F2F #31: Spaghetti Prompting: You’re wasting time with ChatGPT

Yes, I know I promised to share my new project in the last issue. But I gave a training to my employees at MarsBased today about prompt engineering and I have come up with this concept accidentally and I felt the a-ha moment.

It has to be done.

🍝 What is Spaghetti Prompting?

If you are a developer, chances are that you have heard the expression spaghetti code.

We've all seen spaghetti code: it's messy, verbose, hard to follow and impossible to scale and maintain. It usually is the consequence of not having planned what you wanted to develop beforehand and just YOLO-ing through the requirements at top speed trying to solve immediate problems without thinking about the maintainability of the whole thing.

Now meet its AI cousin: Spaghetti Prompting.

It’s what happens when you:

• Start with a vague prompt.
• Get a vague answer.
• Keep replying with half-baked follow-ups.
• Lose context by iteration #6.
• End up closing the tab and blaming the AI.

Result? A 15-minute task becomes a 45-minute loop of frustration.

🛑 Why Spaghetti Prompting fails

Like spaghetti code, it technically works. You eventually get something usable.

But doing it this way is:

• Inefficient.
• Slow.
• Hard to comprehend.
• Mentally draining.

You’re making the AI guess your intent instead of guiding it. By cutting corners and trying to save time on the initial prompt, you will have to write other prompts to correct course and fine-tune ten times more.

✅ Doing it right: Structured Prompting

You need to start with clarity. The best prompts follow this formula:

[Persona] + [Task] + [Context] + [Constraints] + [Examples] + [Desired Output Format (Optional)]

Example:

"You're a senior product marketer. Write a 3-tweet thread announcing our new feature: user roles. Audience: B2B SaaS founders. Tone: confident and helpful. Include a call to action".

Compare that to:

"Give me a tweet announcing my product".

🧪 Spaghetti Prompting in action

I'm going to show a real chat example. I was trying to change the slogan for the MarsBased website.

Generic interactions with ChatGPT usually run like this:

User: What's a good name for our new product?
ChatGPT: How about “TechBoost”?
User: I don’t like that. Try again.
ChatGPT: How about “Boostify”?
User: It’s for devs, not marketers.
ChatGPT: DevFuel?
User: It’s too generic.
ChatGPT: What does the product do?

See the issue?

Now here’s the structured version:

You're a branding expert. Suggest 10 name ideas for a developer-focused platform that speeds up CI/CD pipelines.
Requirements: short, punchy, techy. Audience: engineering teams. Avoid overused buzzwords like “cloud” or “AI”.

Or even better, use a ChatGPT thread where you throw in your half-assed prompts and it helps you to build the real ones:

Copy it from here for your convenience:

We will use this thread as a prompt refiner. I will input prompts and you will guide me through questions to get all the necessary information you need to create the perfect prompt for each task. Once you have the information, you will rewrite the prompt and explain it to me in terms of every item in its structure so we can fine-tune and do flawless prompt engineering.

This thread will reply with a set of questions to gauge more information off you. Once answered, it'll spit back the refined prompt. Use this final prompt instead.

💡 Takeaways

Some parting thoughts:

• Time is leverage: One good prompt = 10 iterations saved.
• Clarity is a superpower: If you can’t describe what you want, neither can the AI. Think things through (difficult to pronounce, if you're not native!).
• Treat AI like a smart intern: Give it context, constraints and feedback.

Before you hit “Send”, ask:

• Have I defined the role/persona?
• Is the task specific enough?
• Did I give enough context?
• Are there any constraints (tone, length, style)?
• Can I include examples?
• Optional: Have I defined the output format?

Spaghetti Prompting isn’t just inefficient; it’s a symptom of unclear thinking. And while we're at it, I'll claim to have invented this expression 🤘

Treat prompt writing like product specs or instructions you'd give to an agency or a freelancer: the better the input, the better the output.

Or, as I like to put it: shit questions deserve shit answers.

This week's edition is pretty long (more than I initially expected on an impromptu idea like this one), so I'll leave out the recommended content + updates + asks and all of that for the next issue. If you miss it, drop me a line so I can feel guilty about it! 😬