Start with three inputs: the list of sources to monitor, the topics that define relevance for your work, and the output format you want. Then configure a scheduled task — in Claude, Cowork, or a similar platform — to run those inputs on a daily or weekly cadence.
The Three Building Blocks
Building a research agent sounds technical, but the logic is straightforward. You’re essentially telling an AI system: “Check these places, look for these things, and tell me what you found in this format.” Those three elements — sources, topics, format — are the only decisions you need to make. Everything else is setup.
For sources, think about where the information you need actually lives. Competitor websites, specific YouTube channels, LinkedIn profiles of key voices in your niche, Reddit communities where your audience asks questions, or specific Google search terms that surface fresh content. You can start with three to five sources and expand as you learn what’s most useful.
Configuring the Agent Step by Step
Step one: write your relevance brief. This is a paragraph describing your work, your audience, and the specific topics you want monitored. Be precise: “I’m an educator teaching AI skills to coaches and consultants aged 45+. I want to monitor: new AI tool announcements relevant to non-technical educators, competitor course launches in my space, and trending questions about AI in online teaching communities.” Step two: identify your sources. List the specific URLs, YouTube channels, or search terms the agent should check. Step three: define the output. Should it be a bulleted list, a narrative summary, a prioritized top-five? Specify the format so the agent produces something you’ll actually read.
In a Claude Cowork environment, this becomes a scheduled task with those three inputs baked in. The agent runs on the schedule you set, applies your relevance brief as its filter, and writes the output to a file or community post. You review it, refine the brief if something important got missed, and the agent improves each iteration.
What This Means for Educators
The most effective research agents are built in an afternoon and refined over two weeks. Expect the first few reports to be 80% of what you want — good but not perfect. The refinement is quick: if something important got missed, add the source. If irrelevant content appears, add an exclusion. By week three, most educators have an agent that produces a report they genuinely look forward to reading each morning.
The Simple Rule
Build the simplest version first: three sources, one relevance brief, one output format. Run it for a week before adding complexity. Research agents that are over-engineered at the start become confusing to maintain and produce bloated reports. Start minimal, prove the value, then expand. The best research agent is the one you actually read every day.