← Claude
Jack Roberts
Learn 95% of Claude in Under 12 Minutes
⏱ 11 min video · 3 min read22 Jun 2026Worth watching
TL;DR
Jack Roberts walks through 7 progressive levels of using Claude — from basic chat to autonomous multi-agent systems — showing how most people use only 5% of its capability. The video is a practical orientation to Claude Code, connectors, sub-agents, skills, and a custom operating system that can run Claude autonomously on your behalf.
Key points
1
Claude has three interfaces: Chat (no file access), Co-work (sandboxed file access), and Claude Code (full access) — the video focuses entirely on Claude Code as the most powerful option.
2
The first prompt in any conversation is the most critical: frame your intent clearly and ask Claude to interview you with questions rather than just answering a surface-level query.
3
Claude can be connected to external tools like Gmail, Google Drive, and Firecrawl via connectors, enabling it to triage emails, scrape web pages, and interact with your real-world data.
4
Multi-agent parallelism (spinning up sub-agents) can deliver 5-10x more output per unit of time compared to single-threaded question-and-answer use.
5
A Claude Code operating system can run autonomously, track spending across AI tools, surface unused skills, and generate recommendations based on your conversation history — even while you are not actively using it.
Actionable insights
→
Start every Claude session with a structured intent prompt: state your goal, ask Claude to interview you, and tell it to challenge you — this unlocks far more targeted output than a single question.
→
Use connectors (Gmail, Google Drive, Firecrawl) to give Claude hands in your real-world environment, and always apply the principle of least access — only grant the permissions Claude actually needs for a given task.
→
Use the sub-agent command to run parallel research tasks simultaneously rather than sequentially, and use Graphify to map codebases so Claude navigates them faster and uses fewer tokens.
→
Find and install skills from GitHub repos (search the desired skill + GitHub on Google, copy the repo URL, and tell Claude to turn it into a skill) to give Claude predefined recipes for recurring tasks.
→
Set Claude to require your approval before sending emails or taking irreversible actions — the approval/auto-approve permission split is available inside the connectors settings.
Notable quotes
“Most people use Claude like a Google search which is only 5% of what it can actually do.”
“Imagine having the world's best, most capable builder on your building site, but he doesn't have a single tool — wouldn't be very effective.”
“We can ask Claude to spin up sub-agents, which is like using five lanes on a motorway instead of one.”
Worth watching?
✅
Worth watching the full video?
Worth watching if you use Claude regularly but have never touched Claude Code — the live demos of connectors, sub-agents, and the operating system are clearer on screen than in text, though the key concepts are fully covered here.
Saved you some time? The creator still deserves a like.
Want this for your own channels?
Add the channels you follow. Every new video summarised and in your inbox the moment it drops. From £4/month.
Try it free