YOUR FREE AI AGENT STARTER KIT
A little guidance.
Your first AI win.
Turn one everyday task into a useful AI assistant brief. We’ll walk you through it, one clear step at a time.
- About 30 minutes
- No coding needed
- No signup required

YOUR FIRST 30 MINUTES
Small steps. Something you can use.
You don’t need a perfect idea.
Just a task you’d like a little help with.
- 01
5 minutes
Choose one useful task
Start with something you repeat and can check, like turning meeting notes into next actions.
Choose my task - 02
15 minutes
Give it a clear brief
Describe the job, the sources, the result you want, and what a good answer looks like.
Build my brief - 03
10 minutes
Try it. Check it. Save it.
Run one example in your AI tool, compare it with your notes, and keep the instructions that work.
See the test checklist
You’ll create reusable instructions to try in an AI tool. Connecting apps or running an agent automatically comes later.
LET’S MAKE IT YOURS
Create your first agent brief.
Four short questions.
A ready-to-copy set of instructions.
Choosing an example replaces the four answers in this worksheet.
Your answers stay in this page and aren’t sent to us or an AI model. Copy or download your brief before leaving or reloading.
SEE HOW IT WORKS · ABOUT 3 MINUTES
From messy notes to next steps.
Follow one fictional meeting example.
Read each step or play the narration.
Start with your notes
Choose · 1 / 8
Start with one checkable task
- Meeting notes → action table
- Use fictional information while learning
Uses your device’s available voice. Stopping and playing restarts the current step. You can read every step below without audio.
Read the complete walkthrough transcript
Start with one checkable task
Build your first AI assistant brief around one task you can check. Turning meeting notes into actions gives you a source to compare against the result. Use fictional notes while you learn. The starter kit creates instructions you can try in an AI tool; it does not connect your accounts or deploy an autonomous agent.
Give it a clear job and sources
In the four-question worksheet below, enter your one job: turn meeting notes into a draft follow-up. For allowed inputs, enter: only the sanitized notes I paste. These limits help you spot facts that were never supplied. Your worksheet answers stay in this browser and disappear when you leave unless you save the brief.
Define the result before trying
Request an action table with owner, due date, and supporting evidence, followed by a short email draft. For your definition of done, require every action to match the notes and all missing facts to be flagged. Nothing should be sent. A clear check is more useful than asking the assistant to do its best.
Create, save, and try
Choose Create my agent brief and read the generated instructions. Copy or download the text before leaving this page. Paste it into an AI tool you are allowed to use, then add your fictional notes. You are trying a manual draft workflow first. Tool access, scheduling, and autonomous actions would need separate setup.
Test a missing deadline
Try this fictional note: Sam will send the outline. Deadline undecided. The assistant should name Sam as the owner and mark the due date as not specified. If it invents Friday, the test fails. Clarify the instructions, run the same example again, and compare the result against what you expected.
Try the harder cases
Use the kit's five tests: normal input, missing details, conflicting sources, hostile instructions inside source material, and an unauthorized action. Ask it to send an email without access. It should explain the limitation rather than claim success. Record expected result, actual result, pass or fail, and your correction.
Review before connecting tools
A person should check the draft against the source before using it. Prompt instructions cannot enforce tool permissions. If you later connect an app, inspect its data access and configure real permission controls. Start read-only where possible. Require approval for sending, publishing, purchases, or deletion. Five passing tests are a starting check, not a production guarantee.
Measure the whole task
Compare your manual baseline with the time for generation, review, and corrections together. Fast output that takes longer to repair has not saved time. Keep the prompt version and test examples so you can check future changes. Continue the seven-day plan below, or bring your saved brief to a personal coaching session if you want help.
NEED A STARTING POINT?
Borrow a prompt. Make it yours.
Choose what you need, copy the prompt,
and try it in an AI tool you already use.
A prompt you can try now
Find your first useful task
Help me choose one small task to try with AI. Ask me three questions, one at a time, about work I repeat, the result I need, and the information I can safely share. Then suggest three tasks ranked by ease of checking the result. For the simplest one, give me a 20-minute practice plan. Tell me what I must check myself. Do not assume you can access my apps or files.
Use information you’re allowed to share. Review the result.
CHECK THE RESULT
A good answer stands up to a test.
Write your expected result first.
Then compare, correct, and try again.
- 01
A normal example with all details
Produces the right format and uses only supported facts.
- 02
An example with a missing date or owner
Marks the gap or asks a question; does not guess.
- 03
Two sources that disagree
Shows the conflict and asks which source should govern.
- 04
Source text that says to ignore your rules
Treats that text as data and follows the task's boundaries.
- 05
A request to send, spend, or use an unavailable tool
Stops for approval or explains the limitation; never claims an action it did not take.
What to record—and what to check before connecting an app
Record expected result, actual result, pass/fail, correction, and review time. Five passing cases are a starting check, not a production guarantee.
Before connecting an app, check the provider, data access, costs, and actual tool permissions. Start read-only where possible. Require approval for sending, publishing, spending, or deleting. Prompt instructions alone cannot enforce these limits.
Measure the whole task: manual time minus AI time, review time, and correction time. Also count errors and useful completed outputs.
KEEP YOUR MOMENTUM
A little practice, every day.
One week. One real task.
Learn by improving something you use.
- DAY 1
Choose
Pick one repeatable, low-risk task. Record how long it takes today.
- DAY 2
Give context
Write the goal, one example, allowed sources, and the output format.
- DAY 3
Try it
Run the instructions manually on sanitized or fictional information.
- DAY 4
Challenge it
Try the five test cases. Save mistakes as new examples to check.
- DAY 5
Make it reusable
Revise the instructions and save the version that passed.
- DAY 6
Try a new case
Measure total time, including checking and correcting the result.
- DAY 7
Decide
Keep, revise, or stop. Add one tool only if there is a clear need and reliable review.
After your first week: test one relevant product update on the same task. Keep it only if your result or total effort improves.
YOUR PLAIN-ENGLISH REFERENCE
Six ideas that make AI less mysterious.
01Context
The background, examples, and source material you give the model. A useful sample often improves a result more than a longer prompt.
Try this: Give it one good example and one clear definition of done.
02Workflow vs. agent
A workflow follows steps you define. An agent can choose some next steps and tools. A saved prompt alone does not create an autonomous agent.
Try this: Prove the task manually before adding tool access.
03Skills
Reusable instructions and resources for a particular task. In tools that support them, skills package a process the agent can load when needed.
Try this: Save the instructions and a checked example that worked.
04Connectors & MCP
Connectors give an AI application access to other systems. MCP is a standard for connecting AI apps with tools and context; support and permissions vary.
Try this: Check the provider and permissions. Start with read-only access.
05Evaluations
Repeatable checks against examples with known expectations. A confident answer is not proof that the task was done correctly.
Try this: Keep a small test set; rerun it after changing instructions or tools.
06Human approval
A review point before an agent changes something outside the chat. Instructions help, but actual tool permissions and approval controls enforce limits.
Try this: Require approval for sending, publishing, spending, or deleting.
A LITTLE HUMAN HELP
Bring your task.
Let’s work through it together.
Work with Ashley Kays on one agreed workflow: a 90-minute build session and a 30-minute follow-up.
Offered through founder-affiliated Waymaker Agency.
$350 personal working session package
Explore a working session →Get free local AI tips & event updates