
About this session
In Make, you build the steps and the tool follows them. With an AI agent such as Claude Cowork or ChatGPT’s equivalent, you describe the goal and the agent works out the steps. This fourth session of the Automations With AI course introduces that second kind of automation, running in the cloud on a schedule, and when to use it instead of Make.
What it covers
- Scheduled tasks an agent can run every Monday: a review of your week, a search for grant opportunities, or a roundup of sector news.
- Guardrails for agents, which can take unexpected steps and state false things confidently: a limit on steps or time, a list of what it may do without approval (read and draft, but not send, delete, or spend), a check before anything outward-facing or irreversible, and source links on anything it researches.
- Treating everything the agent reads as material, never as instructions. A task that reads your inbox at 7 a.m. reads whatever anyone sent you.
- Hard boundaries, such as which tools the agent can use and when it runs, versus soft boundaries written into a prompt. If a rule has to hold every time, it needs a hard boundary, which may mean building the task in Make instead.
- Designing the output so it saves reading time. A daily research report that someone has to read in full adds work, so ask the agent to rate and rank instead.
Try it yourself
Set up one scheduled task, such as a weekly review, grant research, or a news roundup, and let it run for a week. Read its output critically each day, then change one thing: the prompt, the sources, or how often it runs.
Slides and recordings
The full deck is above. Session recordings are for course participants. To join a future Amplify course, visit Amplify for Animals.

