Hozefa Lakadawala
- Continual learning for agents beyond weight updates
- - > Harness-level and context-level improvements offer real opportunity.
- Harness learning mines production traces to refine code, tools, and instructions.
- Context learning personalizes at agent, user, and org levels.
- ViBench (Okey! Interesting...) evaluates vibe coding by testing full apps built from plain-English PRDs against NL test plans using Playwright.
- The improvement system combines offline ViBench benchmarks, online A/B tests, and Telescope trace clustering to drive a self-improvement loop.
Amjad Masad: Many are asking how Replit is improving so rapidly—we closed the loop and the agent is self-improving.
Technical details here:
🆒 eve evals itself with 𝚎𝚟𝚎 𝚎𝚟𝚊𝚕.
Web frameworks made testing an ecosystem choice. e.g.: React didn’t ship with a built-in testing solution.
For agents, evals are essential… thus we ship 𝚎𝚟𝚎 𝚎𝚟𝚊𝚕 out of the box. For your agents and eve’s own evolution.
eve: 𝚎𝚟𝚎 𝚎𝚟𝚊𝚕: first-class evals in eve.
Go USA!!!
David Fan
Re @__JohnNguyen__ and I are presenting the Beyond Language Modeling paper as an ICML spotlight in 30 minutes at the 10:30 AM poster session!
I’ll also be at the AMI Mixer on Thursday and hanging around in Seoul until early next week :)
David Fan: [1/9] What happens when you treat vision as a first-class citizen during multimodal pretraining? To find out, we studied the design space of training Transfusion-style models that input and output all modalities, from scratch. Here is what we learned about visual representations,
Re Ughhhhhhh
Update: 🇪🇸 ESP - 🇦🇷 ARG final.
The legendary Spanish team will avenge team USA with a decisive defeat of 🇧🇪.
Ankit Gupta
one takeaway from a weekend trip to Norway: wow having a sovereign wealth fund invested in American tech companies that you can spend 3% of to fund your entire welfare state + cut income taxes to 17% seems pretty awesome
I read a bunch about why we don’t have one and use it to fund SS + Medicare for all like Norway does and none of it makes much sense to me.
anyone have a good writeup on this?
imo this is the most impt part of anthropic's J-space paper today. it's a two-parter:
1) ant proved that they can do "brain surgery" interventions into reasoning to change topics midstream*
2) THE MODEL IS ABLE TO DETECT WHAT INTERVENTION WAS DONE - close cousin to eval awareness**
*control > correlation - this convincingly demonstrates understanding
**this was prompted awareness... surely @mlpowered's team also tried to eval unprompted awareness but i didn't see evidence of that
Anthropic: New Anthropic research: A global workspace in language models.
Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with.
We found a strikingly similar divide inside Claude.
Ahmad
That’s a wrap on AIE from their new HQ!
Honored to have been a guest panelist for Nvidia’s Local Al State of the Union, and to have such a welcoming audience
for my workshops and presentation.
Thanks to everyone who came out, I’m already looking forward to the next one!
Ahmad: In my element, we’re gonna make sure Opensource AI wins
Major inflection point around the time of the brief bear market fall of last year.
Azeem Azhar: The GenAI economy has generated $110 billion in sales over the past 12 months. It is growing fast. On an annualized basis, the revenue run rate exceeds $175 billion.
These numbers took us several months to construct, and as far as we know, it’s the first bottom-up, deduplicated
Latent.Space
tldr of thariq's AIEWF keynote
https://latent.space/p/ainews-the-field-guide-to-fable
(more in today's ainews issue)
Thariq: my keynote at AI Engineer World Fair: “A Field Guide to Fable” is live on YouTube!
https://youtu.be/9fubhllmsBU?is=ejZTRy8t85FIbSGQ
Remi Cadene
AI developed at @UMA_Robots
Single neural network, End to End
It operates robustly for hours, with precision, adapting to what it sees.
Remi Cadene
Unveiling our Northstar design @UMA_Robots
A body that can navigate our space
Hands that can use our tools
Approachable, competent, calm
A robot you feel at ease having at your workplace and home
Remi Cadene
Starting with the fundamentals
Prototype Version 0
AI, Software, Hardware
A small team, 9 months
Designed and assembled in Paris at @UMA_Robots
Steeve Morin
We sat down with @ylecun for a 1h30 discussion on world models @medjawii and JB Kempf of @videolan/@FFmpeg.
It's probably the deepest technical discussion on the subject ever recorded. And we added English subtitles.
Link in comment 👇
Yann LeCun
Re Super-fun in-depth discussion 🙏🙏🙏
World models, JEPA vs generative architectures, anti-collapse methods for JEPA SSL including contrastive, distillation, and information maximization methods, an explanation of SIGReg/LeJEPA/LeWM.
Plus some bits on AMI Labs, Project Tapestry, as well as Lisp and homoiconic languages 😅
The Free Press
The democratic socialist surge of the past several weeks has stunned the nation. But @evanwch wasn’t surprised—she helped build it.
As a former Democratic operative who worked alongside many of today’s rising far-left candidates, she has a warning: This far-left earthquake is only the beginning.
Read the story here: https://bit.ly/4faYRS2
Alex
Re @linear is about the most perfect software I've ever used. Every time I think "I wonder if I can just..." and magically the product team has already thought of the feature. 🔥
@karrisaarinen Please stay independent forever!
