"show me with an ascii diagram" is my favorite prompt of the day
turns walls of text into a much more easily digestible thing
doesn't require a fancy web-based custom harness and doesn't create artifacts / side effects
Lots of people are dunking on this but I think a subtlety is lost
Slop is bad in both, but humans are far better at recognizing (and therefore more likely to fix) frontend slop
So vibe coded front ends end up ok, but only because the human is better at frontend slop detection
Frontend devs know this — bugs get reported to the frontend devs first, they end up triaging everything
"reading fatigue" is real. models produce so much text that I spend most of my day reading, and very little time deciding or acting
I've tried things like "terse mode" or "caveman mode," but still find i'm fried from reading halfway through the day
I've started to notice that agents, even with the same boot docs, change personality by project
Backend agent: rigid, assertive, pedantic. iOS agent: creative but careless
Is this an emergent property of the training data? Or me projecting my own biases? Who can say...
MB retweeted
Waking up to agents that did work for me over the night is still something that amazes me
I’m sure at some point we’ll all take this for granted but let’s just reflect for a moment on how incredible this is
🙏🙏🙏
wtf do you tell young technical people to learn today
my hunch is learning to code will be valuable the way learning assembly was valuable after high level languages existed. helps a little but not required
CS still valuable, but curriculum needs to zoom out?
What are you using Fable for right now?
For literally everything I’m doing, Opus 5.5 is both better and cheaper
Opus also beats the OpenAI models in my limited experience
Vibe coding is used as a pejorative, and I get why, but it's just another name for something we've always done
We just used to call it "quick and dirty" or "prototyping" or "shitty mvp"
Agents operate that way if you prompt them that way, but they're more capable than that
MB retweeted
North Korea's hackers dumped $750M of stolen crypto on DEXs, but it cost them $107M.
MEV bots & Traders made over $50M. ETH hit $3,135 on-chain and AMM pools were temporarily emptied for their reserves.
So what exactly happened to the liquidity, and where did the $107M go?
MB retweeted
AI decompilation is not "clean-rooming"
a clean room works for one reason: the person writing the new code never saw the old code
the AI writing the new code is reading the old code--it's just reading bytecode or a binary instead of the source code. the binary itself *is* indeed the work (object code has been recognized as copyrightable since Apple v. Franklin)
"but the model never trained on the source" (even assuming that's true, which is largely unknowable unless you work at Anthropic) doesn't matter...
what the model does is translate object code into source, and "translation" is the first item on the §101 list of derivative works...in other words, copying object code/bytecode can itself be a copyright violation
of course, failure to use clean room procedures does not intrinsically mean that a copyright was definitely violated! even when copying from others' code, there is a whole analysis about which elements are "expressive" vs. dictated by efficiency, hardware compatibility, industry conventions, etc.--these are not protected by copyright...
but yeah stop saying that using AIs to decompile stuff is a "clean room"...ridic
Every C-Suite in the world is asking how they can get AI to recommend them over competitors. SEO (gaming the algorithm) no longer works, and you can't buy your wait into a frontier model
What does work? Actually being better than your competitors. AI isn’t fooled by your tricks
Reddit's earnings call has exposed the new reality for thousands of companies.
"AI mode does not drive traffic. Not for us and not to anybody as far as I know. It used to be a traffic driver. Now it's not." - Reddit CEO.
The reality is that AI is fundamentally changing the way that humans access the internet.
Just 24 hours prior to this statement, Google's stock erased -$480 billion on the same issue: AI is disrupting the traditional search model.
For decades, the internet operated on a simple model. Search engines directed users to websites, and those websites monetized the traffic.
AI is destroying this business model, and it's happening at an accelerating pace.
The traditional model of internet traffic is rapidly becoming outdated.
The internet has entered a new era.
MB retweeted
code review isn't going anywhere. if anything it will become THE most important piece of software in the modern sdlc.
2 big reasons why.
1. automerge (self-healing software) will soon be a reality.
most of the people that don't believe this either still read code diffs or used to think not reading diffs was preposterous. its coming.
2. software will be written at 10k tokens/sec, for virtually free.
writing code hasn't been the bottleneck since 2023 - its code quality. you can solve code quality bi-directionally. write good code to start with / review code once done.
both will happen because code quality is the bottleneck.
therefore, code review stays.
Models are changing so fast and getting so good, you have to build your harness and context system to be "deconstructable"
As with code, changes that remove cruft are often more valuable than changes that add things
Sometimes a tweet makes you say "welp, back to the drawing board"
No. We've got to stop applying old human-centric patterns to working in the new era.
delete all your wikis.
have an agent answer questions from source instead.
at spawn we have:
- zero dashboard tools
- zero docs
- we do not make or use summaries
everything we do is primary sources + intelligence
Nothing to go stale
Nothing to organize
Nothing to navigate
Always exactly up to date
Perfectly personalized
The answers get better with every model
If we want to see a chart, we ask for it, and we have it in a minute
If we want an answer to a question, we ask it, and we have a to-the-minute correct answer
MB retweeted
Mind-bending number. 1 in 7 Americans w/ $100m+ in assets could one day owe their fortunes to the AI cartels
Young, idealistic autisticrats who will reshape the startup scene, philanthropy, lobbying and culture
Fun to think about how they'll deploy their cash (likely into longevity, bio, space, defense, humanoids, de-extincting dinosaurs, etc.). But it's even more fun to imagine the cultural and political institutions they'll help fashion
@balajis's Network School @ns offers a tangible template here. They'll have the capital not just to dream up new institutions, but to experiment with entirely new ways of organizing society
