What I Learned From Tennis Rally That Changed How I Think With AI
The one mindset shift that unlocks AI's true potential.
Every AI expert talks about "prompt engineering" like it's the holy grail.
Craft the perfect input, get the perfect output. Be specific, provide examples, define your role, set clear parameters.
I followed this advice religiously for months. I collected hundreds of "proven prompts" like trading cards. My Notion database was a work of art—categorized by use case, format, and complexity.
But my best AI conversations? They happened when I threw out the templates and just... talked.
The newsletter idea that got more than 10k views came from me saying "I don't know what to write about this week, help me find new ideas" and following wherever Claude wanted to explore.
This contradiction bothered me for weeks until I remembered something from tennis that changed everything.
The rally is where the magic happens
I used to play tennis religiously in high school. If there's one thing I learned that I can share with you, is this: anyone can practice a perfect serve, but great tennis happens in the rally.
People come to matches to see the heated rallies, not just flawless serves!
The longer the rally, the bigger the payoff.
It's true the serve gets the point started. But the rally—that back-and-forth exchange where you read your opponent, adapt to unexpected shots, and build toward something neither player could plan—that's where the magic happens.
You can't script a rally. You start with intention, then respond to what comes back. You read the court, then adapt to the changes. Each shot builds on the last, creating possibilities that couldn't exist without the exchange.
That's when it clicked: AI conversations work exactly the same way.
The two ways people use AI
Believe it or not, there are two types of people who use AI. Most people fall into the first category. They treat AI like a vending machine or a better version of Google:
Insert perfect prompt
Get "somehow" perfect output
End conversation
But the people getting transformative results fall into the second category. They treat AI like a tennis partner:
Start with rough intention
Build on what AI gives back
Rally back and forth until something emerges
What makes them different is a simple mental shift: they start conversations to think better, not just get answers faster. This changes everything about how they interact with AI. Instead of crafting perfect prompts, they ask messy questions. Instead of accepting first responses, they push back like a tennis player who knows the rally is where the magic happens—always looking for the next exchange that reveals something neither human nor AI could see alone.
What conversation-first actually looks like
Let me show you the difference with real examples from my work:
Content Creation: Template vs. Conversation
The prompt way:
"Write a newsletter about AI productivity for knowledge workers. Include 3 practical tips, personal anecdotes, and a clear call to action. Tone should be conversational but authoritative. 1500-2000 words."
Result: Generic productivity content that could have been written by anyone.
The conversational way:
Me: "I don't know what to write about this week. Given what you know about my newsletter, how can you help me?"
Claude: "What's been on your mind lately? Any frustrations or discoveries with AI?"
Me: "Actually, I've been noticing I get better results when I stop trying to engineer perfect prompts..."
Claude: "That's interesting—most advice goes the opposite direction. What changed for you?"
Me: "It's like... remember playing tennis? The best points weren't the perfect serves..."
Result: This exact post you are currently reading!
Research: Extraction vs. Discovery
The prompt way:
"Analyze these 5 articles about AI browser tools. Extract key features, pros/cons, and recommendations. Format as comparison table."
The conversational way:
Me: "I'm trying to understand this new AI browser space. What should I be paying attention to?"
Claude: "What's driving your interest? Are you looking to switch browsers or understand the market?"
Me: "I tried Dia for a week and it felt different, but I can't articulate why..."
Claude: "Different how? Walk me through a specific moment when it clicked."
The conversation revealed insights I never would have found with extraction-mode prompting.
How to shift from prompts to conversations
In my experience, the most important thing is to stay open to new possibilities. Learn to let go of controlling every AI response. Instead, follow your curiosity and see where it takes you.
Here are my favorite ways to do it:
1. Start with intention, not instruction
Instead of crafting detailed commands, share your actual situation and let AI help you clarify what you need.
Instead of: "Generate 10 content ideas for my newsletter"
Try: "I'm struggling to find interesting angles this week. What do you think my audience might be wondering about?"
When you're not even sure what you need help with, try the reverse interview:
"I want to improve my content strategy... ask me clarifying questions to help figure out what I actually need."
This turns AI into your thinking partner from the very start, helping you find what you're really trying to accomplish.
Follow-up prompts that deepen the conversation:
"Why do you think that topic would resonate with my specific audience?"
"What angle on this topic would surprise people who think they already understand it?"
"What questions are people probably asking about this that I could uniquely answer?"
2. Build on what comes back and chase tangents
Treat every AI response as a building block, not an endpoint. When AI gives you something interesting, dig deeper. Follow unexpected paths rather than steering back to your original agenda.
When AI suggests something intriguing:
"That's surprising—why do you think that resonates?"
"Wait, that's interesting—can we explore that angle instead?"
"What would the counter-argument be to this approach?"
"That reminds me of [related concept]. How do these connect?"
Your best insights often come from paths you didn't plan to explore. When AI mentions something unexpected, chase it down instead of returning to your checklist.
