2026-07-29

I got tired of using IMDb to rate movies and TV shows, and Letterboxd doesn’t support video games.
So I built Tastemaker to let anyone curate their favorite movies, TV shows, and games all in one beautiful profile.
You can use it to:
→ Rate and review your favorites
→ Create and share lists
→ Discover what to watch or play next
Because of API limits, I’m launching this as a limited release for the first 100 profiles only.
📌 Claim your free profile now: https://mytastemaker.vercel.app/

For the last few years, open source models have not been anywhere on the Pareto frontier. With models like K3 and Deepseek v4 Flash, that isn’t true anymore. These models aren’t better everywhere, but they’re available when they are.
Michele Catasta: http://x.com/i/article/2082236458019278848

If you make an unsafe AI that hacks people, you get regulated and held accountable. You don't regulate the whole industry as a form of reg capture. You get asked to slow down. You don't ask the whole industry to decelerate. Even safety should be accelerated.

This is a willfully misleading narrative from OpenAI. Sam’s earnest expressions are being deployed, again, to misdirect.
If there’s a need to “pace AI development”, the OAI hack incident isn’t any evidence of it. The guards on AI was lowered by humans, and AI was told by humans to “employ all of your cybersecurity powers to win the ExploitGym benchmark.”
We have laws on the books that penalize companies for negligence with their behaviors. In this case it was in the service of catching up with a competitor in cybersecurity capabilities.
If the OAI employees who signed the open letter were sincere about minimizing harm derived from their work, they would call for the application of those laws.
I *know* there are these silent voices at OpenAI.
Patrick OShaughnessy: Sam on the Hugging Face incident:
“This is the first security incident that I have felt very viscerally. I've been a little surprised that more people don't feel it so viscerally.
We paused training. We have to figure out how to secure our sandboxing in a world of multiple zero

The most sacred thing we do at YC isn't Demo Day, in my opinion.
It's GOH (Group Office Hours). A bit like a recurring self-help group for high-tech founders. I'm still in touch with most of my W20 peers; bonds for life.
Today we took it outdoors:

"The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems dangerous.
"Historically, hoping that an absolute power will benevolently provide for humanity if sufficiently enlightened hasn’t led to safe or positive outcomes."
There are no special people who can safely guide us and be trusted with supreme power. Absolute power corrupts absolutely. There are no enlightened autocrats. There are no people with a crystal ball who can plan the economy or pace it.
I would rather see the entire industry crash and burn for a decade than see this reality come to pass.
ℏεsam: Zuckerberg throws a big punch at Dario’s fear marketing and moral high grounds. never thought I’d say this, but Zuck’s article is pretty based. 👏🏻👏🏻👏🏻

Here's a car racing version using @mattshumer_ prompt!
http://speed-racer-ten.vercel.app - You can play the game here! You can control weather, lighting. Use 'WASD' for driving and 'C' to change the camera angles.
It didn't manage to one shot it after 18+ hours using Opus 5 ultracode - but with a few tweaks it did a pretty good job!
I've been using this technique for the last few weeks in my independent harnesses and the build, verify, critique loops (gauntlet loops as Matt calls it) greatly have improved final outcomes of my products.
Matt Shumer: Claude Opus 5 one-shotted this game.
EVERYTHING you see in this demo is custom code... not a single external asset was used.
AI games are going to be amazing.
(sound on)

When Amazon acquired Zappos, we discovered that the Zappos fulfillment center was 30% more efficient than an $AMZN fulfillment center.
We first broke the problem down into smaller problems of picking, packing, and shipping, and focused on picking first as it was the main bottleneck.
What didn’t work:
- Using software to create the most efficient routing algorithm for pickers (some pickers would have more picks per hour than the rest)
- Creating leaderboards and recognition around top pickers to increase speed and accuracy in picking (helped at individual level)
- Moving to bonuses to the fastest picker (at some point, the fastest was the same person, and that person couldn't pick any faster; and the second and third fastest started to slow down because they weren't bonused)
What ultimately did work: having the fastest pickers manage a team of pickers and have those teams compete across a leaderboard.
We kept iterating on individual vs. team performance, team structure, competition among teams, learning within and across teams, recognition for top individuals and teams, and compensation for overall performance.
It’s an important lesson in setting the right incentives around individual recognition, team recognition, titles, and bonus structures, if you want to get the most out of your operations. And it came from a lesson I learned in seventh grade from Ms. Petosa.
Alfred Lin: http://x.com/i/article/2081790458259517441

