AXHUB LECTURE · INTRO L2
How to put AI to work
— like briefing a new hire
You don't need to know "prompt engineering." If you've ever briefed a new hire on a task, you already have the instinct you need. This lecture moves that instinct into five habits.
1Don't hand over the whole thing
You don't tell a new hire "just go write the report."
Same with AI. Throw the whole thing at it and you get a flat result.
Break the task into steps and the result changes. Real-user reviews reach the same conclusion — short step-by-step instructions beat one grand request.
(1) In this Excel, find only the items that changed 20% or more from the previous month
(2) Organize the found items into a table with a guessed cause
(3) Summarize in 3 lines for reporting to the manager
Pick one of your tasks, write it out as (1)(2)(3), and run it in order.
Check: did you verify each step's result by eye before moving to the next?
2Give the context first
You don't put a new hire to work without introducing the company.
Before the request, tell it three things — who we are, who will read this, and what tone it should have.
The same request produces different sentences once the reader is decided.
We're a neighborhood Pilates studio. The reader is a member who hasn't come in 3+ months.
In a no-pressure tone, write a 4-sentence notice about a re-enrollment discount.
Run both versions and compare the results — you'll feel the difference in your bones.
Check: does the instruction include both "to whom" and "what tone"?
3Tell it what a good result is
A new hire doesn't know the standard in your head. Neither does the AI.
Showing one good example is faster than ten lines of explanation.
[paste the good example]
Now handle the item below by this standard.
[paste the new material]
Check: did you show "an example of a good result" at least once?
4Always review the result
You don't send a new hire's first report straight to your boss.
A person sees the AI's output at the end too. Numbers, names, dates, and links especially — always check them against the original.
This isn't inefficiency — a structure where a person reviews only the exceptions outperformed approving every one by more than double (analysis of 51 cases).
I'll check them against the original.
Having the AI pull the checklist speeds up your review.
Check: among what you sent out today, is there a number that went out unchecked?
5Keep what you taught
Training a new hire leaves behind training material.
Stack the instructions that worked, the requests that failed, and your tone standard in one place.
This accumulation stays even if you switch tools. 77% of the hardest part that decided success or failure was operations like this, not the model.
Format — task name / full instruction text / cautions
Check: did the instruction you made today go into the drawer?
When it doesn't work
Next lecture → L3. A shop owner's first AI — start with review replies · Full contents
Sources: Stanford Enterprise AI Playbook (review-exceptions-only +71% vs approve-each +30% · 77% of the hardest part is beyond technology) · real-user reviews (step-by-step instruction is more efficient) · "AI is close to a new hire" (solo-operator review). Full sources are in the axhub.net case library. The practice prompts are example lines to follow along.