How I AIClaude Cowork tutorial for non-engineers | JJ Englert (Tenex)
Claire Vo and JJ Englert on a non-engineer’s guide to Claude Cowork projects, skills, and automation.
In this episode of How I AI, featuring JJ Englert and Claire Vo, Claude Cowork tutorial for non-engineers | JJ Englert (Tenex) explores a non-engineer’s guide to Claude Cowork projects, skills, and automation Cowork is positioned as a practical “get work done” layer between simple chat and full developer tools, letting non-engineers orchestrate AI agents with a familiar UI.
At a glance
WHAT IT’S REALLY ABOUT
A non-engineer’s guide to Claude Cowork projects, skills, and automation
- Cowork is positioned as a practical “get work done” layer between simple chat and full developer tools, letting non-engineers orchestrate AI agents with a familiar UI.
- Projects are explained as ordinary computer folders plus shared memory, enabling consistent context, faster prompts, and better outputs across multiple related tasks.
- JJ shows how connectors (Gmail, Slack, Calendar, Drive, Notion, etc.) let Cowork both take actions and ingest personal/work data to personalize skills like email drafting in your own voice.
- The episode walks through building a “daily operating system” from scratch: creating a project, generating reusable skills (email style, thinking partner, multi-persona review), and chaining workflows.
- They emphasize progressive trust and permissioning—trading information for productivity—using granular connector permissions and “draft-only” safeguards as adoption ramps up.
IDEAS WORTH REMEMBERING
5 ideasTreat a Cowork project as a folder-backed workspace with memory.
Create a dedicated folder for any initiative (work or personal), attach it as a Cowork Project, and Cowork will retain shared context across chats/tasks so you don’t restart from scratch each time.
Start with a “brain” file to stabilize quality and tone.
A simple markdown file describing your preferences, collaborators, and working style gives Cowork durable context and reduces repetitive prompting; it’s transferable across tasks and projects.
Use a workspace map to cut token usage and improve navigation.
Have Claude summarize the folder structure into a map so it can jump directly to the right subfolders/skills instead of re-reading everything, improving speed and consistency.
Connectors unlock both automation and personalization.
Connecting Gmail/Slack/Calendar/Drive/Notion enables actions (drafting, organizing, prep) and also allows Cowork to learn patterns (e.g., your writing style) from your existing data.
Create an email-writing skill from your sent mail—then enforce “draft-only.”
JJ demonstrates analyzing the last 30 days of sent emails to produce a voice/style guide skill, while adding project instructions like “Never send; only draft for review” to manage risk.
WORDS WORTH SAVING
5 quotesEven though you might not be a developer, you're now an AI orchestrator.
— JJ Englert
A project is a folder on your computer.
— JJ Englert
Without that workspace map, sometimes Claude has to ingest all of that stuff, which takes up more tokens.
— JJ Englert
All my friends are agents.
— Claire Vo
It’s a trade of information for productivity… think about this as progressive trust.
— Claire Vo
QUESTIONS ANSWERED IN THIS EPISODE
5 questionsWhat exactly should go into a “brain” markdown file to get better Cowork outputs, and what should you avoid including for privacy/security?
Cowork is positioned as a practical “get work done” layer between simple chat and full developer tools, letting non-engineers orchestrate AI agents with a familiar UI.
When should you use a Project versus a one-off Task in Cowork, and what kinds of work degrade when you mix too many tasks into one Project’s shared memory?
Projects are explained as ordinary computer folders plus shared memory, enabling consistent context, faster prompts, and better outputs across multiple related tasks.
Can you show a concrete template for a “sub-advisory” skill (boss/peer/customer) and how you’d keep personas realistic without overfitting or hallucination?
JJ shows how connectors (Gmail, Slack, Calendar, Drive, Notion, etc.) let Cowork both take actions and ingest personal/work data to personalize skills like email drafting in your own voice.
What connector permission settings do you recommend for beginners (e.g., Gmail read-only + draft creation, always-ask vs always-allow), and how do you phase access over time?
The episode walks through building a “daily operating system” from scratch: creating a project, generating reusable skills (email style, thinking partner, multi-persona review), and chaining workflows.
How do you evaluate whether a skill is actually improving outcomes (email response time, newsletter engagement, fewer revisions), and how do you feed “good vs bad” examples back into the system?
