Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
The Cowork Setup That Saves Me Hours Every Week
TL;DR
Claude Co-work is pitched as an operator, not a chatbot — the creator says the real shift is that it works inside the Claude desktop app with access to your actual files, apps, and multi-step tasks instead of just answering one prompt at a time.
The setup matters more than the demo — he recommends a paid Claude plan, the desktop app, all capabilities enabled, notifications on, training-data sharing off, and especially connectors like Gmail, Notion, Slack, Google Drive, Chrome, and Mac file system access.
Global instructions are a force multiplier — by loading persistent context like who he is, what he works on, preferred output style, and even rules like “don’t use m dashes in emails,” he makes every co-work session start with the right assumptions.
A simple Gmail-to-Slack workflow already saves real time — in one demo, Co-work scans emails related to This Week in AI and guest booking, identifies threads needing replies or follow-up, and sends a concise Slack DM, which he plans to make a scheduled task.
The bigger win is ongoing research automation — he built a Notion-based AI funding pipeline that has collected fundraising news for nearly 300 companies over about two weeks, using scheduled scraping, deduplication, and source scoring.
The standout detail is that Claude manages source quality, not just data entry — it tracks which fundraising sources are performing, assigns a “scrape priority,” and benches weak sources like tech.eu EU when they stop surfacing deals.
The Breakdown
Why Co-work Feels Like Hiring an Assistant
He opens by saying most people use Claude like a search engine, but Claude Co-work is a “huge unlock” because it behaves more like a fast assistant that can actually do work. The key distinction is that it runs in the Claude desktop app and can read, write, and edit files on your real computer rather than in some cloud sandbox.
The Three Differences That Actually Matter
He boils the gap between regular chat and Co-work down to three things: chat is one-message-at-a-time, has no file access, and each reply is mostly independent. Co-work, by contrast, can run multi-step tasks, use your real files and connected apps, and lets you message it while a task is still running.
The Settings He Always Changes First
Before the flashy use cases, he walks through his default setup: turn on notifications, switch off “help improve Claude” in privacy, and keep an eye on usage limits that reset roughly every 5 hours but also have a weekly cap. In capabilities, he says to enable everything so Claude can be more autonomous, generate artifacts, execute code, and create files.
Connectors Are the Whole Game
He calls connectors “definitely the most important part” because they make Claude act like another employee instead of a fancy text box. His stack includes Calendly, Canva, Gmail, Google Calendar, Google Drive, and Notion, and he also highlights system-level connectors like Claude and Chrome, Mac file system control, PDF viewer, and Anthropic’s PowerPoint and Word integrations.
Global Instructions: Quietly the Most Powerful Trick
He emphasizes global instructions as the context Claude reads in every Co-work session. His version includes who he is, what he works on, the tools he uses, how he likes outputs formatted, what to avoid, plus project context for Launch Accelerator, Found University, and This Week in AI.
Demo 1: Turning Gmail Triage Into a Slack Briefing
The first demo is refreshingly practical: he asks Co-work to search Gmail for This Week in AI and guest-booking threads, figure out what needs attention, and DM him the summary in Slack. A few minutes later, it produces a tidy breakdown of what needs a reply and what deserves follow-up soon, and he notes that the same task can be scheduled hourly or daily with different levels of caution.
Demo 2: A Funding Research Machine Feeding Notion
The second demo is the bigger flex: a research workflow for AI funding news that writes directly into a Notion CRM. Starting from blank Notion pages, he had Co-work build one database for startup fundraising data and another for sources, including fields like round size, investors, source, dates, and even a performance score for each source.
Source Scoring, Daily Scrapes, and 300 Companies Tracked
What really impressed him was that Co-work created a “scrape priority,” effectively benching sources that had gone cold instead of wasting time scanning everything every day. His scheduled task, called “daily AI funding scrape,” checks the top 20 sources, deduplicates deals, writes them into Notion, alerts the This Week in AI Slack channel, and over roughly two weeks has helped track fundraising news for almost 300 companies.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
