The future of education
6 min read
I lead AI & Data at Hubla, a platform where people sell their knowledge through digital products. I spend a lot of time thinking about how AI could change the way people learn from them, especially here in Brazil.
Someone knows how to position a business, grow an Instagram account, or make videos that sell. They turn that experience into a course. Hubla provides the infrastructure to sell and deliver that knowledge.
I wonder how much of that experience will still look like a course in a few years.
My instinct is that many people would rather start with something they want done. They have an account that isn't growing or a product they struggle to explain. The request might be "fix my store and write next week's posts." I want the AI to help me finish that work, and I think I can learn from the decisions we make along the way.
I could be projecting my own preferences. The popularity of Reels doesn't prove that nobody wants a long lesson anymore. But I think there is room for an experience that starts with the learner's situation and develops from there.
Imagine following an expert whose explanations about positioning make sense to you. You like how she looks at a business. You would like her to look at yours.
Say I sell handmade pottery. People like my posts, but orders are slow. I share my Instagram and store with her AI and ask it to improve the pages and prepare my next posts.
The conversation shows it reading my Instagram profile and following the link to my store. I can open each action to see the material it reviewed and its findings. My bio describes my process but never says I take orders. On the mug's product page, "made to order" has no production time attached. That still doesn't tell us why sales are slow. It asks how long a mug takes to make and what buyers usually ask.
I explain that production takes three weeks and most messages are about whether a gift will arrive in time. It writes a clearer bio and adds production time to the store copy. Then it prepares a calendar, captions, and a Reel storyboard. I have drafts I can review and use.
Beside the gift Reel, it explains why it included the production time. I ask for a shorter opening and wonder whether we should leave the wait out. It writes a shorter version that keeps the timing visible, explaining that gift buyers need it to decide.
I would want that calendar inside the conversation. I could move a post, change the offer, or point to Thursday and ask why that topic belongs there.
The AI could generate two Reel openings and recommend one. If I don't understand the difference, it could show both versions in a short demonstration. I could pause at a frame and ask about that moment, or use the version it recommends and keep working.
All of this could happen through chat. The conversation would have a calendar and unfinished work in it, with room to inspect and change them. I wouldn't want every answer to become another paragraph I have to translate into something usable.
Here's my Instagram and store. Orders are slow. Rewrite the pages and prepare my next posts.
Read Instagram profile
Handmade ceramics. A little of my process.
This describes the studio, but never says you take orders.
Read store website
Made to order. Choose your glaze.
The offer is clearer here, but there's no production time before checkout.
These gaps may leave buyers unsure. They don't explain slow sales on their own. How long does a mug take, and what do buyers ask?
Three weeks before shipping. Most DMs ask whether it will arrive in time for a gift.
I've drafted the page edits and next posts below. I included production time so gift buyers can plan ahead.
- Bio draft
- Handmade mugs, made to order. Three weeks to make yours. Choose a glaze below.
- Store draft
- Ships in 3 weeks. Delivery time is additional. Buying a gift? Check delivery before ordering.
Your next posts
Week 1- MonHow ordering worksPin the steps and production time
- ThuA mug worth planning forReel · Answer the gift question
Thursday's idea becomes a Reel opening.
A gift, made for them
Thursday · Opening shotThree weeks to make. Their glaze to choose.
Next shots: glaze choices, then production + delivery timing.
Make the Reel opening shorter. Can we leave the wait out?
Try: "Your glaze, handmade. Ships in 3 weeks." I kept the timing because your gift buyers need it to decide. They can see whether it fits before starting an order.
The owner gets usable drafts and can ask why the AI made those choices.
The length of each explanation could depend on what I need. Sometimes a sentence would be enough. Sometimes I would want a 45-second video, then another example, then a longer explanation because I still don't understand. A small question could turn into an hour of learning without requiring me to commit to that hour in advance.
A 2025 study of ReelsEd, involving 62 university students studying programming-related subjects, found better immediate quiz results with AI-selected short clips than with longer videos. It did not measure long-term retention or test newly generated explanations.
I wouldn't turn every explanation into a Reel. An editable storyboard might help me compare ideas more easily than another video.
Even if a general assistant can produce a calendar, I might still want to work through the problem with someone I trust.
I may have followed her for years, recognize the kinds of businesses she understands, or prefer how she explains a choice. Her name gives me a reason to start. But I would want to know what she contributed to the advice I receive.
Uploading a course gives an AI material to draw from. Shaping how it diagnoses an account asks more of the expert. She would need to explain what she looks for, when a common tactic is a bad fit, and which questions she asks before recommending anything. By reviewing its advice, she could catch mistakes in how the AI applies her method or discover where the method itself needs changing.
Her recognizable voice wouldn't tell me whether she shaped those decisions. Her involvement also wouldn't mean she had personally reviewed my account.
MasterClass On Call offers AI coaching built with experts such as Chris Voss and Mark Cuban. In Brazil, Nizan Guanaes's NIZAI aims to make his creative thinking accessible. I'd want to bring them an unfamiliar problem and see how the expert's approach shapes the advice.
When I think about Hubla's future, I imagine an expert selling an experience that does useful work for people while helping them understand her approach. Someone could return to the same project, with the expert's material and their previous decisions available in the conversation. The AI could prepare the next version, explain a change, and help them use it.
Some days I would accept the calendar and move on. Other days I would want to understand why the AI changed the offer. Both should fit in the same experience. I don't need a lesson before I can get a useful result.
Anthropic's study of university students using Claude found both collaborative work and requests for direct solutions or finished outputs. The researchers analyzed conversations without measuring whether students became more capable afterward.
In a randomized study of high-school mathematics in Turkey, students using a relatively unrestricted GPT interface performed better during assisted practice but worse than the control group when the assistance was removed. A version designed with teaching safeguards largely mitigated that harm. That study tested school mathematics, including performance without AI. It gives me a reason to check what people learn alongside the work we help them finish.
When I add bowls to the store, I want to bring them into the same conversation. We already worked through how I sell and what customers ask. The AI could use that context to draft the product page and first posts. I could ask why it changed the approach for this product while we get the launch ready.