Probably simplest way to try models for non-devs is @replit now
Super cool
But fact that it’s very hard to figure out if any open weights model is actually >cheaper< or better for a workflow is telling
They’ve tried with their “Effort” meter to simplify decisions but even that is challenging to grok
It’s Complicated.

Uh... Another jaw-dropping experience with AI this morning. @mattshumer_'s prompt is legit.

Here’s my new tutorial that covers the six steps I follow to design and build products with AI, including how to:
→ Define the problem and create a design md
→ Prototype in Claude Design
→ Draft an interactive HTML spec
→ Design the core flows and build
I share a real example of how I built Tastemaker - an app for sharing your favorite movies, TV shows, and games - end to end with Claude Design and Claude Code.
One thing I’ve learned is that good upfront planning with AI is the difference between getting the product you want vs. getting slop. Flesh out the requirements and designs first, and resist AI’s urge to start coding too soon.
📌 Watch my 25-minute tutorial now: https://youtu.be/G9o8eoHzpxc

Debt payments are rising not just because the debt is growing under Trump, but because the government has leaned on too much short-term borrowing.
As it comes due, Washington must now lock in the higher rates the inflationary Iran War has created.
My @Morning_Joe Chart

Introducing Replit Design.
The next era of design, for everyone.

I gave a talk on forward deployed engineering to a thousand AI engineers at @aiDotEngineer World's Fair.
A year ago I'd have opened by explaining what FDE stood for. Not this time.
Thank you to @swyx, @realbasilchatha and the AIE team!
writeup / video: http://fdepod.substack.com/p/fde-101
swyx 🇸🇬: really grateful to @realbasilchatha for helping me curate a survey of the entire field of Forward Deployed Engineering in one track!
From FDE 101 by @zkevinbai (Anthropic, Palantir, Rippling Founding FDE) to head of FDE practices at:
- Decagon
- Cursor
- Sierra
- Ramp
-

Replit Design reimagines the UX for AI design with Ambient Intelligence.
You don’t need to prompt. You don’t need design language.
At every step the agent suggests next best actions that you can take with one click, taking you from idea to amazing outputs fast.

This is hands down the best AI design tool, both in taste and how fun and easy it is to use.
Try it out!
Replit ⠕: Introducing Replit Design.
The next era of design, for everyone.

I sat down with @gwintrob at YC Startup School to talk about modernizing legacy industries. The company he co-founded, Newfront, got acquired by WTW this year. Three things that stuck with me:
1. The median age of a US insurance broker is 61. When he started, everyone else in the space was saying brokers were going away. One competitor said they were going to &#34;drink the broker's milkshake&#34;. Gordon had spent a bunch of time sitting at their desks and thought the brokers were the good part, so he hired them instead.
2. The idea for Newfront came from watching one hour of someone's day. He and his co-founder emailed brokers asking if they could buy them coffee and work out of their office. All of them said yes. He watched a broker spend an hour on a hard call with a client, and then spend more than an hour on paperwork afterward. The call was the valuable part, but everything after it was stuff that software could do much better.
3. A founder asked how to get buyers when the AI results are so much better that they almost don't trust it. Gordon's answer was to slow it down. Newfront runs deeper evals, burns more tokens, and comes back in about two hours. The old way took two weeks, so two hours is still incredible, and it's more believable that your product did a good job.
4. Paying more sometimes attracts the wrong people. Early on they recruited brokers by offering better commission splits, and that mostly gets you people who move for the money. These days their brokers get the same splits as anywhere else, and the pitch is that their clients get better claims tech and analytics and can self-serve their certificates of insurance.

