← Back to "AI for Good" Nonprofit Workshop Resources
<aside>
🧭
Chiv Heng Consulting — www.chivheng.consulting • LinkedIn
</aside>
<aside>
⚠️
Good to know: Cowork is included with every paid Claude plan (Pro and up) and runs on web, desktop, and mobile. If you do not have a paid plan, the free NotebookLM is a fine first step. For most nonprofits, the real question is budget, not readiness.
</aside>
Last reviewed: July 2026
Think of Cowork as the difference between an assistant who answers questions and one who completes the task for you.
The three ways to use Claude
- Claude chat — an assistant who answers questions. Best for quick questions, brainstorming, and drafting.
- Claude Cowork — an assistant who does multi-step work. Best for file-based work, research, and recurring tasks.
- Claude Code — a developer's power tool for building software (needs coding comfort).
Cowork sits in the middle: agentic help for people who do not code.
What Cowork actually does
You describe an outcome, and Claude makes a plan, does the work, and checks in as it goes. What makes it different from a normal chat:
- It works with your real files. It can read and create Word, Excel, and PowerPoint files on your computer, not just text in a chat window.
- It connects to your tools. Gmail, Google Drive, Calendar, Slack, and a growing list, so it can pull from apps you already use.
- It can browse. For research or current information, it searches the web and folds findings into your draft.
- It runs on a schedule. Set up recurring work (a Monday briefing, a deadline tracker) that runs without you asking each time. These now run in the cloud, so your computer does not need to stay open.
- It learns your process. Save reusable "skills" that teach it your format or voice, then run them on demand.
Why it matters for nonprofits
Cowork is the clearest example of the shift from chatbots to agents for people who do not write code. It moves AI from "help me draft this" to "do this task," which is exactly the jump the Delegation Framework describes. Used well, it takes repetitive, multi-step work off your plate.
Three starter use cases (keep them low-stakes)