How I AII gave Claude Cowork my entire job. It handled most of it. | JJ Englert (Tenex)
CHAPTERS
- 0:00 – 2:54
Why Cowork turns you into an “AI orchestrator” (multi-agent work as a solo operator)
JJ and Claire open with the core promise: Cowork lets non-developers coordinate multiple agents in parallel, assigning permissions and attention where needed. They highlight how sub-agents create fresh perspectives via distinct personas—useful for review, critique, and confidence when working alone.
- •Cowork as an orchestrator layer: manage multiple concurrent agents
- •Sub-agents provide independent personas and ‘fresh context windows’
- •Using persona-based feedback to strengthen work before sharing
- •Solo-founder/remote-work reality: AI ‘roundtable’ replaces costly synchronous feedback
- 2:54 – 5:11
From skepticism to daily driver: what Cowork is for (and why it’s different from chat)
Claire explains her initial skepticism—Cowork felt like a UI on top of Claude Code—then notes the product’s rapid improvement and adoption by non-technical users. JJ frames Cowork as the bridge from “ask AI” to “AI does work,” especially once connected to business tools.
- •Claire’s original ‘who is this for?’ confusion and why it changed
- •Cowork’s appeal to non-technical knowledge workers
- •Difference between Chat mode (advice) vs Cowork (actions)
- •Cowork for business productivity; Claude Code remains builder-focused
- 5:11 – 6:51
Where Cowork lives: Claude Desktop modes and the first click into Cowork
JJ situates Cowork inside the Claude Desktop app (Chat / Cowork / Claude) and explains why desktop matters for real work execution. They clarify Cowork’s ability to access the computer and browser to take real actions, beyond summarizing or drafting text.
- •Claude Desktop has three modes; Cowork is the ‘do work’ tab
- •Desktop-first: better access to files/actions than mobile
- •Cowork can operate on desktop and in browser (reservations, organizing, etc.)
- •The psychological shift: from chatting to delegating execution
- 6:51 – 10:44
Projects are just folders: the foundational mental model (plus the ‘brain’ file)
JJ introduces Cowork’s core abstraction: a project is simply a folder on your computer that Claude can work within. He shows his ‘brain’ markdown file—working preferences, collaborators, and style rules—so every task starts with strong context instead of repeating prompts.
- •Project = folder; files are the durable context layer
- •‘Brain.md’ as reusable instructions (preferences, people, tone)
- •Portable context replaces fragile ‘one magic chat thread’ workflows
- •Cowork as an on-ramp to Claude Code concepts without terminal use
- 10:44 – 12:31
Workspace map: helping Claude navigate your file structure efficiently
To reduce token waste and improve consistency, JJ uses a ‘workspace map’ that summarizes the folder layout and what’s inside. Claire reinforces this as a general technique: ask Claude to orient itself to any repo/folder quickly and produce a navigational index.
- •Workspace map summarizes folders/files for fast retrieval
- •Reduces unnecessary context ingestion and token costs
- •Improves reliability by narrowing Claude’s attention
- •Reusable trick: ‘What’s going on in here?’ for any shared folder/repo
- 12:31 – 15:02
Live build: creating a ‘Daily Operating System’ project from scratch
JJ creates a new folder, opens it in Cowork, and prompts Claude to help build a daily operating system for email, Slack, and decision support. Claude responds by asking clarifying questions, demonstrating how projects get shaped interactively rather than via one perfect prompt.
- •Step-by-step: create folder → open in Cowork → start task
- •Prompt outlines goals (email help, Slack help, thinking partner)
- •Claude asks setup questions (storage, interaction mode, support style)
- •Key lesson: you can start simple; complexity can grow over time
- 15:02 – 18:47
Project interface vs tasks: shared memory, chaining work, and focus
JJ explains why the Project view matters: multiple tasks inside a project share memory and context, enabling continuity across sessions. Claire clarifies the two-layer organization—folder context plus project memory—along with project-specific instructions that guide all future work.
- •Projects enable ‘shared memory’ across multiple tasks
- •Move an existing task into a project to inherit continuity
- •Project-level instructions tune behavior per workstream
- •Better results: less context sent, faster prompts, more consistency
- 18:47 – 19:47
Connectors setup: linking Gmail, Slack, Calendar, Drive, Notion (with permissions)
JJ demonstrates connectors as the primary unlock: one-click integrations that let Cowork read and act within your tools. He emphasizes permission granularity—allow/deny/ask—and shows how connectors support both action (drafting) and ingestion (learning from your data).
