I'm really excited to share these upcoming episodes for my podcast soon.

One small ask: My birthday is on Thursday, July 9, and it would be amazing to finally hit 100K YouTube subscribers by then.

I’m at 98.5K now, so pretty close :)

📌 If you’re looking forward to these episodes, consider subscribing here: https://www.youtube.com/@PeterYangYT?sub_confirmation=1&utm_source=newsletter-header

I appreciate your support!

Damn they better have some armored vehicles to escort these English players out of the stadium 😅

What a game wow what a game

Hardly a day goes by without new levels of evil dropping from this one country.

Parody Jeff: THIS IS SATANIC ✡️🇮🇱

An Israeli soldier throws a stun grenade into a car carrying a Palestinian family, trapping them inside to absorb the blast.

Reports confirm one of the children inside is now permanently blinded for life.

"People assume prototyping requires writing code, but Image Gen is simply much faster."

Rohan (Codex PM) showed me exactly how he does it:

"I took a screenshot of the Codex composer where you can type in a new message.

Then I asked Codex to use Image Gen to prototype 4-5 ideas to improve project selection.

I can then ask it to take the first image and make a prototype of it on a Codex site.

We usually think of Image Gen as useful for things like turning my hair blue. But it can create amazing digital mocks too."

📌 Full episode: https://youtu.be/fAdFE7y6K2o

Tired of winning

Steve Rattner: Labor force participation has fallen to 61.5% — the lowest since 2021, erasing the entire post-pandemic recovery.

Since Trump's inauguration, over a million people have left the job market — which explains part of the reason the unemployment rate dropped last month.

Gaming is such as tough space.

This paragraph jumped out for me:

"We will reduce management layers to no more than 5, and where possible, 3. We will deliver success through a flatter organization that is built around makers (individual contributors focused on building), player-coaches (leaders who remain deeply involved in the work while developing their teams), and directly responsible individuals (DRIs) who own key decisions and outcomes."

ASHA: This is an important email I sent today to all employees at XBOX:

Team,

We are beginning the most significant restructure in XBOX history. After careful consideration, I've made the difficult decision to reduce our team by approximately 3,200 throughout FY27. This will include

“Developers even haven’t fully warmed up to the level of ambition you can have with Codex."

From Rohan (Codex PM):

"If you go into any large enterprise, maybe the top 5-10% percent most AI-pilled frontier engineers are maximally leveraging Codex. But then there’s a long tail of people that are still pair programming and doing less delegation.

Engineers have seen AI incrementally expand its capabilities over the last three years. But if you look to a non-developer, it’s kind of been just ChatGPT this whole time.

The type of things they’re asking ChatGPT are very different than what you can actually delegate to Codex.

Almost everything you need to do professionally should almost always just immediately have Codex start doing it.”

📌 Full episode: https://youtu.be/fAdFE7y6K2o


Peter Yang: "You cannot walk from point A to B inside OpenAI without hearing Codex get mentioned."

Here's my new episode with @TheRohanVarma (Codex PM) on how he uses Codex as his everything app to:

→ Prototype designs with Image Gen
→ Turn repeated work into automations
→ Turn project

Fable 5 will leave Claude subscriptions tomorrow at midnight. Here are 5 use cases worth trying before then, with prompts you can copy and paste:

1) FIND FABLE-WORTHY WORK
You're Fable 5, the most capable model available. Look through my projects, docs, and your memory, and list the top 5 tasks genuinely worth running on you. Rank them with a one-line reason each. Don't do the work yet.

2) GET LIFE & BUSINESS ADVICE
You're my business advisor. Read my plan doc and pull live data from my connected tools. Write a one-page assessment of my business and the top 3 things to focus on for the next 3 months, plus what to drop and why.

3) MAKE YOUR PROJECT SHIP-READY
I'm about to ship this project. Find everything wrong with it first. Read the whole codebase and hunt for real bugs, broken edge cases, and anything that'll break in front of a user. List each issue with how to reproduce it and the fix. Hold a high bar.

4) PLAN THE NEXT BIG THING
I want to plan a big project: [describe it]. Don't build yet. Lay out the full plan: phases, key decisions, risks, and open questions. Flag anything that could sink it, and make it clear enough that a cheaper model could execute step by step.

5) REFACTOR YOUR AI SKILLS
Refactor my AI skill system. Read all my skills, map where they overlap or conflict, then clean them up: tighten workflows, cut bloat, and standardize structure. Keep each skill working and check in at major milestones.

