We never said we don't upload any user data to the cloud. We said the code (file contents) specifically doesn't.
Paxel is here to help you, and over time as local models get better, we'll be able to do even more locally. Can't wait, tbh.
one popular theory is that research paper alpha* and lab publishing ~died when researchers realized that instead of fighting with marketing depts they could simply walk out the door and get >$100m for their legally protected tacit knowledge gained
california non-noncompetes have a bigger impact on knowledge spreading than github, arxiv, and huggingface combined
*btw this is a motivator for me to set up @aidotengineer as a product-centric industry conference to complement the paper-centric research conferences
roon: @beffjezos the sheer dollar quantity of ip leakage from openai has been just tremendous, probably in the trillions. openai truly has been quite open
Whenever I don’t use codex for a task, I ask myself why and usually realize that there’s some missing context, I needed to write a skill, or I just didn’t think to use it.
Rarely is it because the task is outside of the capabilities of the model. Overhang right now feels large.
idea - universal basic ai:
1 share of xai, oai, and ant to each US citizen.
cost
SpaceXai: $135.00 × 349 million = $47B
OpenAI: $733.54 × 349 million = $256B
Anthropic: $930.45 × 349 million = $325B
total $628B cost is 8.5% of US Govt budget, 62% of defense budget, roughly same as ~$700B of TARP program during the Financial Crisis (mostly given to big banks, this would be a lot more decentralized)
There should be a way to filter or sort all my Codex threads in different ways vs. only by project.
Like filter or sort by:
- All waiting for approval
- All currently working
I'm trying to keep it to 10 threads but it's already getting unwieldly. wdyt @ajambrosino ?
Peter Yang
this agentic coding crack is more addictive than video games smh
this agentic coding crack is more addictive than video games smh
Good night
Mirko Monti
I have to admit that Gstack by @garrytan is really cool (https://github.com/garrytan/gstack)
loving the /office-hours skill
very useful to test your business ideas
Ronin
okay i'm just gonna share this..
*that's part of workflow how i automated my solo marketing agency
been using this free repo for a few weeks. ships in hours what used to take days
23 ready skills. product → design → QA → docs
an 18-year-old won a hackathon with this repo last month. in 2 hours
while 80 others were still figuring out what to build (cheat code for planning)
made by @garrytan (president of YC). 108k stars. you probably haven't found it
[ what blew my mind ]:
the setup takes 3 commands.. not 3 days
> 23 structured skills covering every phase: think → plan → build → ship
> /office-hours - YC-style review before you write a line of code
> /plan-eng-review - full architecture from one product decision
> /design-shotgun - 4-6 design variants in parallel, opens in your browser
> /qa - opens a real browser and clicks through your flows. finds bugs in the running product
> GBrain - persistent memory between sessions. the agent never forgets again
[ how i'm using it ]:
i run a fully automated marketing agency. solo. zero employees
gstack handles a lot of what used to need 5 hires:
> /office-hours scopes client launches in 5 min instead of 30
> /design-shotgun ships 6 landing page variants in one session
> /qa catches broken automation flows before clients see them
> /document-release keeps client deliverables updated automatically
> GBrain remembers every client's voice + history across sessions
what used to eat my mornings now finishes in an hour
[ the part nobody's talking about ]:
most developers using Claude Code are leaving 80% of its potential on the table.. they treat it like autocomplete
gstack treats it like a development system. product thinking, architecture, design, QA, docs - all passed between phases automatically
you stop re-explaining context every step
[ and the craziest part ]:
Rakhatbek (the kid who won) said it would have been "practically impossible to build a game like this in 1 hour 45 minutes manually"
3 commands. 1 markdown file. 23 skills
most developers will never install it.. not because it's hard, but because they think it's just another tutorial
Repo: https://github.com/garrytan/gstack
100% FREE, OPEN-SOURCED
full breakdown here ↓
Noisy: http://x.com/i/article/2063288163141828608
Yes, if you build a foxconn factory for intelligence, the workers can't do smart things
Reusable intelligence means the agents are not boxed. Something bad happens? it's not hemmed in by code. It can just do the right thing.
Spock Woz: @alphabatcher i mostly agree.
the bigger shift isn't markdown skills replacing code. it's moving from hardcoded workflows and UIs to reusable intelligence.
the agent should be able to generate the logic and interface it needs at runtime, while infrastructure stays deterministic underneath.
Peter Yang
"If you're still manually reviewing every line of code, you're the bottleneck."
Here's my new episode with @kunchenguid, an ex-Meta L8 engineer who now ships up to 40 PRs a day with AI agents. Instead of manually reviewing code, he built an agentic engineering system that includes:
✅ Lavish, his free tool for annotating AI's plans as visual HTML artifacts
✅ gnhf or "good night, have fun," his free orchestrator to get agents working towards a goal while you sleep
✅ No Mistakes, his free AI code validation pipeline for catching errors before merge
Some quotes from Kun:
"If I spend a lot of time crafting detailed plans, then the agents can work for much longer."
"Every time I encounter friction in my workflow and I don't find an existing tool that can solve the problem, I just build something myself."
"To really scale how much we can get from the agents, we have to move ourselves out of the loop as much as possible."
📌 Watch now: https://youtu.be/88B6DimMD2g
Links to Kun's free agentic engineering tools below 👇
David Breslauer
Gbrain by @garrytan is so good, I've now moved by brain onto a dedicated server, that Codex, Claude, OpenClaw, and Hermes all connect to.
