Lesson 2: Treat AI Like a Collaborator, Not a Vending Machine
Lesson overview
Most people use AI in one direction: request, receive, done. This lesson explains why that approach limits the quality of what you get — and what a more conversational, iterative approach looks like in practice.
What this means
A vending machine gives you what is in slot B4. You do not get to tell it that B4 looks a little stale, or that you wanted something different after all, or that the portion was too large.
Most people use AI the same way. They put in a request, review the output, and either accept it or start over with a new request. That pattern misses most of the actual value.
AI is more like a capable colleague with a very broad knowledge base and an unusually high tolerance for revision. You can tell it what is not working. You can ask it to try a different approach. You can give it more information after seeing what it produces. You can redirect it mid-conversation.
That back-and-forth is where the real quality improvement happens.
Why it matters
One-shot prompting is good for simple tasks where the request is clear, the output is short, and the stakes are low. For anything more complex — a document, a plan, an explanation, a communication piece — one-shot prompting produces work that requires significant editing because it had to make too many assumptions without your input.
Iteration is not a sign that the first request failed. It is part of the process. The first output tells you something: what AI assumed, what it prioritized, what it got wrong. That information helps you give better direction in the next round.
What most people do wrong
Starting over every time instead of redirecting
When output misses the mark, many people delete it and write a new prompt from scratch. This throws away the information you just received. A weak first draft is more useful than a blank page — it shows you what needs to change.
Framing every request as a final product
If you go in expecting perfect output on the first try, you will consistently be disappointed. If you go in expecting a useful starting point that will need shaping, the same output looks much more workable.
Not using follow-up at all
A surprising number of people do not know they can say "that is too formal, try again." You can give feedback in natural language. You do not need a new prompt format. Just tell it what is wrong.
Rewriting the output manually without teaching AI what changed
If you fix the draft yourself without telling AI what you changed and why, the next request will probably make the same mistakes. Explaining the fix gives AI the information it needs to do better next time.
What better looks like
Strong AI use is a conversation, not a transaction.
Here is what an iterative workflow looks like:
Round 1: Submit the initial request with as much context as you have.
Round 2: Review the output and identify what is off — tone, structure, length, focus, accuracy. Give specific feedback. "The tone is too formal for this audience. Rewrite it to sound more direct and plain-spoken."
Round 3: Review again. Now the output is closer. Ask for any final adjustments: "This is good. Cut the third paragraph — it repeats the intro."
Round 4 (optional): Save the successful exchange as a reference. Note what you told AI that made the difference. This becomes the basis for a better first prompt next time.
Weak example
Request: Write a product announcement email.
Output: Generic announcement email is received. The tone is slightly formal, the subject line is predictable, and it is longer than needed.
What most people do: Accept it, manually edit it, or delete it and type the same request again.
Strong example
Request: Write a product announcement email for a new integration we just shipped. The audience is existing customers who are technical users — they know our product well. Keep it short (under 200 words). Tone should be direct and conversational. Subject line should be specific, not promotional.
Output: Shorter, more focused email. Good direction but still a bit stiff.
Follow-up: "Good start. The subject line is still too generic — try something that names the specific integration. Also, the last paragraph is unnecessary. Cut it."
Output 2: Now the email is on target.
This is a two-round conversation that took less time than manually rewriting the first draft.
Practical exercise
Start with this weak first draft and improve it through exactly two rounds of follow-up prompts. Do not rewrite it yourself — only give AI instructions.
Weak draft:
Communicating well with your team is important for success. When team members understand what is expected, they are more likely to perform well. Here are some tips for better team communication: 1) Be clear. 2) Be consistent. 3) Listen actively. 4) Give feedback regularly. If you follow these steps, your team communication will improve.
Round 1 prompt: Identify something that is weak about this draft and give AI one instruction to improve it.
Round 2 prompt: Identify one more thing that is still weak and give one more instruction.
Compare the final output to the original.
Reflection prompt
- Do you typically treat the first AI output as a starting point or a finished product?
- Think of a time when you got mediocre output and edited it yourself. What would you have said to AI to get that result directly?
- Is there a recurring task where you could benefit from a two-round approach instead of one-shot prompting?
Key takeaway
AI is most useful when you treat it as a thinking partner and work through rounds of direction and refinement. The first draft is information, not a final answer. Iteration is how strong output gets built.