3 agents. no code.
15 minutes each
no api keys. no terminal.
What an agent is
Ask claude to research.
Read it. approve it.
Ask for an outline.
Review it. write it. edit it. format it.
Four steps. every time.
An agent runs all four in one pass. One input. One output. The whole difference.
Three types
no code. any of them.
Type 1. Chat agent.
Built in Claude Projects. A system prompt runs the full workflow on its own.
Type 2. File agent.
Built in Claude Cowork. Reads a folder, processes every file, saves the output.
Type 3. Scheduled agent.
Built with Cowork and /schedule. Runs on its own schedule. Nobody has to start it.
Agent 1. Research to article.
15 minutes. one prompt.
Open projects
Click Projects in the sidebar. Create a project. Name it Article Agent.
Paste the system prompt
Research, pick an angle, outline, write, review. Five steps in one response.
Type a topic
Claude hands back a finished, reviewed article from a 3-word input.
try: ai agents for beginners
Agent 2. File processing.
15 minutes. one folder.
Open cowork
Grant access to the folder you want processed.
Paste the instruction
Read every file. Summarize it. Rate it. Save the summary.
Let it run
Claude processes the whole folder and builds one ranked master file.
works on contracts. notes. receipts too.
Agent 3. Scheduled morning.
runs while you sleep.
Open cowork
/schedule
Paste the instruction
Check gmail. Sort it. Draft replies. Check the calendar. Save the brief.
Set the time
Every weekday. 7am. Save it.
3 minutes to scan instead of 30 to dig.
Make them better
three rules. one month.
01Correction rule. every fix becomes a new instruction.
02Example rule. upload your best output as reference.
03Feedback loop. review weekly. update the prompt.
3 to 5 times better output after a month.