←
Vincent van der Meulen4,612 posts
Vincent van der Meulen avatar

Vincent van der Meulen

@vinvan

co-founder @mainframe. prev: @figma @diagram @facebook.

◷ Joined August 2013▧ 498 Photos and videos
Vincent van der Meulen avatar
Vincent van der Meulen@vinvan·10h

there’s a lot of talk about the end of software, generative ui, etc. but not many in-depth conversations about what that future might look like! for users, but also for us designers. it’d be cool if someone did a mini series on the topic. and visits a bunch of people at the frontier to try and figure it out :) e.g. talk to Jason Yuan and Eugenia Kuyda about how an OS might generate everything on the fly. sit down with Evan Conrad and Chris Paik and ask them what made them write their respective “End of Software” essays, and what they’re predicting next. visit the Runway office and imagine what would happen if all ui becomes ai video. how does one tame such a ui? talk to Nils Edison about “ui as content” and what a Spotify for software might look like. and, if possible, visit Bret Victor at Dynamicland to see what he thinks. whether he has a pov on where it’s all headed. while we can’t know for sure, i think there’s tremendous value in speculative futures. especially since they give us a sense of what future we’d actually want. if someone makes this series, i’d love to watch/listen! (attached some questions i’m personally curious about!)

Post attachmentPost attachmentPost attachmentPost attachment
Vincent van der Meulen avatar
Vincent van der Meulen@vinvan·Nov 14

i have ai psychosis and completely changed my schedule this year to maximize agent output. what I’ve learned:
 1. if you have N working hours, it’s now beneficial to spread those out as much as possible instead of work continuous blocks. aka work more days and start earlier/end later, but with more breaks. models can work autonomously for days but still need your direction, and you need to sometimes unblock them. frequent touch points are important for that. e.g. much better to find out at 7am that your models got stuck, and get two hours of progress before your work day, than find out at 9am. similarly, a quick check-in on sunday morning sets up your models for a day of work. this doesn’t have to mean actually working more hours if you don’t want to. a five minute chat from your phone is enough. 2. there’s now an entire second workday that can happen while you’re asleep. but it only happens if you make sure your models can work overnight. it’s crucial you spend the end of your workday setting up your overnight agents for success. my lazy life version looks like this: every night at ~9pm i go on a run, dictate what i think should happen while i'm asleep, and dump the transcript into grok bot. it takes 10 minutes and results in eight hours of work. 3. prioritize short, high intensity work sessions over long hours. when you’re in front of your computer, all your time should go to unblocking your agents with *high quality* ideas and decisions, e.g. doing Matt Pocock style grill me sessions. alternating a few high quality planning sessions with touching grass is more valuable than spending twelve continuous hours babysitting agent traces. ofc it’s beneficial if you’re able to combine intensity with long hours, but that’s harder. it reminds me of how athletes train: workout heavy multiple times a day and focus on recovery in between. 4. do whatever you need to maximize your capacity for idea generation and decision making. don’t make stupid health and sleep sacrifices in pursuit of productivity. you’re no longer getting paid for brute forcing solutions. that’s what agents are for. it’s much better to get good sleep than 996 if the latter makes you even a tiny bit less focused or creative the next day.

Jordan Singer avatar
⟳ Vincent van der Meulen retweeted
Jordan Singer@jsngr·Nov 13

collaborative product building in mainframe.app

0:34
Vincent van der Meulen avatar
Vincent van der Meulen@vinvan·Nov 12

i just saw the sickest new ui pattern in a long time 🤯 i haven't seen anything so obvious but interesting since conductor's worktree switcher. @atlou_ cooked. next week is going to be incredible for @mainframe.

Jordan Singer avatar
⟳ Vincent van der Meulen retweeted
Jordan Singer@jsngr·Nov 10

we design and build @mainframe inside of Mainframe here's our full frontend and backend running on our canvas

0:14
Vincent van der Meulen avatar
Vincent van der Meulen@vinvan·Nov 10

i now get to spend my days talking to my childhood design heroes and call it 'customer research' 18 y/o me would be very jealous!

Yuki avatar
⟳ Vincent van der Meulen retweeted
Yuki@yukiasa_·Nov 10

Been using this every day for the last 3 weeks. It just makes so much sense 🙂‍↕️

we’re opening the waitlist for @mainframe today. it’s a new design tool for designers who ship.

Post attachment
Jordan Singer avatar
⟳ Vincent van der Meulen retweeted
Jordan Singer@jsngr·Nov 8

i think about this a lot you can slop clone anything, but you can only copy a point in time, not the seed of original ideas or the dedicated pursuit

I remember @tayroga once told me “they can copy the point but not the vector” and I still think about it

Vincent van der Meulen avatar
Vincent van der Meulen@vinvan·Nov 7

we’re entering an era where every product insight will get copied the moment it’s public, by agents that are working 24/7. what i think that means for the next months (wip): 1. spend literally all your time talking to customers 2. do whatever you need to do to perfect your moat: human decision making go on walks, jam with friends, take long showers, and spend time with out of distribution (old?) ideas, take care of yourself. your ideas and judgment as to how the pieces connect still make the difference. 3. maximize compute so you can ship faster