digest
Short Summary
Main recommendation: Progress from using Claude as a simple one-off chat tool to deploying it as an autonomous system, moving through six levels from amateur to agent orchestrator. The speaker, Dan Martell, says most users stay at level one and never use Claude's memory, projects, integrations, skills, or agents, which he claims leaves most of its capability unused. Each level builds on the previous one by adding persistent context, connected tools, repeatable automations, custom software, and finally agents that run work without constant human input.
Key Points
- Level 1 (Amateur): Using Claude only for single questions with no memory or projects; the speaker says this uses about 5% of its capability.
- Level 2 (Regular): Organizing work into Claude projects per role, client, or workflow so Claude retains context across sessions.
- Level 3 (Integrator): Connecting Claude to tools such as Gmail, Calendar, Drive, Slack, and Notion so it can retrieve and act on information directly.
- Level 4 (Operator): Turning repeated work into system prompts, skills, and scheduled Co-work tasks, with the human reviewing and approving outputs.
- Level 5 (Builder): Using Claude Code to build loops, tools, and real software applications, with plan mode used before building.
- Level 6 (Agent Orchestrator): Designing a main agent that directs specialized sub-agents that run departments or workflows, with the human "on the loop" rather than "in the loop."
Details Worth Noting
Concrete instructions:
-
Level 1: Add to prompts, "Before you answer, ask me any questions that you need to perform this task properly," so Claude interviews you first. Ask Claude to "Check your work" to catch its own mistakes.
-
Level 2: Click "create a new project" on the left side and name it after your role (the example used is "marketing"). Prompt: "Interview me to build a master prompt for my role as a marketer." Save the resulting master prompt file, then add it as a project file along with sample data, examples, and processes.
-
Level 2 (advanced): Ask Claude to interview you to create a system prompt for a specific workflow, then paste it into your custom instructions so output quality and format stay consistent.
-
Level 3: Use connectors to link Gmail, Drive, Slack, Notion, and other systems. Use visualizations to build graphs and mockups in chat, and interactive artifacts for clickable mini apps. Paste text into the chat and ask Claude to place it in Composer, an in-chat document editor. Install Claude in Chrome to have it carry out browser tasks you would otherwise do manually.
-
Level 4: Have Claude interview you to create system prompts for recurring outputs. Save repeated workflows as skills (slash commands) and use them; the speaker's rule is to make a skill for anything done more than three times a week. Use Co-work to schedule recurring tasks, such as a nightly summary of tomorrow's emails and calendar. Chain skills so one skill calls another.
-
Level 5: Type "/plan" and describe your idea so Claude asks questions and writes a plan you approve before any code is written. Type "/remote-control" in the terminal to connect to the Claude app on your phone and keep coding remotely.
-
Level 6: Start with one main orchestrator agent, create specialized sub-agents that each own a workflow, and connect Telegram to talk to the agent by phone. Add a critique agent to review outputs such as copywriting or research and return notes for revision.
-
The speaker describes a Claude Code "hackathon" in which team members with no programming background built working software; he says English is the new programming language.
-
He names three categories of building in Claude Code: loops (recurring server-based jobs), tools (disposable one-off builds), and apps (production software).
-
He names Apex (apex.host) as his own platform that uses Claude as its main agent model with added security and easier interaction.
-
He states he has used Claude for over 1,000 hours and that reaching the Claude Code level puts someone in the 0.04% of the population.
-
Closing challenge: commit to using one Claude feature consistently for 30 days in a row.
Bottom Line
Claude's real value comes from building persistent context, connected tools, and automated workflows around it, and the highest-leverage step is to stop treating it as a search box and start designing repeatable systems.
