3 Things I Learned About AI That Save Me Hours Every Week
Over the past years, I’ve been learning and writing about AI.
One thing I noticed is that we often waste time and energy chasing the latest AI news and trying tools that we’ll never actually use—instead of focusing on mastering what’s already available. While learning new things is always valuable, our time is limited, so it’s best to invest it in something that will truly make a difference.
Here are three things I learned about AI that save me many hours every week.
#1 Invest your time in learning only a few AI tools (here are the ones that matter)
Every week, there’s a new AI tool promising magical results to solve one or multiple tasks. There are so many that it’s hard to know which to learn first.
The good news is that you don’t need to learn them all, but choose the minimum number of tools that can cover most of your needs.
AI tools can be split into two categories:
General-Purpose Tools: Tools that are good at multiple tasks
Task-Specific Tools: Tools that were designed for one specific task
There are two general-purpose tools I think everyone should use: ChatGPT and Gemini. Both are good tools for tasks like writing, coding, solving math problems, searching the web, generating images, etc. Master at least one of them.
General-purpose tools are enough for most users. Still, there are more powerful options available in each specific niche.
Before you learn a task-specific tool, follow these steps:
Identify your needs
Find what tools can cover your needs
Compare them and learn to use the winner tool
For example, a few months ago, I needed a research tool to help me when writing. ChatGPT, Perplexity, and Gemini have a Deep Research tool that could help me with that. After comparing them, I found that ChatGPT’s Deep Research is the best tool for me, so I learned how to use it.
That’s how you find a task-specific tool.
Below is my essential task-specific toolkit. Most office workers will find at least one of these tools useful:
Google’s AI Studio: It has a real-time screen sharing feature that helps you explore, troubleshoot, or streamline any program or task on your computer
Perplexity: I use it for lightweight search. It’s my replacement for Google search
NotebookLM: It helps you digest information (PDFs, Google Docs, links) with fewer AI hallucinations
For more info about these tools, demos, and why I think they’re must-haves, check out this article I wrote.
#2 Create, reuse, and manage prompts efficiently
Just like we use English to speak with another person, we use prompts to talk to AI. People usually struggle with prompts because they:
Try to use all the elements of a good prompt—or use none of them
Don’t reuse or manage their prompts (saves a lot of time in the long run)
A good prompt has the elements below, BUT we rarely need to include all of them. Most of the time, we’ll only need to include the task and context in our prompt. Also, as ChatGPT gets updated, adding some elements becomes less necessary.
Here’s a prompt example that only has task and context:
I’m a 75kg man who wants to gain 5kg of muscle in 1 year. Build a 1-year training program to follow. I don’t have previous experience and I can train 4–5 days per week (60–75 min per session)
Here’s why you don’t need the other elements:
Examplar: When we're unfamiliar with a topic, we rarely have a good exemplar to show ChatGPT to generate a better response
Persona: Adding the “you’re an expert in … “ in persona isn’t that necessary in newer versions of ChatGPT (you could add some extra details, though)
Tone/Format: They’re nice-to-have elements, but they won’t drastically improve the response you get
Don’t get me wrong—I do create more robust prompts that include the elements mentioned above. In fact, many of them are stored in a prompt database that I frequently reuse (more on that later). I typically create these detailed prompts for tasks and projects that require high precision. While I don’t create them every day, whenever I do, I make sure to save them for future use.
For example, if you’re a personal trainer, you’ll need a more robust prompt that you can use with different clients. In this case, you could add an exemplar, persona, format, and tone as shown below.
Task: Build a 1-year training program for a client
Context:
[x]kg [man/woman] who wants to gain [5]kg of muscle in 1 year
Can train [x] days per week
[x years] of experience
Exemplar:
<You attach a workout planner template as a reference for the 1-year training program>
Persona:
You’re a bodybuilding coach with 10+ years of experience preparing recreational lifters for physique goals. You stay evidence-based, cite key references where relevant, and individualize programs.
Format:
<More details of the table output for the 1-year training program>
Tone:
Think of motivating and knowledgeable coach who is encouraging yet direct.
To sum up, always include the task and context, and reserve the other elements for work that requires detailed explanation and precision in the response.
This last robust prompt is a good example of a template that can be saved in a database and reused later with new clients. An easy way to reuse and manage prompt templates is with text expanders.
Mac users have a built-in Text Replacement feature that can help with this. Just go to Apple menu → System Settings → Keyboard → Text Replacement (in text input). The window below will pop up. Click the + button to add a new prompt shortcut.
After that, every time you write !prompt, you’ll get your prompt template.
This built-in Mac feature is a good start, but as you add more prompts, your list becomes messy and there’s no easy way to manage them on your Mac.
There are more powerful apps that can help you manage your text expanders, such as TextExpander (Windows/macOS), Beeftext (Windows), and Alfred (macOS).
This article is exclusive to paid subscribers (see extra perks here). To continue reading, please consider becoming a paid subscriber 👇