2026-03-20

me and everyone around me prompt codex and chatgpt with voice. the more decisions you can spit out to codex the better your code will be, so you're only interface limited
the "everyone will use speech" guys were right, just 234 products and 18344 softwares too early

If you are attending @APSphysics March meeting, come learn more about thermodynamic computing!
Our work on taming non-equilibrium thermal electron fluctuations in silicon is now accepted in Physical Review Applied.
Read more here:
https://journals.aps.org/prapplied/abstract/10.1103/5l2z-3f1h

People of pi. Do you feel experimental? Want to try a new edit tool? Stuff this into your ~/.pi/agent/extensions folder. Use it with your preferred model(s) for a while. Report back if it works.
Example: GPT 5.4 prefers rewriting entire files sometimes over doing multiple small edits. This mostly solves this for me.
https://gist.github.com/badlogic/30c35e985b5ca59f8e6f69a66520d487

Just completed our annual SOC2 audit using Vanta.
They audited every MDM config, device destruction certificate, patched vulnerability.
Honestly, it was painful. But we passed each test and I know we do right by our users.
Delve is a known "shortcut".
Never considered it.
Ryan: Delve, a YC-backed compliance startup that raised $32 million, has been accused of systematically faking SOC 2, ISO 27001, HIPAA, and GDPR compliance reports for hundreds of clients. According to a detailed Substack investigation by DeepDelver, a leaked Google spreadsheet

“If your $500K engineer isn’t burning at least $250K in tokens, something is wrong.”

Apple added AI coding agents from Anthropic and OpenAI to Xcode 26.3 two weeks ago. Today it's blocking the two biggest independent vibe coding apps from updating in the App Store.
Replit just raised $400 million at a $9 billion valuation. It generated $240 million in revenue last year and is targeting $1 billion in ARR by December. Since January, when Apple froze its updates, Replit dropped from first to third in the developer tools download chart. Two months of blocked updates did what no competitor could.
The stated reason: "longstanding App Store rules" against apps running code that changes their own functionality. The actual targets: embedded web views that preview generated apps, and the ability to create software for Apple devices.
Think about what Apple is asking. Replit has to open its generated apps in Safari instead of inside its own app. That kills the seamless build-test loop that makes the product work. Vibecode has to remove the ability to generate iOS apps entirely. Apple's message to Vibecode is that you can exist in the App Store as long as you don't make things for the App Store.
Apple's Services division did $108 billion last year. The App Store contributes an estimated $40 billion of that. Every app built through Replit or Vibecode that ships as a web app instead of a native iOS app is revenue Apple never touches. The 30% commission on a $0 App Store listing is $0.
Vibe coding tools let a marketer or a small business owner describe an app in English and have it running in minutes. That capability routes around the App Store entirely. Apple saw 50 million Replit users building apps that don't need App Store approval, don't pay the 30% cut, and don't require Xcode.
So Apple put vibe coding in Xcode and froze it everywhere else. The policy isn't about code execution rules from 2009. The policy is about who gets to be the on-ramp.
Wes Roth: Apple has quietly halted App Store updates for popular AI "vibe-coding" applications most notably the $9 billion startup Replit and mobile app builder Vibecode.
After months of pushback, Apple is reportedly demanding major UX changes.
Replit is being asked to force its

OpenClaw and Pi together are in the top 10 of all time software breakthroughs.
Chrys Bader: folks who are calling @openclaw pure hype are telling on themselves
openclaw is like the early internet, it's raw, unrefined, and takes a little doing to get things to work, but when you figure it out, it's transformative.
here are some real use cases that are having material

Responses to this are hilarious.
Codex implemented this correctly in 15 minutes btw
Theo - t3.gg: Just let Opus go for over an hour on a new feature. When it was done, I asked how I can test it.
20 minutes later, it realized I can't test it because it did the whole thing entirely wrong.
Idk how you guys use this model every day for real work 🙃

this is art

我正在用gstack验证我的健康app的想法
看看gstack到底有啥优秀之处
其实我之前也跟这些AI讨论了很多
---
gtack安装步骤：
1. git clone 下来，安装 bun， 如果你没装
2. 然后 ./steup --host auto
自动安装复制skills到你电脑，它会自动检测你用的哪个ai，你不用管
3. 然后发送：
/gstack-office-hours + 你的产品idea
然后你就拿到YC验证Startup的知识杠杆了

time to clone Claude Code channels: -13 days.
and anthropic even named it the same as the preexisting pi extension (not saying they cloned it, and even if, GG!).
that's hilarious.
JB Cornac: @badlogicgames https://www.npmjs.com/package/@e9n/pi-channels
Already exists and it’s even called the same lol

