Most people know they need to automate with AI, but they have no clue where to start.
They hear about workflows, agents, and ten different AI tools and feel immediately overwhelmed.
Too many tools, too complicated, and too much friction for getting started.
So, they resort to doing nothing and end up wasting hours per week on manual tasks that AI could've handled if they'd followed the right steps.
In this article, I am going to give you the exact three-step framework I use to identify, design, and build any automation in under 10 minutes - no coding or expensive AI tools required.
This framework applies to any AI automation you want to build for any aspect of your life.
Stick around until the end of the article (bonus section), where I give you two free AI automation resources.
Contents
I: What do I automate?
II: Designing the solution
III: Execution layer
Bonus: my \free\ automation resources bundle
Step One: What do I automate?
Before you touch a single tool, you need to know what you are actually automating. This might sound obvious, but most people skip this step entirely and end up drowning in options with zero useful automations actually built.
I want you to conduct a time audit exercise: write down every task you do on a daily, weekly, and monthly basis. Do NOT filter anything out at this stage.
The goal is just to get it on paper (or organized digitally); it doesn't matter whether or not you think it can be automated.
Three categories: Daily tasks. Weekly tasks. Monthly tasks.
Tip: actually go through a full day/weekly routine so you can accurately time log your tasks.
Once you have your list, go through it and ask two questions for each item:
- Can this be fully automated?
- If not, can AI handle 50-80% of it, so I only handle the final step?
That second question matters most. A lot of people look at their list and think nothing is automatable because they are thinking only about full automation. But partial automation is just as valuable.
To give you guys a tangible example: I use AI in many ways to help with my writing here on X. However, I do not want AI to write my posts end-to-end. So, I use AI to handle all the research, scraping, grammar checking, and grunt work that goes into hitting the publish button.
The curation and taste are still mine, but I leverage AI systems to automate tasks that don't require human touch.
Pro tip: Use voice prompting to build this list. Open Claude or ChatGPT and explain that you want to map out everything you do daily and weekly in order to automate it. Ask it to prompt you with questions. You will remember things through conversation that you would never think to write down.
An example time tracker log:
At the end of the article (bonus section), I give you my exact template - don't worry about creating your own.
Step Two: Designing the solution
Now, you have your list. What I want you to do now is just pick one thing to start with.
Ideally, this is a high-leverage daily task that, if automated, could save you 30-60 minutes+ per day.
For me, automating my entire content research system (via Slack + Perplexity Computer + OpenClaw) saved me hours per week.
For you, this could be spreadsheet work, email management, calendar management, to-do list tracking, accounting, creative design work, etc.
The biggest mistake people make at this stage is trying to automate everything at once. They look at a list of 20 tasks and start building half-functional workflows for everything.
Don't do that. Instead, pick the highest-leverage item and commit to automating just one workflow per week.
In a perfect world, that means in 2-3 months from now, you'll have 10+ automations running at once.
Once you have your task, use this prompt in Claude (or any LLM) to reverse engineer exactly how to automate the workflow you need.
I also give you access to the completed mega prompt in the bonus section.
Fill in your specific details, and you will get a clear workflow (the steps, the tools, and the order for executing them).
One important note on constraints: most people do not need complex automation. They do not need webhooks, APIs, and five different tools chained together. All you really need is something simple that works.
To make this tangible, I'll give you real examples below (with the tools I personally recommend).
Step Three: Execution layer
Two tools handle the majority of automations I run. Both use agent architecture, meaning they do not just answer questions; they execute actual workflows on your behalf.
If you're a complete beginner, just pick one of these tools:
- Perplexity Computer
Since its release, Perplexity Computer has quickly become one of my most-used AI tools. What makes it different from a standard LLM is that it leverages multiple AI models inside a single workflow.
It can use Gemini for research, Claude for synthesis, and GPT for structuring. It automatically selects the best model at each step, so you don't have to manage any of it.
A real example from my own workflow: I use it to build a daily content brief for the AI Edge team. The prompt tells Computer to research developments in AI, identify the strongest ideas, and package them into a ready-to-use brief. It spins up multiple sub-agents running in parallel, synthesizes the outputs, and delivers the brief automatically.
What used to take my team and me hours of manual research now runs automatically every morning.
Perplexity Computer's only downside is its cost. It works extremely well and is genuinely useful, but that productivity comes at a premium. If you don't have the means to invest in a premium tool, resort to the alternative option below.
2. Manus
Manus is the more accessible alternative, and for most workflows, it performs just as well. Same agent architecture (it can research, synthesize, create files, generate images, and automate recurring tasks), but at a lower price point than Computer (with usage limits).
A practical example: give it a prompt to generate a daily intelligence brief, scan for the top developments across your niche, and deliver a concise summary to a log file.
The thing I like about Manus is that it can generate images, build slide decks, and handle visual outputs (useful if your workflow involves thumbnail ideas, presentation drafts, or visual research).
using Manus alone.I've created many viral infographics on@aiedge
For 90% of the automations you will want to build, sticking with Perplexity Computer or Manus will handle it entirely.
The only exception to using these tools as your daily drivers is building highly specific workflows that may require an API connection, feeding live data into a dashboard, and/or syncing data across multiple platforms.
For those workflows, the automation architect prompt from Step Two will flag what additional setup is needed.
I also recommend using Cowork to handhold you through those more complex workflows.
Bonus: free resources
As promised earlier, I made two free resources to get you started on your AI automation journey.
First, I built a template to assist with the time-logging exercise I outlined in Step One.
Secondly, I built a mega prompt you can use across any LLM to help you identify exactly which tasks you should automate (Step Two).
You can find the above resources here:
Closing
That is the entire framework. It's simple, repeatable, and it compounds over time in a way that genuinely changes how much you can get done with AI.
I hope you found this article valuable and practical. I'm aiming to deliver content exactly like this every single week.
If this is the kind of content you want on your feed, follow me at @aiedge_ and @milesdeutscher.
Across both accounts, I publish at least 3 articles a week on the hottest topics in AI.
Lastly, if you can, please Like/Repost this article so others can find it 💙
