Pinned Tweet
Software is buggy largely b/c programmers jump straight into writing code. Few write even a sketch of what their programs should do before coding.
Code draws attention to individual pieces instead of the bigger picture of how your program fits together.
theatlantic.com/technology/a…
Every company should have a Brain — Garry Tan, Y Combinator youtu.be/eBUyTS7SzV4?si=sAgc… via @YouTube
Wyn Van Devanter retweeted
Scientists put kids through 100 hours of reading, then scanned their brains. New wiring had physically grown inside the language regions. Communication between brain areas sped up by a factor of 10. Kids who didn't read showed zero change.
That was a 2009 Carnegie Mellon study. It gets wilder.
In 2013, Emory University scanned 19 students every morning for 19 straight days while they read one novel chapter each night. Mornings after reading, the brain areas responsible for understanding other people's emotions lit up with new connections. So did the region that processes physical sensation. Their brains were simulating what the characters felt, as if it were happening to them. Those changes stuck around for 5 days after they finished the book.
Now flip to scrolling. A massive review published in Psychological Bulletin last September pulled together 71 studies covering 98,299 people. Heavy short-form video use (TikTok, Reels, Shorts) showed a clear pattern: worse attention, weaker self-control, and more anxiety. Consistent across teenagers and adults, across every platform tested. Oxford didn't name "brain rot" its 2024 Word of the Year for nothing.
A 2024 brain wave study found that people hooked on short-form video had weaker activity in the front of the brain, the part that controls focus and impulse control. Separate brain scans showed the same thing: heavy scrollers had less activation in the exact regions that deep reading strengthens.
UCLA neuroscientist Maryanne Wolf has been studying this for decades. Humans were never born to read. There's no gene for it. Reading is something we invented, and it hijacked neurons that were originally meant for recognizing faces. Over time, it built entirely new brain circuits connecting language, vision, and emotion. But those circuits only survive if you use them. Stop reading, and they fade. Wolf's conclusion is simple: screens built for speed produce a speed-wired brain. Books built for depth produce a depth-wired brain.
One honest caveat: most of these studies are snapshots, not long-term tracking. People who already struggle to focus might just prefer short videos. But the same pattern showing up across nearly 100,000 people is hard to shrug off.
The tweet repeats the line seven times. The research backs it up with brain scans, EEG data, and white-matter imaging across tens of thousands of people.
Wyn Van Devanter retweeted
Junior and mid-level engineers can no longer push AI-assisted code without a senior signing off at AWS
Amazon is holding a mandatory meeting about AI breaking its systems. The official framing is "part of normal business." The briefing note describes a trend of incidents with "high blast radius" caused by "Gen-AI assisted changes" for which "best practices and safeguards are not yet fully established." Translation to human language: we gave AI to engineers and things keep breaking?
The response for now? Junior and mid-level engineers can no longer push AI-assisted code without a senior signing off. AWS spent 13 hours recovering after its own AI coding tool, asked to make some changes, decided instead to delete and recreate the environment (the software equivalent of fixing a leaky tap by knocking down the wall). Amazon called that an "extremely limited event" (the affected tool served customers in mainland China).
Wyn Van Devanter retweeted
Dario Amodei just revealed that the AI training bottleneck everyone is worried about doesn’t exist anymore.
The industry spent years obsessed with scraping the open web. More data. More text. More human output to feed the models.
Amodei: “I don’t think data is quite the most central thing anymore.”
The shift is fundamental.
Amodei: “Static data is becoming less important. A lot of the data we use today is RL environments that we train on. Dynamic data that the model creates itself.”
Not scraped. Not licensed. Not written by humans.
Generated by the model through pure trial and error.
When you train on complex math or agentic coding, you don’t feed it a textbook. You give it an environment.
The model experiments. Fails. Adjusts. Tries again.
Amodei: “You’re getting some math problems and the model experiments with trying the math problems.”
It generates its own experience. Millions of iterations. Each one building on the last.
No human required.
This destroys the entire narrative around AI hitting a data wall.
You cannot throttle a competitor by locking down copyright. Cannot slow the race by putting up a paywall.
When a model learns through its own synthetic experience, the open web becomes irrelevant.
The only true bottleneck left is compute.