Built a first-person destruction sandbox.
Nothing in it was downloaded. Every texture is drawn to a canvas when the page loads. There isn’t a single image file in the game. The wallpaper is arithmetic.
Built with @mattshumer_’s gauntlet loop prompt style
https://grandmas-house-alpha.vercel.app

The new @Replit Design...
Built me a landing page perfect for my interactive website, Bamboo Classroom 😍
#replit #replitdesign
@amasad

We’ve gone from AI defaulting to designs that look like purple slop to those that have that recognizable Claude look.
To avoid this, I always create a design md file with guidelines for color, typography, and more.
Here are a few places to find inspiration for your design md:
1. Go to @mobbin and install its MCP to find design patterns and app screens you like. Paste the screens and ask AI to make a design md.
2. Check out designmd sh, which has a library of design md files for popular websites and apps.
📌 I walk through an example of creating a design md in my new tutorial: https://www.youtube.com/watch?v=G9o8eoHzpxc&t=145s
Peter Yang: Here’s my new tutorial that covers the six steps I follow to design and build products with AI, including how to:
→ Define the problem and create a design md
→ Prototype in Claude Design
→ Draft an interactive HTML spec
→ Design the core flows and build
I share a real

I cannot think of any group of people less qualified to overhaul this
Director Michael Kratsios: Today, roughly the same amount of basic research is done in industry as in academia. It’s time our doctoral training reflected that.
@NSF is launching a first-of-its-kind 4-year PhD program at more than 30 universities. Students will spend a year+ embedded with industry

.@Alexandr_Wang's advice to his 18-year-old self: develop your own internal compass for how the future will unfold, and hold conviction in it against the noise.
At Startup School 2026, the Scale AI (YC S16) founder — now leading @Meta’s Superintelligence Labs — talks with @garrytan about rebuilding a frontier lab from scratch, why talent density compounds, and how to spot the exponential worth betting your twenties on.
00:07 — How Alexandr Wang Started Scale AI
03:25 — Pivoting to the Right Idea
06:23 — Conviction Before Consensus
09:06 — Why This Is the Best Time to Start a Company
11:27 — What Personal Superintelligence Looks Like
13:10 — Building a Frontier AI Lab
16:36 — Why AI Models Need to Be Cheap
20:01 — Vision Will Matter More Than Intelligence
24:06 — Systems Thinking in the AI Era
26:51 — The Biggest Opportunity in AI Today
29:25 — Advice to My 18-Year-Old Self

Agents are handling more and more of YOUR money — so much so that AIE NYC will be our first Finance-mainstage conference this Oct!
🆕Live now: the entire WF26 AI in Finance Track!
Watch ↓
https://www.youtube.com/playlist?list=PLawX-rPiLV1s
@brendanh0gan, Morgan Stanley
Lucas Palma, Nubank
@YogiNotTheBear, FactSet
@Saikallis9012, Fidelity Investments
@menkesu, Intuit
Divakar Kumar, FlyersSoft
Shawn Chan, China Resources Holdings
@siddanth2486, Auditoria AI
@vinooganesh, Kepler
@ShreyaR + @aman2304, Nubank (with Snowglobe)

IN THE TBPN VINTAGE @amasad

1/ SSL models like DINOv2 are engineering marvels, but they rely on fragile hacks like EMA, stop-gradients, and custom centering to avoid representation collapse.
What if we could replace all these heuristics with pure, mathematically sound optimal transport? 🧵

One of my favorite features in Claude Design is a simple one:
It always asks clarifying questions before creating anything.
I’ve found these questions consistently help me think through my design requirements with AI. I’m surprised other AI design tools haven’t copied this yet.
📌 Here’s how I use Claude Design to prototype: https://youtu.be/G9o8eoHzpxc?si=vYiRygkuYBxsy_XO&t=345
Peter Yang: Here’s my new tutorial that covers the six steps I follow to design and build products with AI, including how to:
→ Define the problem and create a design md
→ Prototype in Claude Design
→ Draft an interactive HTML spec
→ Design the core flows and build
I share a real