Don’t Overthink It: Building Your First AI Agent in Claude Cowork
This month Andy Bass of InnovateAI led a hands-on lunch and learn on building AI agents in the SURGE Training Room at The Next Center in Holland. It drew such interest that he gave it twice, on September 17 during Tech Week Grand Rapids and again on September 24. His title said it all: “Don’t Overthink It, Just Build: Your First AI Agent.” He was right. The hardest part is getting started.
Generative AI versus agents
Most of us began with generative AI. You type a question into ChatGPT or Claude, and it writes an answer. It is a very capable conversation partner, but you still do the work between the answers: copying, pasting, saving files, and deciding what comes next.
An agent takes the next step. You describe the outcome you want, and the agent plans the steps, uses the tools you have connected, checks its work, and delivers the result. Generative AI answers. An agent acts.
What that looks like in Claude Cowork
Claude Cowork is Anthropic’s agent for everyday knowledge work. It runs in the Claude desktop app, can read and write the files you give it access to, and works with connected tools such as Google Drive, Gmail, and your calendar.
Here is a real example. I asked Cowork to adapt an article I had written for our Rotary district newsletter into a piece for The Whittaker Report and save it in my newsletter folder. It read my Word document, found the right folder on our shared drive, rewrote the piece for economic developers, kept my photos, and saved a Word file where I keep my drafts. I reviewed it, asked for a few changes, and it was done. I never copied, pasted, or saved a thing.
Global instructions: Your standing brief
The most valuable thing I have done in Cowork is write my global instructions. Think of them as the onboarding memo you would hand a new staff member. Claude reads them at the start of every session, so you never have to repeat yourself.
Setting them up takes a minute. Open Settings in the Claude desktop app, select Cowork, click Edit next to Global Instructions, paste in your text, and save.
Mine runs about two pages and covers:
- Who I am and what Whittaker Associates does
- Where the company is headed, so advice lines up with our strategy
- Tone and voice: warm and conversational, and keeps my voice when editing
- Formatting rules: Word documents, clean layout, and no em dashes
- Research standards: verify facts against primary sources, cite them, and flag anything unconfirmed
- Where my files live and how I like to work: one or two clarifying questions, then the outcome first
That research rule matters for economic developers. Our credibility depends on accuracy, so I tell Claude plainly to separate what is confirmed from what is inferred.
Start small
Andy’s advice matches my experience. Pick one repeatable task, write clear instructions, test it, fix one thing, and test again. Your first agent does not need to be perfect. It needs to be useful.
Andy has posted his workbook, starter prompts, and test cases free, with no sign-in required: Build Your First AI Agent, InnovateAI.