The 4 AI Tools I Actually Use

I just counted 47 AI tools in my bookmarks folder. Half of them don't exist anymore.

The other half I've never actually used for anything meaningful. And yet every morning, my feed explodes with "🔥 This AI tool will 10x your productivity!" posts from people who clearly haven't moved past the free trial. In my experience, 95% of AI tools are solutions looking for problems. But the remaining 5% can genuinely transform how you work.

Today, I’d like to give you a peek behind the curtain of what these are for me and how I use them in my day-to-day work.

Let's dive in!

Want more insights on how to use AI in a profitable way? Check my daily notes:

Avoiding Tool FOMO

Before we dive into the tools that actually matter, let me share a framework that'll save you from AI tool FOMO forever. I call it the Integration-Automation Matrix, and it's how I evaluate every AI solution that crosses my desk.

There are essentially four types of AI use cases you can build, and understanding where each tool fits will change how you think about AI implementation.

Low Integration + Low automation = Assistant 
Think ChatGPT copy-paste workflows. You manually feed it information, it gives you output, but it can't access your systems or do anything on its own.

High Integration + Low Automation = Copilot 
Microsoft Copilot in Outlook knows your emails and writing style, but it still waits for you to hit "generate." It's deeply connected but not autonomous.

Low Integration + High Automation = Autopilot
That website chatbot answering FAQs 24/7. It runs independently but exists in isolation – just a snippet pasted on your site with no connection to your actual systems.

High Integration + High Automation = Agents 
The holy grail. These tools connect to your systems AND run autonomously. Think customer support bots that can actually solve user requests like resetting their password.

Most AI tools die because they're stuck in the wrong quadrant for what people actually need.

The four tools I'm about to share either nail their quadrant perfectly or help you move between them strategically.

Here they are:

1. ChatGPT - My Swiss Army Knife

Let’s start with the obvious one – and probably the only AI chatbot I’m confident will still be around in five years.

ChatGPT is my daily driver. Not because it’s revolutionary, but because it’s reliable. While others are chasing the newest model or platform, I’m over here appreciating that ChatGPT just works.

The UI is clean. There’s virtually no usage limit (I’m on the Pro plan). And it handles 80% of what I need without breaking a sweat.

Calling ChatGPT a chatbot honestly feels like an understatement at this point. Yes, you can chat. But you can also:

• Run small use cases through Custom GPTs

• Use code execution (Python, plotting, etc.)

• Chat with or extract values from documents

• Generate high-quality images

• Impeccable speech-to-text

• Top tier voice generation

• Automate small tasks

For example, one of my Custom GPTs runs a 25-point checklist on my newsletter drafts. It reviews tone, clarity, formatting, and checks I’m not veering into lecture model.

Here’s how that looked for this one:

I still use Claude sometimes. Mostly because out of habit or because old projects live there. But I hit rate limits too often (even on a paid plan). I used to prefer Claude’s writing style—it felt more elegant. But since 3.7 and 4, it started sounding more generic to me. I can go as well with ChatGPT at this point.

ChatGPT sits perfectly in the Assistant quadrant – low integration, low automation. But that's exactly what you want for a daily GenAI Swiss Army Knife that just works.

2. n8n – My Workflow Workhorse

If ChatGPT is my assistant, n8n is my operations team.

Once something works in ChatGPT I often move it to n8n to go from “this works” to “this works on its own.”

n8n is a popular no-code workflow builder that sits between Zapier and a full backend stack. It’s flexible, scriptable, and it doesn’t force AI into places it doesn’t belong.

That’s my biggest gripe with many new "AI workflow builders": They wrap everything in AI, even tasks that should just be a simple if-else statement or spreadsheet update. Just because you can use an LLM doesn’t mean you should. n8n gets that.

It also handles agentic behavior really well (using Langchain under the hood). For example, I can simply create a standard n8n workflow (say, calculating a price from a spreadsheet), then register it as a callable function (tool) for an AI agent. Works like a charm: short, clean, and incredibly powerful. (Feels like MCP, but without the hype.)

Bonus: you can self-host the open source version. That makes it enterprise-ready and accessible—two things you rarely get in the same sentence as “AI tool”.

One example how I use it every day: 

• I drop transcripts, newsletters, or voice memos into a Google Drive folder.

