Stephen

MTS @skypilot_org | Founding Design prev @nuoninc | HCI | Dev Tools & AI Infra

[Stephen @stephenkevn]

San Francisco never ceases to amaze.

Recent Camera roll:

[Stephen @stephenkevn]

Wow, there's some amazing work in these replies, thank you! 🙏🏽

Just to clarify, this is a full-time, in-house role (San Mateo, remote for exceptional folks).

If that's you, apply here:
https://t.co/E5TCFIVL53

[Stephen @stephenkevn]

Hey Design twitter! I'm looking for a motion designer who's obsessed with craft and storytelling.

know anyone? would love an intro!

[Stephen @stephenkevn]

@mamuso Love the gallery!

[SkyPilot @skypilot_org]

Zongheng & Zhanghao after squeezing 25% util out of your $100M GPU fleet

[SkyPilot @skypilot_org]

Check out the blog: https://t.co/KpdgJxM2mL

[Stephen @stephenkevn]

@prekesh Love the framing on these.

[Stephen @stephenkevn]

Did someone say everything bagel? 👀

Training, inference, agents, and sandboxes all on one run.

[SkyPilot @skypilot_org]

RL is Everything, Everywhere, All at Once.

A single RL run spans training GPUs, inference engines, agent loops, and sandboxes, all feeding each other.

We put the whole stack on an everything bagel, and wrote an illustrated tour of it, its tradeoffs and failure modes.

[Stephen @stephenkevn]

@tylerjdunn @continuedev This is huge! Congrats!

[Ty Dunn @tylerjdunn]

Everyone has been asking me what I’m up to after the @continuedev acquisition

The answer: I wrote a paper with Disintegrator titled *The Superdark Factory: Toward the Full Automation of Software* that was published in Antikythera (MIT Press) today

https://t.co/XstbhbEHBS

[Stephen @stephenkevn]

Guess it's summer in San Francisco.

[Stephen @stephenkevn]

Local models had to settle for .gguf extension cause Adobe illustrator called .ai first

[SkyPilot @skypilot_org]

Introducing SkyPilot Agent Sessions.

Start an agent, close your laptop. It keeps working on your own K8s cluster, and the dashboard tells you the moment it needs you: from the CLI, the browser, or your phone.

[Stephen @stephenkevn]

Time to center that div.

[Claude @claudeai]

We’re introducing Claude Fable 5.1 and Claude Mythos 5.1.

They're the world’s most advanced models for coding and knowledge work.

[Stephen @stephenkevn]

Legendary run!

[Tim Cook @tim_cook]

Sending lots of love to the Apple community on my last day as CEO. My title changes tomorrow, but the love I have for the Apple community never will. Thank you for being a constant source of inspiration. My gratitude is endless, and I’m excited for the next chapter!

[Stephen @stephenkevn]

@MadhavJivrajani Super excited!

[Stephen @stephenkevn]

Golden goat coffee wait time. 📈

[San Francisco Chronicle @sfchronicle]

Cognition, the company behind an autonomous AI agent designed to help write and test code, has signed a lease for 333 Brannan St., a 180,000-square-foot building. https://t.co/l4lPZyxiKd

[Zhanghao Wu @Michaelvll1]

OpenAI announced today it will end its agreement providing models to Cursor following the SpaceX acquisition.

It’s tempting to read this as another chapter in the @elonmusk vs @OpenAI drama. There is another take about the market structure.

1. Model access is now a competitive weapon. Anthropic cut off Windsurf during its acquisition talks, revoked OpenAI’s Claude access in 2025, and cut off xAI in January. OpenAI cutting off Cursor isn’t an anomaly. It’s the playbook.
2. When a trillion-dollar provider terminates your access, your business is disrupted and theirs is unaffected. That’s not a partnership. That’s dependency.
3. Cursor anticipated this after the Windsurf cutoff and built its own Composer models on open-weight foundations. That decision is the difference between a painful announcement and an existential one.

Enterprise buyers learned this lesson with open source a decade ago: you don’t adopt it because it’s free, you adopt it because it guarantees a path out.

Open-weight models have to win. Not for ideology. For survivability.

[Stephen @stephenkevn]

@rbccawang This is true.

[Jeff Dean @JeffDean]

Announcing Discovery Loop!

I am very excited to announce that, along with my longtime friends and collaborators @Sanjay_Ghemawat, @OriolVinyalsML and @quocleix, we are founding Discovery Loop (@DiscoLoopAI), a Public Benefit Corporation whose mission is to automate machine learning, science, and engineering to accelerate discoveries and progress. The four of us have worked together for 14 to 30 years, and have helped build some of the world’s most used products, infrastructure and AI models, and we’re excited to turn our attention to this ambitious endeavor.

♾

Learn more at: https://t.co/Rv3LMdLluK

[Stephen @stephenkevn]

@martin_casado I think it’s not thinking ability declining, it’s cognitive offloading. When a good-enough answer is always one prompt away, the reason to keep engaging drops, and that disengagement shows up in brain activity. Mainly affecting curiosity.