3. Admit when you're confused
This is my most favorite. Showing vulnerability sometimes generate better response to AI. In fact, some of the best conversations happen when you're honest about what you don't understand. Confusion is data, not failure.
When you're stuck or confused:
"This feels off to me but I can't articulate why. Help me figure out what's bothering me."
"I'm getting lost in the complexity. What's the simplest way to think about this?"
"This makes sense intellectually but doesn't feel right. What perspective might I be bringing that's creating resistance?"
4. Use AI as your thinking coach
Instead of asking AI to give you solutions, ask it to help you think better about the problem. Use AI as your thinking coach, not your answer machine.
To improve your thinking process:
"What questions should I be asking that I'm probably not considering?"
"How do experts in this field typically approach problems like this?"
"Help me stress-test this idea—where could it fail?"
The key insight: Each of these prompts is designed to extend the conversation rather than end it. You're not trying to extract a perfect answer—you're trying to think better together.
The real talk about implementation
Now, I know what you're thinking—this conversation approach sounds great in theory, but what about the real world?
"Wyndo, this sounds like more work, not less."
I hear you. Sometimes you just need a decent email draft and don't have bandwidth for deep exploration. In those moments, a well-crafted systematic prompt will serve you better than forcing a conversation. The key is choosing consciously rather than defaulting.
"Wyndo, do you understand about context limit? Long conversations make AI forget important details."
True. Exploratory conversations can hit context limits where AI loses track of earlier conversations. When you find something valuable through exploration, capture it and either start fresh with that insight or switch to systematic mode to build on it. Focus on productive discovery over endless back-and-forth.
Here's what I'd suggest: Taking 2 minutes to understand the situation often saves 20 minutes of frustration with generic outputs that don't actually solve your problem.
Develop the skill to have conversations when you need breakthrough thinking, and use systematic prompts when you need reliable execution. Both approaches have their place in your AI toolkit.
Why this changes how you think with AI
When you shift from prompts to conversations, something deeper happens beyond better outputs.
You develop AI fluency. Instead of collecting templates others created, you build the skill to have productive dialogues with any AI model. That skill transfers and compounds.
You think better with AI. The back-and-forth forces you to articulate your thinking, question assumptions, and explore angles you'd miss alone.
You stay curious longer. Prompt-first thinking optimizes for known outcomes. Conversation-first thinking stays open to surprise.
Most importantly: you stop being dependent on other people's prompts. You develop the confidence to start conversations from scratch and guide them wherever your thinking needs to go.
Your challenge today
Stop reading about conversation-first AI and start doing it.
Right now, think of your most frustrating AI challenge this week. The one where you've tried prompts and gotten mediocre results. The one you keep putting off because it feels too complex or you don't know where to start.
Don't craft a prompt. Don't plan your approach. Just open your AI tool and start exactly like this:
"I'm struggling with [your actual challenge] and feeling stuck. Help me think through this."
Then rally. Build on what comes back. Follow the tangents. Ask the follow-ups. Stay in the conversation for at least 10 minutes—that's where the magic happens.
The goal is the productive discovery, not an efficiency.
The people building AI conversation fluency today will have an unfair advantage tomorrow. While others are still collecting templates, you'll be thinking with artificial intelligence—and that fluency will transfer across every work domain.
Your first rally starts now. What will you learn when you stop serving and start playing?
Share it in the comment section.
As one who has hit a few buckets of balls over the years, this post rings a lot of bells. And it brings to mind a line from David Foster Wallace in String Theory:
"…if both guys are good enough so that there are few unforced errors to break up the rally, a kind of fugue‑state opens up inside you where your concentration telescopes toward a still point and you lose awareness of your limbs … all you know then is the bright ball and the octangled butterfly outline of its trail across the billiard green of the court."
Some special moments with AI it feels like that - and for the rest, most shots go into the net or over the fence.
I have been confused since day one about the obsession with the perfect prompt. I don't remember when I began playing with LLMs but I downloaded GPT and input whatever was on my mind, and lo and behold, we had an exchange of ideas. A collab.
Every answer leads to more questions. I summarize in my own words to ensure I understand concepts and ask GPT to drill down on minutiae I don't understand. I push back when GPT says something trite (which is often) and call it out when it omits something important. Recently, while conversing with GPT about why Anthropic is pulling ahead of OpenAI in the Enterprise, GPT did not mention security. What? How can you talk about Enterprise Computing and not talk about Security? I called it out and GPT tried to cover for its blunder and then we engaged in depth on the topic.
None of my interactions with Gen AI required any specialized training, no directions in fine print necessary. Isn't that the point of AI? LLMs are built on decades of NLP (Natural Language Processing) research. We’re engaging with a system designed to interpret and respond to human language.
The Perfect Prompt rhetoric puts pressure on people who are already intimidated and feeling left behind by the barrage of AI everywhere. My advice: play with any of the generative AI models available. It's designed to be user-friendly. The technology itself is super complex, but we don't need to make using it into something esoteric and intimidating.