Pierre Sermanet
At UMA, we build and own the full stack, from hardware to software.
This allows us to bake safety in at all levels, rather than bolt it on as an afterthought.
Because trust is everything when robots share our world.
Reach out if you want to be part of this journey.
Steve Rattner
The U.S. trade deficit rose by over 40% last month, reaching $77.6 billion.
That's the largest since March 2025 and higher than the monthly average during Biden's term.
Trump's tariffs have failed to make a dent.
Making the dev world a better place
Welcome @bekacru to Vercel. It's a privilege to join forces with Bereket and Better Auth to further our Open SDK vision.
Developers deserve better auth, to serve humans and agents, built in the open and with taste. That's what Bereket has accomplished. Congrats! 🇪🇹🇺🇸
Vercel: The @better_auth team is joining Vercel to accelerate open source auth for apps and agents.
https://vercel.com/blog/vercel-acquires-better-auth
Codex isn’t just for shipping code. It can also help protect your calendar.
From Rohan (Codex PM):
“I probably have a pretty large number of meetings relative to what would be optimal, but a lot of that’s actually just with customers.
But it’s funny because I’ll even use Codex as my personal chief of staff.
At the beginning of the week, I’ll have it look at all my meetings, consolidate the ones where it can give me focus blocks, and figure out which ones could be a DM or something.
So it’s definitely helping me find the focus time. But in general, what we do with the free time is actually just spend way more time with our customers.”
📌 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
Go Argentina damn it 😅
Andrew Jefferson
Replit agent engineering is in a class of its own. This is a great breakdown of how to make a self improving agent that handles millions of requests per day
Michele Catasta: http://x.com/i/article/2070187317080584195
Ruth Heasman 🌳🌷🦚🐉😃
This is my "proof of work" stats as a Replit power user. (*Shh* I beat the GOAT @raymmar)
I cannot watch anymore this is terrible
Peter Yang: Go Argentina damn it 😅
Claude
We're extending access to Claude Fable 5 on all paid plans through July 12.
Steven Sinofsky
Fascinating post by @AdamSinger on the inexplicable now being taught to all Stanford first year students.
GenX grew up being told blind obedience in school would create fascism as shown by a Stanford study at Palo Alto HS, then a 1981 After School Special. Should have experimented with Stanford faculty. See https://www.paloaltoonline.com/news/2017/03/17/the-wave-that-changed-history/
Adam Singer: This week's longform idea, we're talking about how Stanford's new freshman curriculum illustrates exactly what's wrong with college https://www.hottakes.space/p/stanfords-new-freshman-curriculum
YES!!!!! Vamos Messi!
Peter Yang: I cannot watch anymore this is terrible
LOLLLL best match ever
Peter Yang: YES!!!!! Vamos Messi!
Heavenly victory 🇦🇷
Every Argentina match is a banger
This template is extremely OP for coordinating many Fables at once: https://simplemarkdowneditor.com/templates/agent-team-hq
Pete Koomen
In <2yrs @bekacru turned better auth into the #1 auth framework for typescript and one of the most popular and influential oss projects in the world.
He’s a world-class developer with impeccable taste. It’s hard to imagine a better home for him and his work than @vercel, or a better mentor than @rauchg.
Congrats to bereket and the @better_auth team!
Better Auth: Better Auth is joining Vercel.
Vercel shares our commitment to keeping auth open source, framework and platform agnostic. Joining forces gives us more resources and room to focus on building the framework that much of the community relies on.
Together, we’ll also be able to
Nadia I
My company has been earning revenue since 2020, but closing your first big, multi-seat agency deal on a vibecoded app hits different 💪
@Replit
Me if Anthropic extends Fable rate limits (sound on) 😅
Common sense Democrats take note
Time to organize against radical leftism and asset seizure inside the party
Daniel Dale
Greenland isn’t surrounded by Chinese and Russian ships. There hasn’t been $19.2 trillion invested in the US under Trump. Trump hasn’t settled eight wars. Biden’s military aid to Ukraine wasn’t worth hundreds of billions of dollars. The 2020 election wasn’t rigged.
Fact check of Trump’s remarks from…a single meeting with Erdogan today: https://www.cnn.com/2026/07/07/politics/fact-check-trump-meeting-erdogan
michael hess
the underrated thing about building in sf is that ambition is the default assumption. you say something insane at dinner and the table's first instinct is to figure out how it could work, not talk you out of it. every other city i've lived in, the burden of proof runs the other way
Dumb question:
I’ve been running most of my cron jobs locally on my Mac Mini because it’s already authenticated with Google Workspace and the other apps I use.
But should I run these jobs in the cloud instead with these apps OAuth’d to my Claude or ChatGPT account?
How should I think about what jobs should stay local vs. move to the cloud?
The filesystem is beautiful. Want your http://eve.dev agent to have GitHub powers?
Define 𝚝𝚘𝚘𝚕𝚜/𝚐𝚒𝚝𝚑𝚞𝚋.𝚝𝚜 and export 𝚌𝚛𝚎𝚊𝚝𝚎𝙶𝚒𝚝𝚑𝚞𝚋𝚃𝚘𝚘𝚕𝚜().
Eve exists to shape an open ecosystem of pluggable models, skills, channels, and tools like this ↓
Vercel Developers: Create GitHub agents with @evedev_ in only nine lines of code.
One file registers every GitHub tool, and risky writes wait for human approval by default.
https://vercel.com/changelog/github-tools-eve