Wasn't joking about this one btw
You can reverse-engineer pretty much any part of Apple platform internals in seconds using Claude or Codex with Hopper MCP
Oskar Groth: @sids7 Connect to Hopper MCP and it’s basically as good as having access to all of Apple’s internal source code

Cursor is raising at a $50 billion valuation on the claim that its “in-house models generate more code than almost any other LLMs in the world.” Less than 24 hours after launching Composer 2, a developer found the model ID in the API response: kimi-k2p5-rl-0317-s515-fast.
That’s Moonshot AI’s Kimi K2.5 with reinforcement learning appended. A developer named Fynn was testing Cursor’s OpenAI-compatible base URL when the identifier leaked through the response headers. Moonshot’s head of pretraining, Yulun Du, confirmed on X that the tokenizer is identical to Kimi’s and questioned Cursor’s license compliance. Two other Moonshot employees posted confirmations. All three posts have since been deleted.
This is the second time. When Cursor launched Composer 1 in October 2025, users across multiple countries reported the model spontaneously switching its inner monologue to Chinese mid-session. Kenneth Auchenberg, a partner at Alley Corp, posted a screenshot calling it a smoking gun. KR-Asia and 36Kr confirmed both Cursor and Windsurf were running fine-tuned Chinese open-weight models underneath. Cursor never disclosed what Composer 1 was built on. They shipped Composer 1.5 in February and moved on.
The pattern: take a Chinese open-weight model, run RL on coding tasks, ship it as a proprietary breakthrough, publish a cost-performance chart comparing yourself against Opus 4.6 and GPT-5.4 without disclosing that your base model was free, then raise another round.
That chart from the Composer 2 announcement deserves its own paragraph. Cursor plotted Composer 2 against frontier models on a price-vs-quality axis to argue they’d hit a superior tradeoff. What the chart doesn’t show is that Anthropic and OpenAI trained their models from scratch. Cursor took an open-weight model that Moonshot spent hundreds of millions developing, ran RL on top, and presented the output as evidence of in-house research. That’s margin arbitrage on someone else’s R&D dressed up as a benchmark slide.
The license makes this more than an attribution oversight. Kimi K2.5 ships under a Modified MIT License with one clause designed for exactly this scenario: if your product exceeds $20 million in monthly revenue, you must prominently display “Kimi K2.5” on the user interface. Cursor’s ARR crossed $2 billion in February. That’s roughly $167 million per month, 8x the threshold. The clause covers derivative works explicitly.
Cursor is valued at $29.3 billion and raising at $50 billion. Moonshot’s last reported valuation was $4.3 billion. The company worth 12x more took the smaller company’s model and shipped it as proprietary technology to justify a valuation built on the frontier lab narrative.
Three Composer releases in five months. Composer 1 caught speaking Chinese. Composer 2 caught with a Kimi model ID in the API. A P0 incident this year. And a benchmark chart that compares an RL fine-tune against models requiring billions in training compute without disclosing the base was free.
The question for investors in the $50 billion round: what exactly are you buying? A VS Code fork with strong distribution, or a frontier research lab? The model ID in the API answers that.
If Moonshot doesn’t enforce this license against a company generating $2 billion annually from a derivative of their model, the attribution clause becomes decoration for every future open-weight release. Every AI lab watching this is running the same math: why open-source your model if companies with better distribution can strip attribution, call it proprietary, and raise at 12x your valuation?
kimi-k2p5-rl-0317-s515-fast is the most expensive model ID leak in the history of AI licensing.
Harveen Singh Chadha: things are about to get interesting from here on

30s from Figma to animated demo. Designers & devs feasting rn. @Replit

1,600 commits. 140,000 lines of code. Built in 10 days.
Then 4,000 users showed up on launch day and the whole thing went up in flames.
@danshipper on the reality of vibe coding at scale:
http://every.to/chain-of-thought/when-your-vibe-coded-app-goes-viral-and-then-goes-down

Much of today’s AI-generated graphic designs look like slop because there is a lack of high-quality, open datasets. This space is a cluster of walled gardens (@figma , @canva , @Adobe ). We’ve built one of the largest graphic design dataset, 1.5 million compositions spanning several categories. LICA is an order of magnitude larger than existing datasets and enables tasks like layer-aware inpainting, structured layout generation, and temporally-aware generative modeling. @huggingface link below.