• n8n monitors that folder, runs an LLM workflow to extract quotes and insights,

• stores them in my personal insight vault.

• This becomes my treasure box I can mine later for content, ideas, or talk material.

Other days, it’s just doing boring but necessary stuff—like processing form submissions, using AI to enrich them, and updating my CRM.

n8n lives in the Autopilot and Agents quadrant: low to high integration (depending on your setup), high automation.

3. Azure ML Studio – My Forecasting Friend

This one's a bit more niche, but it’s a quiet powerhouse – especially for anyone working in forecasting, classification, or time series analysis.

Disclaimer: I don’t use Azure ML Studio every day (anymore). GenAI work has taken over most of my time recently. But whenever I’m working with a client on something predictive this is still my go-to.

The standout feature is the AutoML pipeline for time series. You can benchmark everything from ARIMA to XGBoost to Facebook Prophet—no need to hand-code from scratch or deal with different packages. Just upload or connect your dataset, click through a few steps, and you get performance comparisons, downloadable models, and deployment-ready outputs.

Azure sounds expensive, but getting started isn’t. For AutoML there’s no subscription fee, you just pay-per-compute. A small experiment might cost a few cents. Larger datasets go up to a few dollars. Hosting a trained model on Azure can get pricey, but you can download and self-host it instead.

Azure ML Studio fits into the Autopilot quadrant – low integration (looking from a business workflow perspective), and high automation. You do have to prep your data and get it in there, but once it’s set up, it drives the process for you.

4. Google Colab – The Prototyping Playground

If you’ve never heard about Colab – it’s essentially a Google Doc that runs Python—and that simplicity is deceptive. Underneath is a full Jupyter Notebook environment with zero setup, GPU support, file access, version control, and seamless Google Drive integration.

Whenever I have a code-heavy idea—typically for some data analysis, or just testing some niche automations—I open Colab. There’s no fiddling with environments or worrying about breaking my system. It just runs. Often, I start in ChatGPT, let it generate some code, and then bring it over to Colab where I can repeatedly run it (or use Gemini to make updates).

Talking about Gemini – Google recently integrated it’s state-of-the-art model into Colab and the integration works surprisingly well (compared to other integrations in Docs or Slides). Gemini can read your notebook context, which means it knows the variables, functions, and structure you’re working with. Ask it to write code referencing cleaned_data, and it knows what you mean.

It goes even further: since a few weeks, the built-in copilot not only suggests what to do, but does it. I demoed this in a short video—you can see how it helps you step through data cleaning, analysis, and visualization workflows with minimal effort.

Colab is where I do all my:

• Exploratory data analysis

• Quick software prototypes

• Workflow scripting before moving things to n8n

If you work on more complex or UI-heavy software projects (or you don’t like notebooks for whatever reason), try AI-native IDEs like Cursor or Lovable instead. For starters, Github Copilot for VS Code will also do the trick. Whatever tool you choose – get familiar coding alongside an AI. It’s the future.

With regards to my matrix, Colab mostly lives in the Copilot quadrant, edging toward Agent—especially when I let it create entire notebooks from scratch while I enjoy lunch.

Pick the Right Tool for the Right Job

Tool FOMO is real. But if you understand the Integration-Automation Matrix, you can ignore the noise and focus on tools that actually fit your use case.

• Need a lightweight assistant? Start with something ChatGPT-like.

• Want real automation? Move to a workflow builder like n8n.

• Do classical ML automatically? Use an AutoML tool like Azure ML Studio.

• Trying to bridge gaps? Write custom code using an AI-assisted IDE like Colab.

These are my personal tools, of course.

In client projects – especially enterprises – one of the first goals is often to find your own version of this. Many try to pick one vendor that can deliver everything (looking at you, Microsoft!). But I find that too limiting. For example, the Microsoft Copilot integration so far is… underwhelming – to stay the least.

In any case, the goal isn't to use more tools. It's to use fewer, better ones—tools that actually help you get things done faster, better, or cheaper.

And that’s what these four are for me.

Want to see them in action? I'll use 2 of these tools in my upcoming workshop Dead Lists to Dollars to turn a raw list of emails into sales-ready leads with titles, buying role & estimated deal size. You can check out the details here.

That’s it for today.

See you next Friday!

Best,
Tobias