I've tested a ridiculous amount of AI tools this year.
ChatGPT. Claude. Gemini. Copilot. Perplexity. You name it.
And for a while, I was focused on trying every AI tool I could get my hands on.
I wanted to know what all of them did, how they compared, and whether I was missing something better.
What I found is that the people getting the most value from AI aren't necessarily using the most tools. They're using the right tool for each job.
After spending the last year testing AI tools, building automations, and talking with teams about how they're using AI, I've started thinking about AI in terms of five layers.
Each layer solves a different problem, and together they form the AI stack I use every day.
🎙️ Layer #1: Inputting Information
Everything starts with getting information out of your head.
This is why I've become such a huge fan of voice dictation tools. 🎙️
When you type, you're doing two things at once: thinking and writing.
You write a sentence, stop, reread it, second-guess it, delete half of it, and start again. That stop-start cycle is where most of the friction lives. 😵💫
Talking doesn't work that way. You just get the thought out (even if its messy and long winded).
This matters especially when you're working with AI, because the more context you give it, the better the output.
Most people would never sit down and type three paragraphs of background before a prompt. It feels like too much effort. But speaking that background information out loud on the other hand… takes seconds.
I've been using an app called Wispr Flow because it works almost anywhere I'd normally type:
- ✉️ Emails
- 📃 Documents
- 🤖 AI prompts
- 🔎 Browser searches
But the real benefit isn't speed.
It's that it removes the "blank page" problem entirely. Instead of sitting there trying to write the perfect prompt, you just start talking (or honestly, word-vomiting lol).
📚 Layer #2: Documentation
This is the layer I think most people underestimate.
Not because it's hard. But because it's unglamorous.
Nobody gets excited about writing down "Step 4: click save" 👀
But here's the lesson I keep coming back to:
You can't automate what isn't documented.
And if you ever want AI to take over parts of that workflow, documentation isn't optional; it's the starting point.
You genuinely cannot build agentic or automated workflows without it.
This is one of the reasons I'm a big fan of Scribe, who sponsored last week's video.
I use it in my own business, for accounting procedures, onboarding contractors, SOPs, and even leaving step by step instructions for my software consulting clients.
You just do the task once, and Scribe turns it into a step-by-step guide with screenshots automatically.
That's what I love about it. Documentation no longer feels like a separate project you have to remember to do later. It happens as part of the work itself.
Then once those processes are documented, you have the foundation for automation, AI workflows, and everything that comes after.
And if you'd like to see how Scribe could work for your team, you can book a personalized demo with their team here: www.scribe.how/rowe
💼 Layer #3: Work
The next layer is where most people spend their time today.
This is where AI helps you do your everyday work.
Think things like:
- ✉️ Summarizing emails
- 🤝 Preparing for meetings
- 🔎 Finding information
- 📄 Drafting documents
- ✍️ Researching projects
For Microsoft users, that's often M365 Copilot.
For Google users, that's Gemini.
Here's what makes this layer different from just opening up any AI chatbot:
It already knows your job. 💼
Your meetings. Your email threads. Your documents. Your team.
So instead of doing this:
"Can you summarize this project update?" (after you paste in three emails, a Teams thread, and your meeting notes)
You can just ask:
"Where do we stand on this project?" (with nothing pasted in at all)
And it pulls the answer directly from your inbox, your chats, your documents, and your meetings—because it already has access to them.
A few ways this actually looks day-to-day:
- In the morning, you can ask it to summarize your unread emails so you start the day already caught up
- Before a vacation, you can ask, "What do I need to wrap up before I'm out?" and it pulls straight from your inbox, calendar, and chats
- Before a meeting, you can ask, "What's the latest on this project?" and it gathers the relevant threads, docs, and updates on its own
No copy-pasting. No gathering context yourself. No explaining the backstory.
That's the shift.
AI isn't just smarter - it already knows what you know. 👀
⚡ Layer #4: Taking Action
This is where AI stops assisting and starts acting.
In the work layer, AI helps you complete tasks. In the action layer, it can actually do some of that work for you.
We're starting to see this from Anthropic, Google, Microsoft, and just about every major AI company. Instead of giving you a draft, these tools can work across your email, calendar, files, and apps to get things done.
For example, before a vacation, you could ask AI to prepare a handoff document. It can review your emails, meetings, and messages, create a summary of active projects, draft your out-of-office message, and even go into Outlook and turn it on for you.
Or you could ask it to find unanswered questions buried across your inbox and chat history. It can gather the messages, draft responses, and have them ready to send with a single “approve” click.
That's what makes this layer different. AI isn't just generating content anymore. It's taking action across your tools and handing the final decision back to you.
🤖 Layer #5: Agentic AI
This is the layer everyone is excited about right now: agentic AI.
These tools don't just help you with a task or take actions on your behalf. They're designed to run entire workflows with minimal human involvement.
A few tools already exist in this space:
- n8n – powerful, but built more for developers
- Lindy / Zapier Agents – more accessible, but still require a separate platform and setup
Personally, I'm keeping an eye on Copilot Studio and Gemini Enterprise, which are making agent building much more approachable for non-developers.
But regardless of which platform wins, one thing doesn't change:
Every agent needs context.
And that context usually comes from a documented process.
The teams that get the most value from agentic AI won't necessarily be the ones adopting the newest platform first. They'll be the ones that already know how their work gets done.
That's why I think documentation is the most overlooked layer in this entire framework. You can skip straight to agents, but eventually you'll end up right back at documenting your processes.
And that's really the point of these five layers: the best AI systems aren't built by stacking more tools. They're built by putting the right tools in the right places.
🎬 Video
If you want to see my full breakdown with demos & visuals of these applications you can check out the full video below 👇