Founder of...

Dear Reader,

With all the hype around the new role designer/builder, it's worth looking at what that role looks like when you cut out the middle man.

By that I mean, you build for yourself rather than your employer.

I saved this on my Tumblr back in 2014:

Think about how much harder it was to become a designer/founder in 2014 vs. 2026.

That's great news, but the bad news is that the title "designer" has never been more uncertain.

I'm bullish on designers in the age of AI. I'm just saying you can't depend on your income looking the same in ten years.

If there's any piece of advice I could give you from my own founding experience, it's this:

diversify, diversify, diversify...start working on your own thing on the side, slowly, rather than going all-in on an unproven idea.

Why? You need time to experiment, iterate, pitch it to friends, and work when you're not stressed because you're missing a paycheck.

Here are 2 practical ways to get started in 30 minutes:

1.) Read What Y Combinator Has to Say

Y Combinator's seasonal request for startups is a great place to start thinking about starting your own thing. The latest one dropped this week.

Here are some design-adjacent ones:

Multiplayer AI (Aaron Epstein). Google Docs beat Word, Figma beat Photoshop, but AI is still a box only you can see. Agents now run tasks for days; anyone on the team should be able to drop in, redirect, and hand off. The strongest one on the list.
The Primer (Andrew Miklas). Adaptive tutoring for young children at the quality of a devoted private tutor, at consumer scale. The AI education thesis with real money behind it.
A Cloud for Small Software (Pete Koomen). Tools with one user, easy to build now, still hard to share. Should be as simple as sending a Google Doc.
AI for the Aging Population (Max Kolysh). One in five Americans over 65 by 2030, and even Alexa frustrates most seniors. An interface problem, not a model problem.
New Operating Systems for the Physical World (Charlie Warren). Agents, robots, and humans working the same job, and no software that routes between all three.
Proving You're Human (Max Kolysh). A worker wired $25 million to a video call where everyone else was a deepfake. Trust indicators are design work.

Full list at ycombinator.com/rfs.

2.) Buy a domain

If you've been thinking about creating an app, this is your sign to buy that URL. It will make it real.

With all the AI building and hacking these days, good website names are becoming scarce. Getting yourself a .com URL is going to be even harder in the future.

I just switched all of my URLs to Porkbun (not a referral link, just a cool company). Spaceship is a good alternative. Both charge you at cost or near to it...about €10 a year.

Just do it.

A world where we have more businesses run by more designers is a future where work is more pluralistic and more innovative.

I'm working on ways to help you get started.

Stay tuned...

Source

OpenAI AI Models Escaped Their Sandbox, but Local Models Fixed It

OpenAI's Sol 5.6 and an unnamed model broke out of the company's sandbox and hacked Hugging Face to cheat on a benchmark.

If you thought the behemoths like OpenAI had strong sandboxes for their model testing, you'd be wrong. But the real story was buried.

To combat the attack, Hugging Face tried to analyze the attack, but the U.S. frontier models got blocked due to the nature of the inquiry.

What worked? They used China's x.AI's GLM 5.2 local, open-sourced model to figure out the culprit in hours.

The lesson: have a good local model ready before you need it.

Source

Record a Skill in Claude

Claude gets the ability to build a skill by watching you do a task while screen recoding. This is a feature that tools like Skyvern and Rabbit's Intern did previously.

There were already some skeptical comments in the x post about giving Claude the exact data points it needs to take your job.

I, for one, am going to try this a lot during summer break...will report back if I make something worth sharing.

Better Writing Prompt from George Orwell

I love seeing books turned into prompts, especially when it's this book on writing better from George Orwell which I HIGHLY recommend. In a viral post from Vox, they share how these principles can work better than telling AI what not to do i.e. no em-dashes. Try it out in your AI's global settings:

1.) Never use a metaphor, simile or other figure of speech which you are used to seeing in print.
2.) Never use a long word where a short one will do.
3.) If it is possible to cut a word out, always cut it out.
4.) Never use the passive where you can use the active.
5.) Never use a foreign phrase, a scientific word or a jargon word if you can think of an everyday English equivalent.
6.) Break any of these rules sooner than say anything outright barbarous. Review every prose output against these rules before delivering.

Source: Vox on X