Claude Cowork tutorial for non-engineers | JJ Englert (Tenex)
TL;DR
Claude Co-work clicks for non-engineers once you think in folders, not prompts — JJ Englert frames a “project” as just a folder on your computer plus shared memory, which turns Claude from a chatbox into a persistent work partner.
The real unlock is connectors + reusable skills — by connecting Gmail, Slack, Google Calendar, Drive, and Notion, JJ has Claude analyze real work (like 46 sent emails from the last 30 days) to build an email-writing skill in his exact tone.
Projects turn you into an AI orchestrator, not just a user — inside Co-work’s project view, you can run multiple agents at once, see which one needs permission, and chain tasks together so each new task inherits the memory of the others.
Sub-agents are a practical feedback machine for solo workers and remote teams — JJ’s favorite pattern is spinning up 3 reviewer personas (like your boss, customer, and engineering partner) to critique a PRD, newsletter, or post before a real meeting.
Scheduled tasks make Claude feel like lightweight personal software — JJ creates a daily 7:30 a.m. “morning debrief” that checks email, Slack, and calendar to generate a plan of attack before he starts work.
Start with progressive trust, not full autonomy — both hosts emphasize you don’t need to let AI send emails or run your life on day one; drafting only, permission prompts, and scoped connectors are enough to get real productivity gains.
The Breakdown
From skeptic to convert: why Co-work suddenly makes sense
Claire Vo opens by admitting she originally thought Claude Co-work looked like “a UI slapped on top of Claude Code.” What changed was watching non-technical friends rave about it and seeing Anthropic rapidly improve the UI, especially for people who want AI to actually do work, not just answer questions.
JJ’s core model: a project is just a folder with memory
JJ’s first big simplification is reassuring: a project is literally a folder on your computer. He keeps markdown files like a personal “brain” inside those folders — preferences, collaborators, working style — so Claude stops starting from zero every time and instead shows up already knowing how he likes to work.
Workspace maps and why structure beats giant prompts
One of JJ’s favorite tricks is asking Claude to generate a “workspace map,” a file that explains the folder structure back to itself. That makes Claude faster and cheaper to use because it can jump to the right files instead of re-ingesting everything, which Claire points out is also a great trick anytime you’re dropped into a messy folder or repo and need to ask, “What’s going on in here?”
Building a daily operating system from scratch
To demonstrate, JJ creates a brand-new folder called “daily operating system” and tells Claude he wants help with email, Slack, and decision-making. Co-work responds by asking setup questions instead of forcing a perfect prompt, which is part of the point here: you can enter casually, then let the system become more structured as you go.
The big shift: tasks are chats, projects are shared-memory HQs
JJ then moves that initial chat into a formal project and explains why this matters. In task mode, you can do useful one-offs; in project mode, all those chats share context, instructions, and memory, creating what he calls an “orchestrator view” where you can manage multiple active agents working inside the same initiative.
Connectors turn Claude into a real work assistant
The demo gets tangible once JJ opens connectors: Gmail, Google Calendar, Drive, Notion, and Slack, all added with simple authentication. He asks Claude to analyze his sent emails from the last 30 days and build an email-writing skill from them, and the result is a style guide and voice profile based on 46 real emails — with a safety rule added afterward: never send, only draft.
Skills and sub-agents: reusable workflows with perspective
From there, JJ builds two higher-level skills: a generic “thinking partner” for career decisions and hard conversations, and a review skill that spins up multiple sub-agents with different personas. This is where the human angle really lands: Claire jokes, “All my friends are agents,” but the use case is dead serious — if you’re a solo founder, PM, or marketer, synthetic pre-feedback from your boss, customer, or ICP can dramatically improve your work before anyone joins a meeting.
Scheduled tasks: your morning debrief before you open your laptop
Finally, JJ creates a scheduled task called “Morning debrief” that runs daily at 7:30 a.m., reviews email, Slack, and calendar, and produces a plan for the day. The broader takeaway is that Co-work is less about chat and more about building tiny personal systems — for work, house maintenance, hiring, events, even weddings — as long as you’re willing to organize the context and decide how much trust you want to give the AI.