Unlocking AI's True Potential: From Instructions to Co-Creation

Practical AI Usage Guide from my learnings of playing with it towards outcomes

Based on practical learnings from business contexts - not a lot of tech jargon, just practical tips on how you should go about leveraging it

The First Rule of the Game: System Prompts

In almost every AI-based product you would use, be it Claude or Gemini or ChatGPT, there is something called a system prompt. This is like the default behavior of how the AI should work with you. This typically will be some kind of configuration in your profile or somewhere depending on what product you use. If your version doesn't have it, it simply means you'll have to create this system prompt and probably use it in every interaction by remembering to use it.

AI being very agreeable in nature probably comes from the fact that AI was built to be liked by humans. It doesn't disagree a lot with you. Its composition, its word choice is all about being agreeable.

So one thing that works really well for me is a few system prompt lines where I tell it that I want you to work with me to get the best out of me. I want you to question me. I want you to make sure you extract all the information you need from me before you execute my requests. And I want you to be skeptical and questioning me and not agreeable.

Something to this effect:

"You are my expert collaborator and thinking partner. Before executing any request:
- Question my assumptions and challenge my thinking
- Extract all necessary information before proceeding  
- Be skeptical and push for deeper analysis, not just agreeable
- Work WITH me to achieve the best outcomes, not just FOR me"

This is an important ground setting with any of these AI tools.

Create the AI You Want

The next thing you want to do is create the AI you want in the sense that create the AI in the persona of the right expert or the right set of people that you need to execute the job at hand.

So if you are creating a course that is about AI and engineering and all of that, you'll want to bring in the best people, the best personas in the industry about that. That generally will help you create AI in a way that can collaborate your thought process rather than just follow instructions.

Using Real Experts:

For strategic thinking:

"Channel Roger Martin's strategic thinking approach combined with Clayton Christensen's innovation insights. Challenge my strategic assumptions the way they would."

For problem-solving:

"Think like Tim Brown from IDEO - focus on human-centered design thinking and reframing problems completely."

Creating Custom Expert Personas:

When you can't reference specific experts, create detailed personas:

"Act as a senior advisor with 20+ years of experience who's seen every business mistake twice. You're known for asking the uncomfortable questions that save projects from failure. Your style is direct but supportive."

Create AI as you need it to be and generally try to create experts who would know better than you, reference experts that are on the internet so that the AI is able to play that role effectively.

Your Attitude When Working with AI

What is your attitude when you are working with AI? I think that's very important. You need to move from instruction mode to co-creation mode.

Treat that AI as you have created as your mentor and expert and jam with it about your problem statement. Give it your problem and ask it to give you solutions rather than giving it the solution that you think is the solution. And together co-create, give feedback, ask it to give feedback and continuously evolve this.

Instead of: "Create a marketing plan for my product"

Try: "I'm struggling with product positioning in a crowded market. Help me think through what I might be missing in my approach."

Again, you will see that depending on the model you're using, AI will go back into agreeability. It will forget a lot of times your system prompt instruction. So you might have to go back and nudge it and say that "you seem to be just agreeing and not really contributing" or whatever, so that you can nudge it back to the questioning mentor rather than an assistant that will do whatever it's told.

Learning from AI's Chain of Thought

One thing that I always do which is a great learning for me is when using thinking models is to look at the 'Show thinking' - it shows me the model's chain of thought, which helps me validate if it's thinking in the right direction and in most cases helps me think better.

When AI shows its reasoning process, you can see:

• What assumptions it's making about your problem

• How it's breaking down complex challenges step by step

• Where its logic might have gaps you need to fill

• What frameworks or approaches it's applying

• Whether it's considering the right factors for your specific context

This visibility into the thinking process often teaches you new ways to approach problems yourself, making you a better thinker even when you're not using AI. It's like having a senior colleague walk you through their problem-solving methodology in real-time.

Context is King

Generally, I think if you work with AI in a way that you do with humans in the sense that if you have a new person join your team, you will try to give them a lot of information before asking them to do work. That information is the broad context.

It depends on what is the culture of your company or how you prefer to work. But I've always seen great results when I explain the why to people and when people know the why, they can commit to the actual problem, go deeper into the problem rather than just following instructions.