My agent conversations (and planning in particular) are so much more enriched and opinionated.
Every experiment I run in Codex uploads a summary and conclusion, so our mutual thinking continues to compound.
https://github.com/garrytan/gbrain
Paul Graham
Ideally you want to be dismissed as being on the right by people on the far left and dismissed as being on the left by people on the far right. This is not a sufficient condition for getting things right (choosing positions randomly would achieve it too) but it's a necessary one.
Rep. Jake Auchincloss 🟧
Russell Vought’s proposal to make politics, not peer review, the standard for NIH science grants would be a cataclysm for the American science enterprise.
Jason ✨👾SaaStr.Ai✨ Lemkin
At SaaStr AI 2026, @amasad CEO @Replit and I did an incredible deep dive
A top 0.1% power user + the CEO on stage together!
- Why our AI Agents really work so well
- When an AI Agent can beat a human, and why
- How Replit combines models
- Why your second and third AI Agent will be better
- Why every company will end up with internal AI "Oracles"
- How monorepos are a secret unlock to agent performance
and so much more!
Most engineering processes were built for teams shipping 10 to 15 PRs a month.
@kunchenguid, ex-Meta L8 engineer, is now shipping 40+ PRs a day with agents. That breaks a lot of existing processes behind:
→ Code review & QA
→ Merge gates and team handoffs
When output jumps 10x, the old workflows starts to crack.
The answer is to build systems that streamline planning, validation, review, and merging as much as possible.
📌 Watch him talk more about it here:
https://www.youtube.com/watch?v=88B6DimMD2g&t=2626s
Peter Yang: "If you're still manually reviewing every line of code, you're the bottleneck."
Here's my new episode with @kunchenguid, an ex-Meta L8 engineer who now ships up to 40 PRs a day with AI agents. Instead of manually reviewing code, he built an agentic engineering system that
Peter Yang
Most engineering processes were built for teams shipping 10 to 15 PRs a month.
@kunchenguid, ex-Meta L8 engineer, is now shipping 40+ PRs a day with agents. That breaks a lot of existing processes behind:
→ Code review & QA
→ Merge gates and team handoffs
When output jumps 10x, the old workflows starts to crack.
The answer is to build systems that streamline planning, validation, review, and merging as much as possible.
📌 Watch him talk more about it here:
https://www.youtube.com/watch?v=88B6DimMD2g&t=2626s
Peter Yang: "If you're still manually reviewing every line of code, you're the bottleneck."
Here's my new episode with @kunchenguid, an ex-Meta L8 engineer who now ships up to 40 PRs a day with AI agents. Instead of manually reviewing code, he built an agentic engineering system that
Codex use-cases: “From software engineering and design to data analysis and operations, Codex is becoming an AI teammate instead of just an AI assistant.”
Suraj Sharma: OpenAI just published dozens of real-world workflows showing how teams are using it to automate work.
> Manage your inbox and draft replies in your voice
> Review GitHub pull requests before human review
> Turn Figma designs into production-ready code
> Understand large
Vercel AI Gateway recovers on average over 1T tokens a month 🤯
Much like Stripe recovers revenue with smart retries on failed payments or credit card updates.
And we do it with 0️⃣ zero markup over the labs; adding redundancy, zero-data retention enforcement, observability, usage APIs, caps, …
http://vercel.com/ai-gateway
releasing tmr - the biggest code eval launch of the year
glad to have played a small part in defining the agenda for this very critical next phase in koding
swyx: lol heard a 2nd startup today that has made sales and evals based on this podcast
its fun to be "just an interviewer", but i'm always mindful/humbled by the fact that we are playing with live ammo here. LS has directly and indirectly impacted a lot of peoples' careers and
Kun Chen
i just had the pleasure of going on @petergyang's latest podcast episode and sharing how i use agents productively
we touched on a lot of practical tips of how i tackle the bottlenecks in planning, review and validation and achieve a constant flow state
check it out!
Peter Yang: "If you're still manually reviewing every line of code, you're the bottleneck."
Here's my new episode with @kunchenguid, an ex-Meta L8 engineer who now ships up to 40 PRs a day with AI agents. Instead of manually reviewing code, he built an agentic engineering system that
theartofbace
Just hit 10K MRR in 90 Days
These are 3 things that got us there
1. Get to market ASAP. 60% ready means you're ready
2. Post EVERY DAY on social. No exceptions
3. Build/hire out a team to grow & test quickly
@speaksilq
Big thank you to @Replit @amasad
Replit is about removing all distractions and have you focus on what matters — getting to market and getting the bag. Congrats!
theartofbace: Just hit 10K MRR in 90 Days
These are 3 things that got us there
1. Get to market ASAP. 60% ready means you're ready
2. Post EVERY DAY on social. No exceptions
3. Build/hire out a team to grow & test quickly
@speaksilq
Big thank you to @Replit @amasad
Diana
beyond model size, the more interesting frontier is a thin layer on top: a coding agent that writes an executable world model, checks it against observations, and compresses it toward the *simplest program that fits.
it rides every base-model gain for free. A new s-curve sitting on the bitter lesson
*"simplest program" is what ARC-AGI measures in "skill-acquisition efficiency"
x.com
