Compounding AI While You Work
The “I Don’t Have Time for This” Guide to Prompt Engineering
Enter: The Email You’ve Written a Hundred Times
You know the one. Someone reaches out asking about your products or services, and you start typing out the same explanation you’ve typed a hundred times before. Your process. Your pricing. Why you’re different. You try to make it feel personal, but honestly, you’re mostly copying from the last version you sent and tweaking a few details. Twenty minutes later, you hit send and move on to the next one.
Now imagine next week that same email only takes you three minutes. Not because you sent something generic, but because your AI assistant has begun to understand your business, your voice, and the specific things that matter to different types of prospects. The email that went out felt more personal than your copy-paste version ever did—because you actually had time to think about this particular person instead of grinding through the boilerplate.
And here’s where it gets interesting: the week after, your AI starts to flag things like, “This person mentioned they’re coming from a competitor—based on past conversations, you usually want to address the switching process upfront. Want me to add that?”
This is a promise of AI that many have made, and few have yet to unlock. To get this level of nuance out of your AI systems, it requires large amounts of context and examples for your AI to draw from, and the big P word you’ve probably heard before in this space - Prompt Engineering.
Maybe you’ve seen the courses. The YouTube tutorials. The LinkedIn posts about “mastering ChatGPT” that assume you have endless hours to become an AI expert on top of, you know, your actual job. Frankly, that can seem daunting.
Here’s the thing: you don’t need all that. There’s a simpler way.
Let the AI Figure Out the AI Stuff
That might seem counterintuitive, but stick with me.
The conventional wisdom says you need to master prompting techniques to get real value from AI. Learn the frameworks. Study the syntax. Practice crafting the perfect instructions.
Turns out, that’s not quite right. AI systems are actually better at optimizing and writing prompts than humans are—in some cases outperforming human-written prompts by up to 50% on complex tasks.
They understand their own quirks in ways we simply don’t. Or at least it would take PhD level study to get there. Does anyone here have an extra four years lying around to dedicate to deep academic study of probabilistic information models? I sure don’t.
So instead of spending your evenings learning how to talk to AI, it turns out you can just... ask the AI to get better at working with you. While you do your actual work.
This is learning by doing instead of learning then doing.
Compounding Productivity
Before we get into the how, let’s talk about why this approach - though it will take some up front time investment - is worth it, even when it feels like time is exactly what you don’t have.
You know how compound interest works with money: a small amount invested early grows exponentially because the returns generate their own returns. You can enact that same principle while working with AI systems like ChatGPT, Claude, Gemini.
Each time you finish a task with AI, you can capture what worked and roll that into the next interaction. You’re making a small deposit into a system that pays you back on every future interaction.
The first few rounds might feel like modest wins—saving 15 minutes here, catching an oversight there. Not exactly life-changing.
But those gains stack. By iteration six or seven, you’re not just working a little faster. You’re working at a completely different level.
The Approach: Iterative AI Prompting
Here’s the whole approach, step by step. Give it a shot this week, and let us know how it went.
Step 1: Pick one thing you do regularly.
Don’t try to transform your whole life with AI all at once. Pick one repeatable task—responding to inquiries, writing proposals, summarizing meetings, following up with leads, whatever.
The best tasks to focus on might fall into one or more of these categories
High value workflow - Greater quality & efficiency in this area would have meaningful business upside
Bucket drainers - busy work that has to get done but you don’t like doing it and you want to automate
Network effects - many on your team can benefit from sharing this and iterating on it together
Ideally, you’re finding something that converges in these categories
Step 2: Do the work with AI prompts (as much as you can)
Open ChatGPT, Claude, or Gemini. Don’t overthink it. Upload your files, paste in previous examples, ask questions, go back and forth until you get something good. Tell it exactly what to change. Let it be messy. Messy is fine.
If it can’t quite get what you’re after, take over and finish the task manually, that’s part of the process.
Step 3: Ask the AI to capture what worked.
Here’s the key to the kingdom. When you’re done, ask this exact question:
“Based on my prompts in this chat, create a starter prompt I can use next time to get to this quality of output faster.”
That’s it. The AI will write detailed instructions based on everything it learned about your preferences, your context, and what “good” looks like to you.
Step 4: Save it somewhere you’ll actually use it.
You’ve got options here, from simple to sophisticated:
The quick and dirty way: Copy the prompt into a Google Doc or note. Next time, paste it into a new chat. It’s not elegant, but it works.
The better way: Save it as a reusable template in a custom GPT, a Claude Project, or a Gemini Gem. These let you start every new conversation with your custom instructions and examples already loaded—no copy-pasting required.