📌 Watch my tutorial for a live demo of all 5 use cases: https://youtu.be/5CBnWGP5vIs

Peter Yang: Claude Fable 5 is finally back, but you only have until July 7 to use it on your Claude subscription.

I made a new tutorial walking through 5 use cases worth trying Fable on:

→ Find Fable-worthy work
→ Get life and business advice
→ Make projects ship-ready
→ Plan the next

The ultimate test for coding AI: is software as a whole getting better? Are companies shipping faster? Do you now have apps and games you couldn't even dream of before? Is the software you're getting bug-free?

For me: my own ability to influence my productivity has radically increased. I feel a level of agency and autonomy I never had before. I'm free to create and execute. I can communicate a vision with my colleagues instantly. Personal software is real and a big win, I have a bunch of web tools and macOS apps I wrote just for myself.

For Vercel: it's been amazing to see our CTO @cramforce bring novel software to life with swaths of agents. Genuinely useful software that's having an impact industry-wide emerged nearly autonomously (http://chat-sdk.dev, http://just-bash.dev, http://deepsec.sh). That's just an example. Our PM team ships software and fixes bugs. Our junior engineers and interns ship quality software at scale. On a personal note, AI at my fingertips, as CLI or v0, allows me to keep up with all the product updates. I can build demos, apps and benchmarks on top of @vercel as new products go out.

But the ultimate jury is that people should love our products, expand their usage, refer their friends and colleagues to the platform. So far that tracks well, and it's the main metric I watch (and obviously we take token costs seriously, just like any other infrastructure spend… there isn't and has never been a proclamation to "tokenmaxxx").

RT @sapinker: The data are in: Populism makes countries worse off. "... a country being ruled by a populist is associated with a 10% lower…

We've put together a short history of how Claude Code came to be, told by the people who built it and the early users who helped make it what it is today.

https://www.anthropic.com/features/making-of-claude-code

Memory cost and capacity are significant issues for AI accelerators.

Unlike game rendering, model inference can have a deterministic memory access pattern. You don’t need “random access memory” at all for model weights, and you could tolerate cold-start latencies in the multiple milliseconds, as long as continuous reads were delivered at the necessary bandwidth.

NAND flash is over 100 times cheaper per GB than HBM, so there should be opportunity there, even after giving a flash controller a 1024 bit interface with HBM bandwidth.

You could make a specialized pin protocol that just supported pipelined transfer of full 16KB+ pages from the flash to program-managed accelerator scratchpad memory and improve per-pin performance over HBM, but it might be more convenient to make it still look like a true random access memory with very fragile performance characteristics, where anything but sequential reads falls off a 1000x+ performance cliff.

That has the advantage of automatically using existing cache hierarchies, and providing a natural path to update the flash memory with new model weights. With the stream-to-scratch interface, code has to be completely rewritten before it works at all, while the ram-emulation interface will start off just extremely slow, and you can incrementally sort out the changes for full performance.

There may be cases where there isn’t enough scratchpad SRAM to hold the weights for a layer, which might force you to deploy the old optical drive optimization technique of duplicating data in multiple places on a sequential read to avoid seeking, but there would be capacity to burn.

It might be possible to do something like cuda graph capture to record a memory access trace and have everything magically remapped to a linear sequence, but deploying programmer / agent elbow grease to manage transfers and access in a scratch ram ring buffer would be lower risk.

A split memory system consisting of some channels of flash and some channels of HBM will probably be suboptimal compared to a uniform memory, but it could be much cheaper, and allow much larger models to be run.

I think th case is strong for inference, but you have to stretch more for training. You can still linearize all the weight memory accesses, both reads and writes, but flash memory would quickly wear out from the writes, even if they were all perfectly page aligned. Replacing low-latency HBM with massively parallel cheap(er) DRAM at high latency might still be a worthwhile cost savings.

Many are asking how Replit is improving so rapidly—we closed the loop and the agent is self-improving.

Technical details here:

Michele Catasta: http://x.com/i/article/2070187317080584195

Atlanta-based real estate company saved $100k by replacing Salesforce with their Replit-built CRM.

Oakland will continue to be a case study in virtue signal policies by ideologies that don’t work

The city will continue to crumble until the common sense and business forces unite to vote smarter

Garry's List: Oakland saw a massive wave of small business closures the end of 2025, with owners citing the "increasingly impossible math" of staying open in the Bay Area.

Now the city is considering nearly doubling the minimum wage to $30/hr, making the math even harder.