They emphasize progressive trust and permissioning—trading information for productivity—using granular connector permissions and “draft-only” safeguards as adoption ramps up.
Chapter Breakdown
Why Cowork finally clicked: buttons, business data, and “doing” vs chatting
Claire and JJ set the stage: Cowork initially felt like a UI on top of Claude Code, but rapid iteration made it compelling for non-engineers. JJ explains the key hook—easy one-click connections to work tools—so AI can take actions, not just give advice.
What Cowork is (and who it’s for): AI that can operate your computer and browser
JJ describes Cowork’s capability to access the desktop and browser to perform real actions, from organizing files to making reservations. The chapter clarifies that Cowork isn’t just for file management—it’s an execution layer for everyday work.
First steps in Claude Desktop: opening Cowork and starting a task
The walkthrough begins in the Claude Desktop app, showing how to enter Cowork and create a new task. JJ emphasizes that you can start simply—like a chat—then graduate into more structure as you get comfortable.
Projects are just folders: the core mental model for non-engineers
JJ and Claire demystify “projects” as nothing more than folders on your computer that Claude can work inside. This reframing helps users treat Cowork like a familiar GDrive/Notion organization system—now with an AI collaborator inside.
Creating a personal “brain” file + workspace map for fast, consistent results
JJ shows his reusable foundation: a “brain” markdown file containing work preferences, collaborators, and writing norms, plus a “workspace map” that helps Claude navigate the folder structure efficiently. This reduces token waste and improves output consistency.
Build a Daily Operating System from scratch: folder → task → structure
They create a new “Daily Operating System” folder and prompt Cowork to lay a foundation for email, Slack, and decision support workflows. Cowork asks clarifying questions to shape the system and define how JJ wants to interact (on-demand vs automated).
Project interface and shared memory: chaining tasks like an AI orchestrator
JJ transitions from a single task to the Project interface, showing how to move tasks into a project so they share memory and context. This creates an “orchestrator view” of multiple agents and improves focus and quality by keeping context scoped.
Connectors setup: Gmail, Slack, Calendar, Drive, Notion—plus permissions control
JJ highlights connectors as the critical unlock: Cowork can read and act within your tools once authorized. He also shows granular permission settings—allow/deny/ask—so users can adopt progressively without over-trusting the system.
Email superpower: analyze sent mail to learn your style and build a writing skill
JJ demonstrates using the Gmail connector to analyze the last 30 days of sent emails and generate an email-writing skill that matches his voice. They frame this as eliminating “anti-to-do list” work—like first drafts—while keeping human review in the loop.
Thinking-partner skill: mentoring, decision frameworks, tough conversations
Next, JJ creates a reusable “thinking partner” skill to support decisions, career coaching, and feedback conversations. He emphasizes adding “ask questions if unclear” to force better clarification and outcomes.
Sub-advisory skill: multi-persona feedback via sub-agents (boss/ICP/customer)
JJ introduces a powerful pattern: spawn sub-agents with distinct personas to review work from multiple perspectives, then aggregate feedback. Claire links this to remote-work realities—feedback is expensive—so AI roundtables improve work before human review.
Cowork vs OpenClaw + handling technical prompts and approvals safely
They discuss permission prompts (sometimes showing technical commands) and how non-technical users can sanity-check by asking Claude to explain. JJ contrasts Cowork as his trusted business driver vs OpenClaw as a more personal assistant, depending on access and risk tolerance.
Advanced skill architecture: multi-step newsletter creation with research and evaluation
JJ explains how he composes a complex newsletter skill: interview prompts, web research via sub-agents, section-specific sub-skills, subject line generation, and an evaluation checklist plus advisory-board review. He stresses defining success and feeding good/bad examples to improve quality over time.
Scheduled tasks: daily “morning debrief” from email, Slack, and calendar + broader use cases
JJ sets up an automated morning debrief to scan key tools and produce a plan for the day, including meeting prep. Claire broadens the idea: scheduled AI can push curated info (news, reminders, wellbeing), and projects can extend beyond work (home maintenance, hiring pipelines).
Lightning round: favorite use cases, favorite tools, and prompting habits
JJ shares top Cowork wins (navigating dense project docs), plus tools he likes outside Cowork (Remotion, Pencil.dev). They close with practical prompting guidance: stay polite, use caps for urgency, clarify success criteria, and record “bad outputs” to avoid repeats.
EVERY SPOKEN WORD
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome