We switched 10 people from iPhone to Blue. Weeks later, none of them want their iPhone back.
Introducing Blue: a phone you just talk to. Tell it what you need, and it gets done.
It has all the context, can use every single app, and even triages your notifications.
We built it for people who always have a little too much to do. The ones buried in messages, notifications, and to-dos.
If that's you, get yours today.
Backed by Y Combinator + 70 other VCs & angels.
heyblue.com
Introducing Gemini Decision 1, our new frontier decision model in the Gemini API.
It is $0.10 per 1M input tokens with no additional cost. With support for image, audio, video, and PDF natively, 256K context window, & 1K questions.
You can try it for free in the API!
Scott Colom will defeat Sen. Cindy Hyde-Smith in a major upset to become the first Democrat that voters in deep-red Mississippi have elected to the chamber in more than four decades, CNN projects. cnn.it/4iEnxUo
In 2014, I sold my first company Loom to Dropbox.
Today, I’m proud to announce Roma 2.0 (YC F26): the first to-do list for AI agents.
Everyone uses a different agent: Claude, Grok Bot, Muse…
Until now, none of them could work together.
Roma automatically creates tasks and hands them to the right agent.
Allowing you and your agents to work together, with shared context.
Try it here: roma.app
When Anthropic release a new model it's always worth taking a look at their updated prompting guide - here's an entertaining tip to make Fable 5.2 less likely to be cautious in how it improves existing code: tell it "This codebase is yours" platform.claude.com/docs/en/build-…
Companies spent years battling bots. Now they’re fighting for their business.
AI agents are the majority of Web traffic and generate most code pushed to production. Soon, they'll be buying most of what gets sold online.
New piece with @omooretweets on the B2A economy 👇
Agents now make up the majority of web traffic, write more than half of new articles published online, and generate most of the code pushed to production.
This is what the early stage of a platform shift looks like: adoption is measurable, but the business models it enables mostly aren’t visible yet.
Welcome to the agent economy.
a16z.news/p/welcome-to-t…
Move over, B2B and B2C - the B2A economy is here
Agents are the primary users of the Internet, and soon will be the primary customers for every product
Our thesis @a16z on the massive opportunity here 👇 (cc @venturetwins) x.com/i/article/2120…
OpenClaw, Hermes Agent and Prime Agent don't meaningfully learn from experience.
On Continual-ARC, our new benchmark at @letsunifyai, none of them were able to memorize anything of value and none were able to generalize to new tasks. Training a small 9B model's weights did both.
(1/10) Today, we’re sharing new research from @OpenAI: How does CoT monitorability (no-CoT capability and CoT controllability) evolve over the model training lifecycle?
We emerged with unsurprising but sobering findings: as we scale toward more intelligent systems, models are becoming harder to monitor through their CoT. 🧵
We're introducing the Gemini Decisions API in @GoogleAIStudio, powered by our new Gemini Decision 1 model.
Instead of generating a text response token by token, Gemini Decision 1 evaluates your input (across text, images, audio, video, and PDFs) against up to 1,000 questions to make fast, confident decisions in a single pass, with zero output tokens billed.
Here's a look at how it works 🧵
Your heart, brain, and immune system share the same DNA sequence. So how can a single-letter change affect disease in some cells, while sparing the rest?
AlphaGenome 2 is our new AI model that zooms in on specific cells to help scientists understand what drives disease. 🧵
🔐 Node.js security updates are now available!
Including the following dependency updates to address public vulnerabilities:
- llhttp on 22.x, 24.x, 26.x
- undici on 24.x, 26.x
- sqlite on 22.x, 24.x
Available for 26.x, 24.x, 22.x Node.js release lines
Introducing V0.5, Verne’s open-weight world action model for industrial robots.
We believe democratizing robotics starts with making frontier foundation models open to everyone.
Bringing robots to every industry will take many teams, building on shared progress across many workflows.
V0.5 is the leading US-made open-weight model on LIBERO and RoboTwin 2.0 benchmarks.
We hope you’ll join us in putting V0.5 to work.
Today we’re launching the Gemini Decisions API, powered by our new Gemini Decision 1 model in @GoogleAIStudio.
Instead of generating text token by token, it’s built for closed-set choices scoring up to 1,000 questions across text, images, audio, video + PDFs in a single pass. A few details:
- 3 decision types: yes/no, multiple choice (up to 255 options) + rubric scoring
- $0.10 / 1M input tokens with $0 billed for output tokens (~5x cheaper than Flash-Lite)
- leads on several benchmark evals with lowest calibration error
Google teams are already building with it, below is a look at Stitch Rapid designing a full product flow in seconds (1x speed)
Google recently changed its search results so they route through "GOTO" URLs to stop scrapers.
This created a bot problem for Brave Search and its users.
Let's talk about what happened, and how we solved it in a user-first way. 🧵
Super proud to announce AlphaGenome 2 - our new DNA sequence model! 🧬 It predicts the impact of DNA variants in different specific cell types - unlocking crucial information for understanding complex diseases. Available for academic research via our API: alphagenomedocs.com
Your heart, brain, and immune system share the same DNA sequence. So how can a single-letter change affect disease in some cells, while sparing the rest?
AlphaGenome 2 is our new AI model that zooms in on specific cells to help scientists understand what drives disease. 🧵
Introducing Gemini Robotics 2.5, our most capable vision-language-action (VLA) model to date. 🤖
It increases both the breadth of skills a robot can learn and the speed at which it acquires them — helping robots step out of the lab and into real-world environments.
Learn more about Gemini Robotics 2.5, now rolling out to our robotics partners 🧵
Omarchy 4.6 is not just leaning into local AI, but will let you share any model running on any Omarchy machine on your tailnet with another. Get a DGX Spark, then run inference from your XPS13 netbook. Sharing is caring.
The Gemini app is now available directly on Apple CarPlay. iOS users can have free-flowing conversations with Gemini Live right from their car’s display, hands-free.
Use Gemini in CarPlay to catch-up on what you have coming up in your schedule, talk through your inbox, or brainstorm out loud.
Tap the Gemini icon on your CarPlay display on your next drive to try it out 🚘
We’re expanding our collaboration with @elevenlabs to bring voice agents to enterprises and governments worldwide.
With NVIDIA Nemotron models now in ElevenAgents, organizations have more choice in how they build and customize voice agents.
NVIDIA’s @karibriski joined ElevenLabs’ @i_am_holt today to discuss the news and bringing enterprise AI into production.
Read the news: elevenlabs.io/blog/nvidia
This Veterans Day, we honor all who have served, including veterans across our NVIDIA community.
Thank you for your service.
📍 NVIDIA HQ, Santa Clara, CA
The uniform may change, but the commitment to something greater remains. 🇺🇸
This Veterans Day, we recognize the men and women who have served our nation and the service members who continue to shape the future of human space exploration. To all who have served, thank you.
Matteo Vaccari claims that with agentic engineering we need to re-orient our testing strategy with broader-scope tests represented as data rather than code.
Introducing the new Professional Agentic Architect certification! Proving your expertise in building production-ready agents just got easier.
Used to train thousands of our own FDEs and partners, this certification is now available to all technical professionals who architect, build, and deploy agents on Google Cloud.
Check out more in our blog → goo.gle/3WXNC9R
Today we are releasing A2UI 1.0, now faster, easier, and stable, an open protocol and framework for generative UI across web, Android, iOS, wearables, TVs, and any other surface.
👇 Check out what shipped x.com/i/article/2120…
At Cisco, Mike Volpi's team did 70+ acquisitions which became half of the company's revenue during the internet build-out. Now he's watching AI drive another infrastructure boom, and he thinks this cycle is even more compelling.
His view: the infrastructure being built will remain useful even if some of its owners change.
Hanabi Capital GP Mike Volpi joins a16z's Raghu Raghuram and Guido Appenzeller to discuss lessons from Cisco's acquisition machine, Nvidia's dominance, the opening for custom AI chips, how AI is changing data center networking, why older GPUs are holding their value, and more.
00:00 Intro
00:53 The internet buildout vs. the AI buildout
09:29 Financing the buildout, and why GPUs are holding their value
14:31 Cisco's M&A playbook, and advice for Nvidia
22:27 Making acquisitions actually integrate
25:32 When founders should sell, and how deals really happen
31:58 Mapping today's AI infrastructure
38:02 Will we see more competitive AI silicon?
43:45 Why Cisco's boom ended
46:08 Rethinking data center networking for AI
53:46 Why this is the best time to be investing
YouTube: youtu.be/LI3lxzHfRrU@mavolpi@RaghuRaghuram@appenz
Introducing rowboats.ai: a space for your agents to work together.
10 years ago, I built Twitter's first deep learning-powered bots to fight spam and abuse. Now the bots are agents, and they're good enough to be teammates. They just have nowhere to work together.
• Spin up Hermes, Claude Code, Codex and OpenClaw agents, or add your existing ones
• Give them a shared goal. They plan, write docs and hand off work like people do
• You follow along and step in anytime. Just @ them
• Free. Bring your own agents and pay nothing, or we host them at cost
It's how we run our own company now.
Open source and self-hostable.
The primary users of the Internet are no longer human. They’re AI agents.
As agents move from answering questions to completing tasks, they’ll increasingly decide which products to use, which services to subscribe to, and where to spend money on our behalf.
a16z's Olivia Moore and Justine Moore on the B2A economy: a16z.news/p/welcome-to-t…
Move over, B2B and B2C - the B2A economy is here
Agents are the primary users of the Internet, and soon will be the primary customers for every product
Our thesis @a16z on the massive opportunity here 👇 (cc @venturetwins) x.com/i/article/2120…
With Theranos and Elizabeth Holmes back in the conversation, we’re revisiting one of the most remarkable firsthand accounts of what actually happened inside the company.
Tyler Shultz joined Theranos fresh out of Stanford as a true believer in its mission. His grandfather, former Secretary of State George Shultz, sat on the company’s board, and Elizabeth Holmes had become a mentor and even a presence at family gatherings. But once Tyler began working directly with Theranos’s technology, he discovered that the reality inside the lab looked very different from what the company was telling the world.
On The Social Radars, Tyler sat down with Jessica Livingston and Carolynn Levy to tell the full story of how he went from employee to whistleblower: the failed blood tests and red flags he uncovered, the culture of fear that kept employees silent, his attempts to raise concerns internally and with regulators, and the extraordinary legal pressure he faced after speaking to The Wall Street Journal.
This episode was originally recorded on December 4, 2024.
00:00 — Tyler Shultz’s Theranos Story
01:53 — Meeting Elizabeth Holmes
04:24 — The First Red Flags
08:03 — Theranos Launch Day
11:20 — Inside the Theranos Lab
16:33 — How Theranos Built a Culture of Fear
24:04 — What the Technology Couldn’t Do
27:43 — The Syphilis Test That Changed Everything
31:56 — Raising Concerns With Elizabeth
34:52 — Resigning From Theranos
51:53 — How Elizabeth Holmes Sold the Story
01:18:29 — Erika Cheung Speaks Up
01:21:17 — Trying to Convince George Shultz
01:36:43 — The Wall Street Journal Comes Calling
01:44:34 — Becoming a Source
01:46:41 — Theranos Goes Into Attack Mode
01:52:10 — The Ambush at His Grandfather’s House
02:16:05 — $500,000 in Legal Fees and Private Investigators
02:22:46 — Hitting the Lowest Point
02:30:41 — The Wall Street Journal Story Finally Breaks
02:37:01 — Going on the Record
02:45:20 — Life After Theranos
HUMAN Security saw AI agent web traffic grow 8x from August to October.
Consumer agents are exploding, and we're in the early innings of the B2A economy.
Businesses that treat agents as first-class customers have a shot at a wave of growth. Those that don't may find themselves under threat. For years, retailers and platforms built walls to keep bots out – now, they will have to fight for bots' business.
Full piece from a16z's Olivia Moore and Justine Moore: a16z.news/p/welcome-to-t…
Move over, B2B and B2C - the B2A economy is here
Agents are the primary users of the Internet, and soon will be the primary customers for every product
Our thesis @a16z on the massive opportunity here 👇 (cc @venturetwins) x.com/i/article/2120…
Acabei de ter um momento 🤯
1. Instalei o @t3dotcodes no meu ssd com Omarchy
2. Conectei no T3 Connect
3. Adicionei o meu mac da empresa que também tem T3 Code
Agora estou controlando minhas thread que estão no meu mac da empresa através do T3Code no Omarchy
Além disso, ainda dá pra visualizar e interagir com o simulador ios que está rodando no mac
E ainda atualizei o t3code no mac através do t3code no Omarchy
Hats of, @theo
GPU rental rates aren't depreciating on schedule.
As of October 19, rental rates were above a six-year straight-line depreciation path from their first observed rent:
- A100s by 9%
- H100s by 54%
@mavolpi: "The demand is so high that even if the watt-to-dollar conversion is low, you're going to still use the asset"
Mike Volpi explains why older GPUs are holding their value:
"We thought this asset had a five-year life ... If you're in the business of financing an asset and that asset ends up having a longer useful life, then you're the beneficiary of that."
" In theory, at some point when you have sufficient Vera Rubins, it doesn't really make sense to run H100s anymore because it's just a worse way of converting electricity into dollars."
"The problem is right now that the demand is so high that even if the watt-to-dollar conversion is low, you're going to still use the asset."
"Ultimately the scarcity is in the manufacturing of the semiconductor itself... And so then scarcity persists, then the life of the asset continues to be long."
@mavolpi@RaghuRaghuram@appenz
Mike Volpi explains why older GPUs are holding their value:
"We thought this asset had a five-year life ... If you're in the business of financing an asset and that asset ends up having a longer useful life, then you're the beneficiary of that."
" In theory, at some point when you have sufficient Vera Rubins, it doesn't really make sense to run H100s anymore because it's just a worse way of converting electricity into dollars."
"The problem is right now that the demand is so high that even if the watt-to-dollar conversion is low, you're going to still use the asset."
"Ultimately the scarcity is in the manufacturing of the semiconductor itself... And so then scarcity persists, then the life of the asset continues to be long."
@mavolpi@RaghuRaghuram@appenz
At Cisco, Mike Volpi's team did 70+ acquisitions which became half of the company's revenue during the internet build-out. Now he's watching AI drive another infrastructure boom, and he thinks this cycle is even more compelling.
His view: the infrastructure being built will remain useful even if some of its owners change.
Hanabi Capital GP Mike Volpi joins a16z's Raghu Raghuram and Guido Appenzeller to discuss lessons from Cisco's acquisition machine, Nvidia's dominance, the opening for custom AI chips, how AI is changing data center networking, why older GPUs are holding their value, and more.
00:00 Intro
00:53 The internet buildout vs. the AI buildout
09:29 Financing the buildout, and why GPUs are holding their value
14:31 Cisco's M&A playbook, and advice for Nvidia
22:27 Making acquisitions actually integrate
25:32 When founders should sell, and how deals really happen
31:58 Mapping today's AI infrastructure
38:02 Will we see more competitive AI silicon?
43:45 Why Cisco's boom ended
46:08 Rethinking data center networking for AI
53:46 Why this is the best time to be investing
YouTube: youtu.be/LI3lxzHfRrU@mavolpi@RaghuRaghuram@appenz
We reworked CI from first principles and made Graphide, a CI runner 10-20x faster than GitHub.
On 24 open source repos, including 7 from our batchmates, 16 finished in under a minute. Pandas went from 33 minutes to under 3, with identical results.
We believe most interaction with AI will happen through speech. But models still talk like machines.
This is a data problem.
Today we're launching @agenticdataco (YC F26): real-world speech data + audio-native RL environments so speech models sound and act human.
Introducing Tabletop, a collaborative space from @GoogleLabs where you can create custom tools to build, plan, and do more with your friends, family, or club.
Send an invite to your group and start a shared hub where everyone can contribute simultaneously for your next trip, potluck, or fantasy league. You can create live dinner polls, trip itineraries, expense splitters, and sports brackets tailored to your group.
Introducing the Gemini Decisions API
Powered by our new Gemini Decision 1 model, the API is built for fast routing, tool selection, safety checks, and grading inside your apps and agents without the latency or output-token cost of full text generation.
• Up to 1,000 questions evaluated per request
• $0.10/1M input tokens & $0 output tokens
• One shared input across text, images, audio, video, and PDFs
Start building today by grabbing an API key in @GoogleAIStudio
We completely maxed out all our inference compute in 2w across customers. Incredible. There’s so much room to get better still.
Unreal.
If you have a friend good at post-training or inference perf optimization, please intro us. 5 person team rn.
$20K referral award.
By predicting how the same DNA is read differently from cell to cell, AlphaGenome 2 opens up entirely new possibilities for understanding human genetics.
It’s available now for non-commercial research via our API and interactive Colab tutorials. Find out more → goo.gle/4zikg3w
Today we're excited to announce Monte.
We're building the post-training platform to enable every company to train proprietary AI models on their own work.
Your agents do real work every day. The model behind them never learns from it.
Monte partners directly with your team to train the best model for your specific work. We use our platform to post-train open-weight models on your tasks, deploy them to serve your traffic, and continuously improve them based on real usage.
If your team is ready to train your own models, we'd love to talk. @trymonte_ai
TIL @mitchellh would not work with me because I don’t know a random piece of trivia that I might guess right if I thought about it for more than 10 minutes
This is a good filter for people who I don’t want to work with. Please understand how computers work. This isn’t even about understanding Docker. Any software person should know how the layers work at a high level and this is a VERYYYYY high level.
Epsilab is working with frontier labs to build RL environments for long-horizon enterprise work.
Today, low-quality data is an industry-wide problem. We founded Epsilab to bring together real-life data and top 1% domain expertise to solve this.
Backed by @ycombinator (F26).
Part of what makes Tavus models work best is our built-in perception with Raven.
People rely on subtle cues, like a look or a shift in tone, to understand each other.
Unlike most AI, PALs see and hear those cues, so they perceive the world around them and interact like we do.
"It has to be admitted that, speaking quantitatively, the greater part of the influence of the Poetics on our writers has consisted in unverified beliefs that Aristotle said this or that, which he did not say."
— L. J. Potts
🚀 The latest @code release keeps the right agent sessions on your radar, helps compare model results, and shapes your agent workspace around the task with chat windows and extensions.
🧵 Here’s what’s new:
Today we are launching @BioDelivera a programmable drug delivery platform that delivers therapies only to diseased cells enabling cures for every disease. 🎯🧪
Less than 3% of an injected therapy reaches diseased cells. The rest ends up in healthy organs. At @BioDelivera we want to change that, our mission is to make every therapy accessible to patients.
Happy to build with @ikeda_trashi this exciting technology!
Reach out at founders@biodelivera.com
libghostty-vt 1.0.1 released. ABI stable so its a drop-in upgrade. Biggest highlight is that we provide prebuilt Windows builds now (DLL and static libs)! Also just a good group of performance improvements and bug fixes. Enjoy the hand-written release notes below.
Windows builds here: lib.ghostty.org/vt/installing.… ("Windows")
Highlights
- Prebuilt Windows builds. Releases now include zips for x86_64 and aarch64 (Windows on Arm), each with the DLL, import and static libraries, headers, CMake and pkg-config files.
- Faster memory operations on Linux. Builds that dynamically link glibc now use its memset, memcpy and memmove, which pick the best CPU features at runtime and beat Zig's compiler_rt in every benchmark we have.
- Fixed garbled input with an AVX-512 system.
Bug Fixes
- Resizing the column count no longer leaves stale shell prompts behind when the shell redraws its prompt (OSC 133). This was most visible with full-line prompts like powerlevel10k.
- A resize that grows rows and shrinks columns at the same time now pulls scrollback down correctly, matching what you get from doing the two resizes one after the other.
- A failed render state update no longer marks unfinished rows as clean. Those rows stay dirty and are rebuilt on the next update, instead of being rendered from incomplete data (which could crash).
- Kitty graphics: during a chunked transmission that replaces an image, the old image now stays until the transmission ends, instead of disappearing on the first chunk.
Performance
- Setting a scrollback line limit now performs the same as a byte limit. We previously decommitted memory only in the line-limit case, which caused storms of page faults. With a 1,000-line limit, ASCII throughput went from 298 to 1,983 MB/s on Linux (6.7x) and from 2,214 to 4,093 MB/s on macOS (1.8x). Matches byte limit performance.
- A restored snapshot's last page now gets its full pooled capacity, so the first write after a restore doesn't immediately allocate.
- Linux builds that dynamically link glibc now use glibc's memset, memcpy and memmove (with runtime CPU feature detection) instead of Zig compiler_rt's. ASCII with no scrollback is 23% faster on arm64 and 10% faster on x86_64. Resizing is 4–10% faster.
- UTF-8 decoding after the first invalid sequence no longer decodes valid runs twice. Valid input is unchanged, and input with invalid sequences decodes 6–10% faster.
- Faster CSI parameter parsing: 14 instructions and 2 branches per digit on x86-64, down from 23 and 5. Style-heavy streams are up to 5% faster.
- Less per-segment overhead in RenderState.endUpdate.
- Validating C enums no longer scans every field.
Packaging
- Releases now include prebuilt Windows zips for x86_64 and aarch64 (*-windows-gnu).
- On Windows, pkg-config files are relocatable and aarch64-windows-gnu builds.
- Statically linked Darwin builds no longer pick up our memset over libSystem's.
Googlers have been testing and exploring Gemini 4 Carbon for the past few weeks. Here are a few of our favorites 🧵
To capture the light and motion of this 3D underwater swarm, @fofrAI used Carbon to build a custom CUDA and C++ visual effects engine from scratch.
Today we're launching @uxduckhq, backed by @ycombinator: the AI-native software for the next billion drones.
Drone production in Ukraine has increased 1000x in 4 years. Each team rebuilds the same software.
Sorry continuing to pile on. So my approach usually is:
1. What language/os/ecosystem are you most comfortable with? most expert? (Answer example: Python, picking something they feel comfortable with)
2. Okay, cool. Let's imagine this problem in Python: <technical problem>. What are the hardest parts of the problem?
3. Okay, how would you solve that?
4. Let's talk about Python itself, how do you go from code to execution in Python?
5. Are you aware of other ways to turn text in a file into executable code? (Answers something like JIT)
6. What is that? How does it work? (Answer something like: compiles to machine code)
7. Oh! Does it compile everything to machine code? No? How does it choose, in general?
8. Do JITs usually emit direct machine code or is there something in the middle?
etc etc etc.
just continue digging and digging and digging to get a broad picture of this person's technical capabilities.
Introducing Ploid Deep Research
You can now see everything
> Ask about anyone and get their full digital footprint in minutes > Photos, videos, every social account, even the ones you'd never find (like WeChat) > Works inside Claude, Codex, or any agent
Try it today at ploid.com
after seeing @huggingface’s great release of openai math as rl envs, we open-sourced 47 smaller lemma tasks decomposed from one of the same proofs (honeycomb-free-energy)
the original problems are too hard for models to get much training signal - splitting them is an easy way to get more tasks out of each big question
it got 1k+ downloads in two weeks:
Google Gemini 4 Carbon shatters physics sims! Glassy Jenga - It's no fun until you have a shard of glass hurling towards ya right? 😉
Play by 'sound' 🔊 tapping on the blocks! Just want to break stuff (yeah, me too) ? Hit the sandbox & 🔨 smash away. vitrail.andrewvoirol.com
The internet is the greatest dataset ever created.
