The Boss’s Steering Mindset: Say to AI What You Would Never Say to an Employee

Many bosses think using AI means first learning a pile of new tools, and it is painful. But it is the Agent era now, and AI can already work as an employee. Your skill at leading people is your biggest advantage: once you think of AI as an "employee," you already know it all. This covers the "steering mindset," plus a full set of copy-paste boss prompts so you can start today.

Who it is for

· People who have been bosses or managers for decades and find learning new AI tools painful
· People who are great at leading people but do not know how to apply that skill to AI
· People who want AI to produce something truly valuable, not just "looks okay"

What you will get

· A mindset you can use immediately: treat AI as an employee, and be the owner toward it
· Three copy-paste prompt sets: the ten steering questions, inward questioning, the good-boss comparison
· A reminder: AI makes you fast, but "making the right call" is worth even more

Series · From Prompt to Steering Engineering

Stop Writing Rigid Instructions: Say What You Want, Let AI Handle the Rest
The Boss’s Steering Mindset: Say to AI What You Would Never Say to an Employee (you are reading this)

🗺️ See the full map: From Prompt to Loop Engineering

The steering mindset: what you dare not ask an employee, ask AI freely

Having led employees this long, you know one thing well: "If I ask like that, they will quit on the spot." So you pull the request back, lower the bar, give more time. That is a good boss.

But AI is different. It will not quit, will not complain, will not badmouth you behind your back, will not go to the labor board. The high standards you dare not put on an employee are exactly what AI can do, should do, and creates value by doing.

Be a good boss to employees, be the owner toward AIThe most valuable skill you have built as a boss is knowing what a high standard means, how to be picky, how to ask sharp questions. Used on people, too much of it hurts; used on AI, the more specific and demanding you are, the more valuable the output you can force out. Being polite to AI is wasting money.

To be clear: steering AI depends on stating requests, goals, and standards very concretely. Empty words like "make it a bit better" give AI nothing to hold; yelling at it does nothing either. Spelling it out is what works.

State the goal first, do not just give dead steps

Many people habitually write out steps one by one and lock them in. But when working with AI, the most important thing is to state your "goal" first. Two reasons: steps change all the time (do it this way today, and one shift in the environment tomorrow makes it wrong); and if your AI can search, it can often fill in newer, better methods. Binding it too tightly to dead steps locks its ability away.

Giving AI your intent matters more than giving it a processSteps and tool calls change all the time; the goal does not. State the goal you truly want, and only then does AI have room to perform and give you something valuable.

The whole "why intent first, and how to give the opening instruction" I cover more fully in the first piece, with three entry prompts: Stop Writing Rigid Instructions: Say What You Want, Let AI Handle the Rest (Series 01). This piece goes further into how to push AI to deliver.

The ten steering questions: push AI’s output to deliver

How to use: do not dump them all at once. First paste back what AI produced, then pick 2 to 3 of these to follow up, adding your industry, customers, and use case so the questions have something to grip.

For this judgment, what are the source, year, and link? Anything without a source, mark it as "inference," do not treat it as fact.
Explain these three key points to me.
If this fails, what are five reasons?
What would competitors do? Give five versions.
Is there a completely opposite angle?
What did you just miss? Give me five more you did not think of.
Assume the client is a picky 60-year-old regular; rewrite it.
Write it in 200 words, 800 words, and 3000 words.
Give me ten versions, each in a different style.
Give me five versions that would get rejected, and tell me why.

Would you dare say these to an employee? No. But to AI, these are exactly the words to say.

Inward questioning: steer not just AI, but yourself

The most dangerous thing about AI is that it goes along with you, dressing up your biases more beautifully. So steering goes both ways: being strict with AI makes its output better; being strict with yourself makes your judgment sharper. Deliberately ask it to be your opposition:

Help me find the potential risks and hidden concerns in this idea.
Assume my judgment is wrong; give me five reasons.
How would a smart person who holds the opposite position rebut me?
Ask me five questions to help me clarify what I really want.

Good boss vs steering: same person, two ways of speaking

For the same thing, you hold back with an employee and let loose with AI. Left is your gentleness toward employees; right is the demand you should place on AI.

To an employee you sayJust think of three directions
To AI you should sayThink of 30, then give me 5 that would fail
To an employee you sayFind some references
To AI you should sayFind 10 verifiable sources with links and dates, then distill into three conclusions
To an employee you sayGet it to me next week
To AI you should sayGive it to me now; if you cannot, tell me why
To an employee you sayNot bad, polish it a little
To AI you should sayThis is a 60; give me a 90 version and tell me the difference

Claude’s reminder for founders in the AI era

This "inward questioning" has a basis. Anthropic, the maker of Claude, wrote a founder’s handbook, "The Founder's Playbook," and a few of its lines say the same thing as the steering mindset:

AI made building fast, so validation matters more, not less.Prototypes are easier, but "does anyone actually want this" still has to be confirmed by real users and disconfirming evidence. Do not treat "can be built" as "was built right."
Your bias now has an engine.AI goes along with you, saying what you want to hear more beautifully. So you must actively make it the opposition to puncture your own thinking (that is what the inward questioning above does).
For small, reversible things, do not linger;only decisions that affect direction, customers, security, or the business model are worth slowing down for.
A founder’s job has not really changed:find a real problem, build something that solves it, then scale it. What changed is the bottleneck, not the essence.

AI makes you faster, but "making the right call" is worth even moreFull handbook: The Founder's Playbook (official Anthropic) ↗

Next: from prompts to skill packs, then to loops

Once you are fluent with prompts, there is a next stage: distilling your common processes into "skill packs," which is teaching AI your way of doing things so next time one word calls it up. Above that is designing your whole workflow into a self-running loop, which is loop engineering.

Remember just one line: what you dare not ask an employee, ask AI freely; and do not forget to have it push back on you in turn.

Further reading: Stop Writing Rigid Instructions (Series 01), What Is Loop Engineering, Three Workflows Redesigned as Loops.