And this is where the geopolitical stakes become impossible to overstate.
The nation that wins the compute race doesn’t just build smarter models.
It builds models that generate their own intelligence, compounding on themselves, iterating past every limit human knowledge ever imposed.
We are no longer training AI on the past.
We are letting it simulate the future.
The machine has stopped reading the dictionary.
It’s doing the math itself now.
Wyn Van Devanter retweeted
Claude is releasing startups as plugins.
48 hours ago @claudeai released the Wealth Management plugin and of course I had to dig in and check it out. I've created a thread on how to use it and what are the inputs and outputs. I'm sharing all the PDFs involved.
[1/12]
Wyn Van Devanter retweeted
Sequoia just called the end of an entire go-to-market era and most SaaS companies won’t realize what hit them for 18 months.
Product-led growth was built on one assumption: humans would try the software. The entire playbook since 2010 optimized for human discovery. Beautiful landing pages. Frictionless free trials. Viral invite loops. Slack, Dropbox, Zoom, Calendly. $200B+ in market cap created by winning the user’s first 5 minutes.
None of that matters if an agent is picking the software.
Claude doesn’t care about your hero image. It can’t be impressed by your Dribbble awards. It’s reading documentation, parsing user reviews, checking API reliability, and matching features to use case. All the surface-level polish that convinced lazy humans to click “sign up” becomes irrelevant.
The new PLG funnel isn’t landing page → free trial → activation → conversion.
It’s agent query → documentation scan → feature match → recommendation.
Which means the new moat looks completely different. You don’t need the best onboarding. You need the best documentation. You don’t need viral loops. You need structured data that agents can parse. You don’t need a beautiful UI for the first session. You need an API that an agent can actually call.
The companies that won PLG hired designers and growth hackers. The companies that win agent-led growth will hire technical writers and developer relations engineers.
And here’s the part nobody’s pricing in yet: agents don’t have loyalty. They don’t have switching costs. They’ll recommend Supabase today and something better tomorrow if the documentation is cleaner or the pricing is more transparent. The stickiness that made PLG so powerful, the network effects and learned behavior, doesn’t transfer.
Sequoia is telling you the entire distribution layer is being rewritten. The question is whether your product is optimized for human attention or machine parsing. Most are built for the wrong audience.
Sequoia partner @sonyatweetybird says we're going from the age of product-led growth to the age of agent-led growth.
"You see this most clearly if you're using Claude Code actively. It says, 'Hey, for a database, you should use Supabase. For hosting, use Vercel.' It's choosing for you, the stuff you should be using."
"Product-led growth brought us closer to the vision of 'best product wins,' but ultimately people are still lazy. They can't read all the reviews, and they kind of default to what looks cool on the website."
"Whereas your agent has infinite time to go and make these choices for you. It can go and read all the documentation, read all the user comments, and figure out [what you need] for your use case."
Wyn Van Devanter retweeted
Great overview of where product management is headed
- Problem shaping / clarity
- Context curation
- Evals / taste
Wyn Van Devanter retweeted
Harnessing the Power of Generative AI using RAG by Matt Ratliff nitter.gq/i/broadcasts/1lDGLlDrr…
I don’t tweet much. So it’s pretty meaningful to me when I do: a tragedy I never thought would happen. Apple’s iPhone/Mac sync has been failing me in horrendous ways, causing me to lose messages, reminders and notes. I must take drastic action.
Wyn Van Devanter retweeted
I always enjoy reading Jim’s books. Many know 'Terraform Up & Running,' but my favorite is 'Hello Startup.' With his new book he completes the circle: building your startup, deploying your infrastructure (IaC), and delivering your product (run & maintain software) 🚀
📣 Introducing my new book: Fundamentals of DevOps and Software Delivery. 🎉 It's a comprehensive, hands-on guide to deploying and managing software in production, written for practitioners. The first half is available now, for free, on the Gruntwork blog: gruntwork.io/fundamentals-of…
Just the beginning 🙄
OpenAI has released its first report, which details how bad actors in Russia, China, and beyond are using AI to spread propaganda. (Poorly.) wired.trib.al/Td0oUip
The old idea in coding that well-written code documents itself and to avoid comments is flawed. It's like saying, anyone can tell your important motivations for any action you do.