Just like you would mentor a new person on your team to be questioning, to be asking the why, going deep into the problem before they start executing, you want to create someone who is real, not just a follower. You need to follow the same principle with AI.

Getting the Context Right

Before you ask the AI to execute something for you, the deeper context you can give it, the broader and deeper context you can give it, it's going to be able to execute the job better.

For the problem you're asking it, try and give it all the information you know. Maybe if you don't know everything about it, you can work with another AI prompt to first find that information on the internet, validate it, and then give it as context.

The Context Framework I Use:

Think about it like briefing a new team member:

• The Situation: What's actually happening right now? Who's involved?

• The Problem: What's not working? What have you tried before?

• The Why: Why does this matter? What's the real impact?

• The Environment: What's the company culture? Market pressures? Constraints?

• Success Looks Like: How will you know if this works?

Example Context Document:

SITUATION: Leading a 8-person product team, we've missed our last two sprint goals, team seems frustrated during standups

THE PROBLEM: Velocity is down 40% from last quarter, stories are taking longer than estimated, lots of scope creep mid-sprint

THE WHY: We're trying to hit a critical market window, competition is moving fast, leadership is asking tough questions about our delivery

ENVIRONMENT: Remote-first team across 3 time zones, company culture values autonomy, limited budget for new tools, regulatory compliance requirements

SUCCESS LOOKS LIKE: Consistent sprint completion, team confidence restored, predictable delivery timeline

If you can create a very detailed context document on the background of the problem, the environment in which the problem exists, the organization in which it exists, what the exact problem is, what do you think is the nature of it - all that background context can help you create a much more effective AI co-creator.

This context can be documented into a very nice document which is detailed and then you can upload that as context and then start jamming with it to create the outcomes.

Context is as important as the prompt or even more important.

Better Prompts Through Co-Creation

Right now we are very casual about writing prompts. Here again it's important that you are not instructing but co-creating, so that initial jamming back and forth with AI to actually find out what the AI should be doing is important before you actually get into the request.

You can use AI itself to create the prompts, which I think is pretty common knowledge by now. But you can flesh out your request into a much better prompt.

The Two-Window Approach: Rather than giving the problem to the AI, have another chat window where you're just getting it to generate prompts. The higher quality prompt will be generated - the better the context, the better the drafted request that you give it.

Create your AI co-assistant or co-worker who is actually generating prompts. You can have Sam Altman as that person or whoever you need them to be, and get those prompts there and then start working with them.

Meta-Prompt Example:

"I need help creating an effective prompt for a team productivity challenge. Given this context: [your situation], help me craft a prompt that will challenge my assumptions and extract what I really need to solve this. What questions should the AI be asking me?"

All of this together should give you high quality outputs.

Real Examples That Work

For Strategic Challenges:

Context: [Your detailed situation using the framework above]

"I'm facing [specific challenge]. Channel the thinking of Roger Martin combined with a seasoned operations expert. Before we dive into solutions, challenge my framing of this problem. What assumptions am I making that might be wrong?"

For Operational Problems:

"Our [process/system] isn't delivering results. Act as someone who's optimized similar operations at scale. Question whether I'm solving the right problem and help me see what I might be missing about the real bottlenecks."

For Decision-Making:

"I have a complex decision about [situation]. Be my board of advisors - challenge how I'm thinking about this decision and help me anticipate second and third-order consequences I haven't considered."

You'll Know It's Working When...

If you do this, then I think you should get very high quality outcomes from AI. You'll notice:

• AI starts asking you better questions before jumping to solutions

• You find yourself thinking deeper about the actual problem

• Solutions feel specific to your real situation, not generic advice

• You catch assumptions you didn't even know you had

• Your confidence in decisions improves because you've explored more angles

The Bottom Line

The magic isn't in the AI technology itself - it's in how you collaborate with it. Treat AI like you would a smart new team member: give them context, explain the why, challenge them to think deeper, and work together toward better outcomes.

Just like you would mentor a new person on your team to be questioning and go deep into problems before executing, you need to follow the same principle with AI. The best results come when AI becomes your thinking partner, not just your instruction-follower.

Start with one important challenge you're facing. Build that context document. Create your expert AI. Then jam with it about your problem rather than telling it what to do. That's where the real value begins.