Meet Codex for Students.
We're offering college students in the U.S. and Canada $100 in Codex credits.
Our goal is to support students to learn by building, breaking, and fixing things.
http://chatgpt.com/codex/students

great read from Dan about vibe coding an entire app and then vibe fixing it after it started crashing
Dan Shipper 📧: Last week our vibe-coded agent-native document editor, Proof, went viral.
4,000+ documents were created in the first two days—and then the app started crashing uncontrollably.
I vibe coded all of Proof—so I spent the next week or so not sleeping and watching Codex agents debug

our March demo night @southpkcommons showed us a the future of digital public squares, interactive dramas, and ambient intelligence.
how we live and work is changing...
h/t @spencerc99, @eddiejiao_obj, @annieqxx, @johnjyang, and @shermangrewal for showing us how.

If you care about the future of OpenClaw, be sure to follow Dave - he's on the board with @steipete and they are working together to protect the future of the project. Check out his conversation about the Foundation with @jason and @TWiStartups - https://www.youtube.com/watch?v=KEHMoI00q3w
Dave Morin 🦞: 🦞 Been working with Peter Steinberger (@steipete) on the OpenClaw Foundation structure for weeks. A home for thinkers and hackers and those that want to own their data. Honored to serve as the founding independent board member. This community built something extraordinary, our

i was skeptical of the openclaw concept first, but turned out i was just stupid
very visionary product. perfectly predicted and made it obvious how having access EVERYTHING from ANYWHERE to do ANY work is clearly how the world will run very soon

for a limited time, get 1 month free
(or $20 in credits)

simply gift a friend a month of Replit Agent 4:

- they get 1 month of Core free & you get 1 month free

don’t miss out

You can get Replit Core FREE for 1 month + gift someone Replit FREE for 1 month.
Win - Win 🤝🏻🤝🏻
@amasad @Replit
Replit ⠕: for a limited time, get 1 month free
(or $20 in credits)

simply gift a friend a month of Replit Agent 4:

- they get 1 month of Core free & you get 1 month free

don’t miss out

taking my dog on a walk while replit builds my agent

Claude Code on desktop lets you select DOM elements directly, much easier than describing which component you want updated!
Claude gets the tag, classes, key styles, surrounding HTML, and a cropped screenshot. React apps also get the source file, component name and props

Pleased to collaborate with Sabastian Niles on this byline in @FortuneMagazine. One of the most important questions emerging around agent-to-agent interactions is: when AI agents negotiate on behalf of competing interests—with no human in the loop—what ensures they remain honest? Interestingly, a London banking house from 1832 points the way. https://bit.ly/4spc27e #AgenticAI #TrustedAI @salesforce

Pleased to collaborate with Sabastian Niles on this byline in @FortuneMagazine. One of the most important questions emerging around agent-to-agent interactions is: when AI agents negotiate on behalf of competing interests—with no human in the loop—what ensures they remain honest? Interestingly, a London banking house from 1832 points the way. https://bit.ly/4spc27e #AgenticAI #TrustedAI @salesforce

The PM playbook was built on an assumption that the technology underneath your product is roughly stable
With the current pace of model progress, this is no longer true. Here's how we've evolved the PM role:

When Copernicus proposed heliocentrism in 1543, it was actually less accurate than Ptolemy's geocentric model - a system refined over 1,400 years with epicycles precisely tuned to match observed planetary positions.
It took another 70 years before Kepler, working from Tycho Brahe's unprecedentedly precise observations, replaced Copernicus’s circles with ellipses - finally making heliocentrism empirically superior.
Terence Tao's point is that science needs a high temperature setting. If we only fund and follow what's most state of the art today, we kill the ideas that might need decades of work to surpass some overall plateau.

You can now schedule recurring cloud-based tasks on Claude Code.
Set a repo (or repos), a schedule, and a prompt. Claude runs it via cloud infra on your schedule, so you don’t need to keep Claude Code running on your local machine.