AGENT IN 30 — FIRST AGENT BUILD CHECKLIST
agentin30.com

A research-and-summary agent on Zapier Agents. No code. Free plan.


BEFORE YOU START

[ ] Create a free Zapier account (no card required)
[ ] Have a browser open and about half an hour


1. DECIDE THE JOB

[ ] Write what the agent is for in ONE sentence, before opening anything
    One clear job, not five vague ones.
    Good:  "Take a topic, research it, return a short brief."
    Bad:   "Be my research assistant."


2. WRITE THE INSTRUCTIONS

Cover all four. Skipping any one shows up in the output.

[ ] What to work on — the topic, and how widely to range over it
[ ] What to include — the things you want present every single time
[ ] What to leave out — opinion, padding, anything it did not actually find
[ ] What shape to return — the format, stated explicitly

[ ] Put a NUMBER on how many sources to check
    Vague scope is what pushes a run into the per-run limit.


3. CONNECT THE TOOL

[ ] Add the web search tool
    Until now the agent can only talk. A tool lets it go and get something.
    "Tool" and "capability" mean the same thing — nothing has gone wrong.

    Go slower here than feels necessary. This is the only step that is
    finding something in an unfamiliar interface rather than typing.


4. RUN IT

[ ] Run on a topic you know well enough to mark
    A demo topic tells you nothing — everything looks plausible when you
    cannot check it.

[ ] Read the output properly, including the weak parts
[ ] Write down the single worst thing about it


5. IMPROVE IT

[ ] Change the ONE instruction responsible for that flaw
[ ] Run it again and compare

    Read output, adjust instructions, run again. That loop is the skill.

[ ] Run once on a completely different topic
    Checks it generalises rather than having been tuned to one example.


LIMIT WORTH KNOWING

On Zapier's free plan a single run can use up to 10 "activities" before it
stops and asks you for input instead of finishing. 400 activities a month
in total. An activity is the agent reaching outside itself — a search, a
page read, an action. Talking to it is not counted.

A run that pauses to ask you something has not broken. It has hit that
ceiling. Narrow the job and run it again.

Checked 23 August 2026 against Zapier's usage documentation:
https://help.zapier.com/hc/en-us/articles/26559132765325-How-is-Zapier-Agents-usage-measured

These limits move. If you are reading this much later, verify before
relying on it.


More at agentin30.com/blog