- •Connectors provide access to core work apps and actions
- •Browse connectors; authenticate; configure permissions per action
- •Granular controls: never/ask/always allow
- •Connectors power both execution and personalization via data ingestion
- 19:47 – 24:27
Personalized email writing skill: analyzing sent mail to learn your voice
Using the Gmail connector, JJ has Cowork analyze his sent emails from the last 30 days to build an email style guide and ‘voice profile.’ Claire highlights this as an ‘anti-to-do list’ approach: identify tasks you never want to do again and turn them into reusable skills.
- •Analyze last 30 days of sent mail to model tone and structure
- •Create a reusable ‘email writing skill’ based on real examples
- •Anti-to-do list mindset: automate recurring, low-leverage work
- •Safety guardrail: default to drafts, not sending emails automatically
- 24:27 – 26:19
Being an orchestrator: approvals, technical prompts, and gradual comfort
They show Cowork prompting for permission to run an action, and discuss the moment when non-technical users see terminal-like commands. Claire advises copying commands into a separate chat to understand them, and frames Cowork as a gentle ramp into more technical agent workflows.
- •Orchestrator dashboard shows which agent needs attention
- •Permission prompts can look technical; you can ask Claude to explain
- •Cowork blends friendly UI with Claude Code-style execution under the hood
- •Best practice: read approvals carefully; build confidence over time
- 26:19 – 27:28
Cowork vs OpenClaw: choosing trust boundaries for business vs personal autonomy
JJ compares Cowork with OpenClaw: he trusts Cowork more for business tool integrations, while using OpenClaw as a personal autonomous assistant with limited account access. Claire notes her heavier OpenClaw usage but validates Cowork’s widespread ‘does my job’ adoption.
- •JJ’s split: Cowork for business systems; OpenClaw for personal assistant
- •Trust and security considerations drive tool choice
- •Both can coexist as complementary layers
- •Key theme: match autonomy level to risk tolerance
- 27:28 – 30:06
Thinking-partner skill: mentorship, decision frameworks, and better communication
JJ spins up a ‘thinking partner’ skill to support decisions, feedback, career planning, and response crafting. He stresses adding “ask clarifying questions” to force better context gathering and more tailored guidance from the skill.
- •Skill acts as mentor/coach for decisions and communication
- •Covers: responses, critique, career direction, big-picture thinking
- •Include ‘ask questions if unclear’ to improve output quality
- •Skills = reusable, detailed prompts invoked on demand
- 30:06 – 34:03
Sub-advisory / multi-persona review skill: pre-feedback for PRDs, marketing, and writing
JJ builds a review skill that launches sub-agents with distinct personas (e.g., ICP segments) to critique work from multiple viewpoints. Claire connects this to remote work: AI feedback reduces reliance on expensive meetings and helps solo creators pressure-test ideas.
- •Sub-agents simulate stakeholders/ICPs for objective critique
- •Useful across newsletters, podcasts, social posts, PRDs
- •Remote-first benefit: asynchronous ‘roundtable’ feedback
- •Advanced idea: simulate your boss by researching their perspective
- 34:03 – 41:01
Advanced automation patterns: multi-step newsletter pipeline + scheduled morning debriefs
JJ describes a multi-step newsletter skill (interview, research, section drafting, evaluation, advisory review, and performance feedback loops). He then builds a scheduled ‘morning debrief’ task that scans email/Slack/calendar daily to produce an actionable plan for the day.
- •Newsletter skill as a pipeline: interview → research → drafting → evaluation
- •Define success with metrics; feed good/bad examples to refine outputs
- •Scheduled tasks run on cadence inside a project with full context
- •Morning debrief: triage messages, prep meetings, create daily action plan
- 41:01 – 50:11
Progressive trust and broader project ideas (beyond work) + lightning round
They close by framing AI adoption as a ‘progressive trust’ spectrum: start with drafting, then reading, then autonomy. Claire expands project ideas beyond work (home maintenance, recruiting), and JJ shares favorite use cases and tools (Remotion, Pencil.dev) plus prompting habits.
- •Progressive trust: trade information for productivity at your comfort level
- •Projects can be personal (home maintenance) or operational (hiring pipelines)
- •JJ’s favorite Cowork use: navigating dense docs/files for complex projects
- •Lightning round: Remotion for programmatic video; Pencil.dev for design; prompting via clearer success criteria
