skills (Entropy/Entropic + Cloud Desktop)

  • In Cloud Desktop app → Customize → Skills, browse pre-built skills from Entropic
  • Example highlighted: a “skill creator” skill for generating new skills quickly

Level 2: Skills as “Bundles” with Richer References

A skill can be more than a single skill.md:

  • A “thin” skill may contain only one skill.md
  • A more complex skill (e.g., /robo) may include:
    • a folder of multiple files
    • a skill.md that acts like a router pointing to other reference files

Uses visual/design references, such as brand guidance (fonts, color palettes).

Demonstrates generating a branded PDF guide quickly using /robo.

Level 3: Trigger Skills “Headlessly” (Outside Chat Sessions)

  • Uses a Claude feature called Claude P to run a skill via a oneshot prompt
  • Motivated use cases:
    • run a “cleanup” skill when devices slow down
    • integrate skill execution into dashboards/internal apps

Key point: minimal extra technical tooling is required—just knowledge of Claude P.

2) Memory (Workspace Organization for Agent Performance)

Problem Addressed

As workspace/context grows:

  • retrieval becomes slower
  • the chance of hitting plan/usage limits increases
  • navigation and retrieval degrade over time (example: a “second brain” folder with ~60,000 files)

Level 1: Raw File Workspace

  • Start with a folder/workspace containing many files
  • Works initially, but degrades later

Level 2: “Agent-Optimized” Organization with Router Files

Instead of human-centric folder naming/navigation, the creator recommends router markdown files:

  • A central router like claw.md that provides context for “departments”
  • Department-level routers (e.g., content.md) that list:
    • relevant skills
    • reference files

Rationale:

  • Agents parse quickly, so the best structure is to give direct pointers to what matters

The video also includes a prompt approach to generate router files via Cloud Code.

Level 3: Visual “Second Brain” System

Adds visualization so users can:

  • see how files/folders connect
  • search and preview faster than traditional file explorers
  • understand systems better when explaining to others (helps visual learners)

3) Routines (Scheduled Tasks That Run Agent Actions)

What Routines Are

  • Routines = scheduled tasks that send a prompt to the model at set times

Level 1: Local Routines in Cloud Code (Computer Must Be On)

  • In Cloud Desktop → Routines, draft routines in natural language
  • Example:
    • daily YouTube → Substack newsletter drafting at a specified time
    • uses artifacts generated by the routine for later review

Benefits:

  • increased confidence because routines can use a custom skill, reducing editing to “minor edits”

Limitation:

  • local routines only run when the computer is on

Level 2: Cloud Scheduled Tasks (24/7 Offloaded Agents)

Mentions external options:

  • OpenClaw (popularized earlier)
  • Grockbot (new, paywalled, expensive)

Chosen solution:

  • Hermes agent (24/7 always on via a cloud computer)

Critical Integration Detail: Sharing Skills/Context with Hermes

If Hermes has its own machine, it must access the same workspace/context (skills + memory files).

  • Uses Syncthing to sync files between:
    • Cloud Code workspace
    • Hermes computer

Level 3 (Near-Future / Advanced): VPS / Cloud Code on One Persistent Environment

Suggests using a VPS with Cloud Code installed so routines and context live in the same always-on environment.

Claims:

  • avoids external syncing tools (no Syncthing)
  • notes future potential for official integration, constrained by file storage/security

4) Applications (Connect Agents to Tools + Build Your Own)

Level 1: Connect via Cloud Desktop “Connectors” UI

  • In Cloud Code desktop → Customize → Connectors, connect apps through UI
  • The creator argues this is less efficient than dynamically searching connectors

Level 2: Search and Select Connectors Using a Skill

Uses a custom skill called “search connectors”, which searches for:

  • official connectors
  • community-made connectors (CLI, APIs, MCPs)

Example: for Adobe Premiere

  • checks official availability
  • recommends an open-source GitHub repo
  • Claude can scan/check whether it’s safe to set up

Level 3: Build Custom Connectors and Micro Apps

Encourages creating connectors for tools without existing integrations.

Example connector creation approach:

  • uses “CLI printing press” (from Matt V, Lyft co-founder) to generate connectors
  • creator created connectors for FitnessPal and School as examples

Also highlights building “applications” such as:

  • the dashboard itself
  • micro apps (e.g., masonry grid for media generations)
  • a “landing pad” artifacts system driven by Claude-built outputs

Reviews / Opinions / Analysis Included

Key opinionated takeaways:

  • Dashboard visuals are not the main value—workspace/context organization is.
  • Bottom-up learning order: skills → memory → routines → applications

Performance/cost motivation:

  • better structure reduces retrieval overhead
  • improves speed
  • headless/automated runs reduce friction and increase productivity

Integration pain point:

  • always-on agents (Hermes) require synced context to access skills/memory

Main Speaker / Sources

Main Speaker

  • Jay (runs an AI business and one of the largest AI communities; introduced as “Jay” with a background in AI/data science)

Referenced Third-Party Sources/Tools (Not Speakers)

  • Entropic/Claude (pre-built skills, Cloud Desktop/Cloud Code context)
  • Matt VH (mentioned as creator of “CLI printing press”)

Tools/Services Mentioned

  • Hermes, OpenClaw, Grockbot
  • Syncthing
  • VPS
  • GitHub connectors

Original video