The best way - fit for purpose: Tools like Jasper, Missive (for email specifically), or Notion let you save custom prompts and templates for specific use cases like writing emails, managing marketing communications, and streamlining task completion, etc. These add cost, but if you’re doing high-volume work, they can be worth it.
Step 5: Keep feeding it your wins.
Upload your final deliverables as examples. Now the system doesn’t just know your instructions—it knows what good work actually looks like.
Rinse and repeat. Each cycle makes the next one faster. After three or four rounds, you’ll feel the difference.
What it Really Looks (and Feels) Like
All that sounds pretty straightforward, but let’s play it back through using that inquiry response example from the opener to fill out the picture further:
Round 1: Kind of a slog, honestly
Your first session takes a while. You paste in a few inquiry emails you’ve received recently and some responses you were happy with. You explain the different types of people who reach out—some are price-shopping, some are overwhelmed and need hand-holding, some are sophisticated and just want the facts.
You correct the AI’s tone three times. Too salesy. Too stiff. Weirdly formal. It’s a process.
But at the end, you ask:
“Based on my prompts in this chat, create a starter prompt I can use next time to get to this quality of output faster.”
You save what it gives you—whether that’s in a doc, a custom GPT, or wherever works for you. Done.
Rounds 2-3: Planting the seeds
Now you’re building momentum.
You open your saved prompt (or your custom GPT or Project), and the AI already knows your services, your tone, the different buyer types you see. You still tweak things here and there—adding a note to the instructions with specific phrasing to avoid, or clarifying how you handle pricing questions—but you’re getting to good output in half the time.
Most people stop here. And honestly, that’s a solid win already. But there’s more.
Round 4: Reaching the clouds
Something shifts around this point. You’re responding to inquiries way faster than before. But what’s more interesting is what that frees up your brain for (your organic intelligence, as it were).
You’re not having to grind through the mechanical stuff anymore—explaining your process for the hundredth time, remembering to include that paragraph about timelines, making sure you didn’t forget the call-to-action. That’s handled.
Instead, now you can spend your time thinking more about things you never had enough time for: What’s this person actually worried about? What’s between the lines in their message? What would get this conversation to moving towards a yes?
You’re starting to move from doing the work to directing the work.
Rounds 5-6 and beyond: Cooking in the castle (customer inquiries only please, no englishmen named Jack)
This is where it starts to get fun.
You update your instructions accordingly: “Before drafting, help me read between the lines of their inquiry. What might they really be asking? What objections should I address preemptively?”
And the system starts helping you apply that next level of thinking too. It notices patterns you hadn’t even articulated. It starts to flag things unprompted: “They mentioned they’ve been burned before—might be worth acknowledging that directly and explaining how your process is different.”
At this point, the AI isn’t just your typist. It’s more like a thought partner, helping you reach higher levels of consideration than you could. Hard to think critically when you’re just trying to grind through the inbox so you can shut down for the weekend.
Now that’s baked into the process, and the whole thing is happening faster than it did before.
Why This Works - Knowledge Compression
If you’re curious about why this works a bit under the hood, here’s the quick version.
Large language models are basically massive compression engines. They’ve taken an incomprehensible amount of human knowledge and squeezed it into something you can have a nice cordial chit-chat with.
When you follow this process, you’re adding your own layer of knowledge compression on top. Your communication style. Your expertise. Your organization’s weird quirks. Your stakeholders’ pet peeves. All of it gets condensed into a set of instructions and examples that constrain and direct the AI as it works.
And because each iteration adds to that foundation, it compounds. You’re not starting over every time. You’re building on top everything that came before.
The important thing to understand: this isn’t about replacing your judgment. It’s about encoding your judgment so the AI can extend it. You’re still the one who knows what good looks like. The AI just helps you get there faster.
The Bottom Line
You don’t need to take a course or watch a dozen YouTube tutorials to become an AI prompt Engineer. Turns out ChatGPT will always be better at it than you, and that’s okay. You just need to start doing your work with AI as your partner, and iteratively ask the system to get better at doing that exact thing as you go.
The people who are going to see major productivity gains in the AI era aren’t necessarily the ones who consumed the most educational content on LLMs. They’re the ones who started compounding sooner with small investments here and there.
The beanstalk starts with a single seed. Start planting yours this week!
This post was collaboratively brainstormed, researched, and drafted alongside AI (Claude Opus is our go-to lately for this sort of thing), then revised and edited heavily by real humans.
Conversint is a consulting firm helping SMBs and everyday humans unlock real value from AI and other modern tech. To learn more about our services and schedule a free initial consultation, visit our website at Conversint.ai.
If you learned something useful, consider sharing this post with someone else who may also benefit.
For more insights and guidance like this delivered to your inbox regularly, consider subscribing.