How I'd learn AI from Scratch in 2026.
A lot of people ask me the same thing now: how to learn ai? Here is it.
How I'd learn AI from Scratch in 2026.
A lot of people ask me the same thing now: how to learn ai? Here is it.
I've been using AI heavily since ChatGPT blew up in 2023, and the thing that became obvious fast is that most people are still mostly in a back-and-forth inside chatbots.
Step 1: Chatbots
I'd use ChatGPT, Claude, and Gemini side by side for a bit, just to get a feel for them.
How one explains better. How one writes better. How one is faster. How one is more useful when you need something practical done.
For example, I like the way Claude talks much better than ChatGPT. For research tho, Chat beats it.

Claude AI
Gemini is kinda around the corner for everything… It's good for deep knowledge tho, I sometimes give it a try when thinking about problems. Not much else.
Step 2: AI Tools
So after chatbots, I'd start moving real pieces of my actual work into specialized tools.
If you read a lot or do research, try something like NotebookLM.
If you make presentations or need to turn messy thoughts into something visual, try Gamma.
Try also ChatGPT's Deep Research. It's crazy good.

ChatGPT Deep Research
Most people would already be ahead of 90% of people just by getting genuinely good at a small set of tools. Because they're so much better than chat bots for the thing they're specialized on.
Like Zuck says, even if AI stopped making progress today we'd still have around 5 years of innovation. Because we haven't harnessed even 1% of all the model's power.
Start thinking AI-first,
Step 3: Automation
Once I had that base, I started going deeper into automation. Tools like Codex and OpenClaw, more agent-like workflows.
If you don't know what an agent is, is just an AI in a loop until it thinks it has finished and decides to stop, connected to tools.
These tools can be edit code (for coding agents, like Claude Code and Codex), or general tools like managing your calendar, touching spreadsheets, or basically anything you can imagine.
OpenClaw became a hit in 2026 because it made what before was mostly only for coding available into more coding tasks.
And it's awesome. I'm not going to dive deep into what it can do, but it's worth messing around with it

messaging my openclaw at telegram
I think the people who will win with AI aren't the ones who know the most theory. They're the ones who made it part of how they work.
Start simple. Use multiple chatbots. Learn the differences.
That's it.
And this is personal for me, it's a big part of why I'm building Disparity.
I want learning AI to feel practical and clear for normal people trying to keep up with what's coming.
I'm launching a mobile app for it soon; check it out here: https://disparity.app
Hope you liked it and served a great purpose.
Till soon.
메타데이터
- post_id
- cfda4a44707a
- slug
- how-id-learn-ai-from-scratch-in-2026-cfda4a44707a
- url
- https://medium.com/@pedroschott.work/how-id-learn-ai-from-scratch-in-2026-cfda4a44707a
- canonical_url
- https://medium.com/@pedroschott.work/how-id-learn-ai-from-scratch-in-2026-cfda4a44707a
- author_url
- https://medium.com/@pedroschott.work
- status
- ok
- fetched_at
- 2026-06-22 17:31:34