What 3 makers on MakerMap learned about it, in their own words: their best posts on X, filed under Building with AI.
There are currently so many AI hackathons running that I'd stop entering generic ones.
I'd pick competitions where the constraint forces you to learn something.
Examples open right now
> OpenCV + AWS → computer vision
> Nebius + NVIDIA → AI infrastructure
> PayPal → payments + AI
> Qloo → recommendation/taste agents
> Amazon → device experiences
A constraint is useful.
"Build anything with AI" usually produces 500 versions of the same app.
Link in reply.
i’ve been thinking a lot more about what it actually takes to build a defensible business in the age of AI
i think we’re moving from selling software to selling outcomes
and there are 3 things that will make these businesses incredibly difficult to replicate:
1. data
not just owning unique data, but analyzing information at a scale that wouldn’t make economic sense for an LLM to do on demand
imagine continuously analyzing hundreds of thousands of YouTube videos and correlating emerging trends with actual buying signals
an AI could theoretically do this for an individual user, but repeating that entire analysis for every request would be incredibly inefficient
the opportunity is in doing the expensive work once, continuously improving it, and making the resulting intelligence accessible to everyone
2. community
AI can generate intelligence on demand, but it can’t generate a network of real people on demand
think experts, mechanics, suppliers, professionals, or people with shared interests
building a community around your service gives you access to relationships, expertise, trust, and real-world capabilities that an LLM simply cannot reproduce
3. non-standard integrations
not just connecting a few APIs
i’m talking about the messy workflows that actually make things happen
legacy systems, WhatsApp conversations, government portals, purchasing processes, human approvals, and physical services
now imagine combining all three
let’s say you build a service for car wash operators
your data continuously analyzes equipment failures across thousands of locations, correlating them with usage, weather, and maintenance history
you predict a component is about to fail
your community of local mechanics allows you to immediately find someone qualified to replace it
your integrations coordinate the mechanic over WhatsApp, order the replacement component, and handle the entire workflow
the customer doesn’t need another dashboard
they just want their car wash to keep running
and here’s where it gets interesting
every repair generates more data
every completed job strengthens your network of mechanics
every integration makes the next intervention easier
the whole system compounds, becoming increasingly difficult for a generic AI agent or competitor to replicate
that’s what i think makers should be building toward
go hyper niche, combine data + community + integrations, and deliver the entire outcome
the future isn’t just selling software to businesses
it’s executing parts of their business for them
what happened: 12 ai agents read 64 physics papers and built a website where anyone can rerun the experiments, like a textbook that lets you poke the results. that's a big win for learning. the uncomfortable bit: if building the demo costs nothing, the scarce skill becomes knowing which of those 64 papers was wrong in the first place