Kumail Hunaid

Kumail Hunaid

@kumailht.com

http://kumailht.com

622 followers45 following502 posts

I've been building an app that helps you test out several headlines The complex part is buidling: - A SDK that tracks views & conversions while removing bots - Storing a crazy amount of data efficiently - Crunch the data without breaking down - Doing the math correctly

View full thread

Then I'll break the build down into phases, which will help the AI build the project in layers And after every layer, I'll have to do a full code review to check the implementation

All of this is a lot faster than if I had done it myself from scratch pre AI. It has taken months of work and turned it into days or weeks of work. This is going to change the bar for our pace of software development. I'm not sure if it allows us to make better software though.

I've been building an app that helps you test out several headlines The complex part is buidling: - A SDK that tracks views & conversions while removing bots - Storing a crazy amount of data efficiently - Crunch the data without breaking down - Doing the math correctly

And you'd think AI can just build this out, and that is partially true But it's also really good at making terrible architecture decisions that just end up being a sphagetti of code In the end, if I just "vibe coded" this, it'll only "kinda" work, and break in interesting ways

The way I'm solving for this is to plan out the build in incredible detail, so I can review every architectural decision upfront Then I can sit and think about it, ask questions about how it'll break, improve it, and hopefully reach something robust

In the AI era, copywriting will become waaay more important LLMs have proven that writing has personality, and when all writing has the same personality, it sounds inauthentic Professional copywriters, people who have their way with words, will become increasingly important

When I see your AI generated image, all I see is slop. Even when it is very good. It's the same effect as the Stock Photos of the 2000's.

In 2024 when I built a booking process with a map, I didn't realize just how complex it was going to be Having to deal with overlapping bookings, time zones, multiple sessions, booking seat changes, etc Airline seat booking systems must be one hell of a beast

The Chapel Coffee Shop in Romania I love when design turns almost nothing into something Like this gap between two buildings has now become iconic

I love how design is almost free like you can spend extra brain cycles and using the same cost of goods end up with a better product

Before AI, everything was 0's and 1's The limitation is that all logic has to be quantitative With no room for ambiguity Now, every layer allows for some ambiguity and unstructured data This changes all software from the ground up Everything must be re-thought Opportunity.

When I was 19yr old, I started my first job The adults around me were working to pay the bills I was working because I just loved the work I could run circles around them As I get older, I have to remind myself to love the work Otherwise I lose to the 19yr old

I bet you didn't know that Amazon controls on the order of ~190M IPv4 addresses, an inventory worth an estimated $7–8 billion, and reportedly earns on the order of $1 billion/year by renting them out at $43/yr each.

Finding your design style is like a band finding their sound. Some artists spend a decade before they hit the thing that's unmistakably theirs. Design's no different. Designers need to find what resonates with the times A distinct style they can call their own

For 10+ years you could buy a $25 html template off Themeforest/etc and make a professional looking website quickly But people didn't do it because it implied the message wasn't worth the paper (website) it was written on So businesses kept hiring designers This hasn't changed

AI can ship a mediocre UI in seconds, but the gap between "competent generic SaaS dashboard" and "memorable distinctive product design" is exactly the gap that doesn't yield to better tooling.

This is why I'd predict the top rungs of design's ladder may never fully reach "trivial" the way software's does — there will always be a premium on the designers who can answer "what should this even be?" rather than just "how do I render it?"

Every ERA has its design STYLE If you want your designs to look FRESH and not dated It cannot look like anything from the previous era This is what people mean by "taste" It means stealing from designers who've done something fresh But first, you must IDENTIFY what's fresht

The best interface is no interface at all Especially now, when it's cheap to make complex UI The apps that will win will have almost no UI at all

My process as a designer has changed in the last few months In Janurary, I'd start my design with notebook and a pencil Now, I start my design with Claude I generate wireframes, and use them to visually think Then move to Figma to find the right direction