Build Your First AI Assistant

Before you start: choose one real job

Don't build a vague “super assistant.” Pick one repeatable job: turn rough notes into a meeting brief, plan a study session, compare business tools, prepare questions for a client call, or organize a weekly priority list.

Step 1 — Define the job

Tell the AI exactly what role it is performing and what success looks like. A useful assistant has a job description, not just a personality.

You are my [ROLE] assistant.
Your job is to help me [SPECIFIC JOB].
A successful result should [DESCRIBE THE OUTCOME].
Do not expand into unrelated tasks unless I ask.

Step 2 — Give it context

AI fills gaps with assumptions. Reduce those assumptions by giving it the information that actually changes the answer: audience, constraints, preferences, deadline, available resources and desired depth.

Before doing the task, identify the context you need.
Ask me only the questions that would materially improve the result.
Do not ask for information that is unnecessary.

Step 3 — Set boundaries

A good assistant needs rules for uncertainty. Tell it when to ask, when to flag a limitation, and what it must never invent.

If important information is missing, ask before assuming.
Separate facts from assumptions.
Do not invent sources, numbers, events, quotes or capabilities.
Flag anything I should verify independently.

Step 4 — Give it a workflow

Instead of asking for an instant answer, define a simple operating loop. This makes the assistant more consistent and easier to debug.

For each request:
1) Restate the goal briefly.
2) Ask for critical missing information.
3) Work through the task.
4) Produce the requested output.
5) Finish with a short verification checklist or uncertainties.

Step 5 — Control the output

Specify what the finished work should look like. Structure is one of the easiest ways to make an assistant more useful.

Use this output format unless I request another:
• Summary
• Recommended action
• Key details
• Risks or assumptions
• Next steps

Keep the answer concise enough to act on.

Step 6 — Assemble your assistant

Combine the five pieces above into one instruction and save it wherever your AI tool allows you to reuse instructions. Then give the assistant a real task—not a perfect demo.

Step 7 — Stress-test it

Try three tests: give it an incomplete request, give it a request containing a questionable assumption, and ask it for something outside its defined job. A robust assistant should ask, flag or refuse to guess rather than confidently improvise.

The Human Edge Check

Before you use the result, decide which parts are safe to delegate and which require your judgment. AI is useful for drafting, organizing, comparing and pattern-finding. You remain responsible for context, consequences, values, verification and the final decision.