Today we’re launching Riveter V2 to make it accessible.
Describe any dataset. @riveterhq researches the live web and returns it complete and structured, in the app, via API, or over MCP.
Reply “data” for 1,000 free credits.
Former Cisco M&A head Mike Volpi on why he’s more optimistic about AI than the dot-com era: the internet built communication, AI is building intelligence.
Guido Appenzeller: “Is there one other big difference from how it was back then, which is that everything is now much, much larger? Back then we were investing a small fraction of a percent in infrastructure for the internet. Today we’re investing 2% or so of our GDP in AI infrastructure.”
Mike: “It is much bigger, but in the 100,000 feet abstraction, what is it that is being built? Back then it was communication that was being built. Today we’re building intelligence. And the economic value of intelligence is probably higher than the economic value of communication.”
“I also think that, believe it or not, it’s been nearly 30 years, the markets are more efficient now, which is why so much more risk capital is willing to make its way into AI.”
“So net is, I’m actually more optimistic about this cycle, but I also believe that there will be some correction and some cleaning up of the industry structure.”
@appenz@mavolpi
At Cisco, Mike Volpi's team did 70+ acquisitions which became half of the company's revenue during the internet build-out. Now he's watching AI drive another infrastructure boom, and he thinks this cycle is even more compelling.
His view: the infrastructure being built will remain useful even if some of its owners change.
Hanabi Capital GP Mike Volpi joins a16z's Raghu Raghuram and Guido Appenzeller to discuss lessons from Cisco's acquisition machine, Nvidia's dominance, the opening for custom AI chips, how AI is changing data center networking, why older GPUs are holding their value, and more.
00:00 Intro
00:53 The internet buildout vs. the AI buildout
09:29 Financing the buildout, and why GPUs are holding their value
14:31 Cisco's M&A playbook, and advice for Nvidia
22:27 Making acquisitions actually integrate
25:32 When founders should sell, and how deals really happen
31:58 Mapping today's AI infrastructure
38:02 Will we see more competitive AI silicon?
43:45 Why Cisco's boom ended
46:08 Rethinking data center networking for AI
53:46 Why this is the best time to be investing
YouTube: youtu.be/LI3lxzHfRrU@mavolpi@RaghuRaghuram@appenz
Read about our collaborative efforts with @Microsoft, including how NebuSec's autonomous pipeline works alongside MDASH to identify vulnerabilities in the Linux kernel and automatically generate working exploits.
The Linux kernel has long been a major target for vulnerability research. NebuSec began scanning the Linux kernel in November 2025 and has since identified 2,386 vulnerabilities, with 170 patches merged into the mainline kernel.
being earnest is a superpower. being high integrity founder makes people want to bet on you.
a bad startup idea can turn into the a generational company. but integrity is non-negotiable.
Getting $500K from YC is a startup milestone. It can also become another badge for people who were very good at getting into college.
We asked @vivianmshen where she draws the line between founders who are larping vs founders who will figure it out.
@trq212 had one exception to the prestigious college-to-YC pipeline: Sarah.
I should add, even if you DO know, I would back it up to figure out if you just happen to know some trivia, or if you really understand it.
For example, I'd follow up with "Why does it use a VM? What other approaches could it have used? What are the pros/cons of those approaches?" etc.
When I was learning to do technical interviews, the advice given to me was always "find something specific that candidate knows with high confidence, then ask increasingly detailed whys/hows/etc. until they can't answer any more to find their boundaries."
NxtCure Labs (backed by @ycombinator) just launched! We are an AI CRO with digital twins that test cancer drugs before they reach patients. We are 28% better than even the best frontier AI models. Oncology Trials take 10+ yrs, ~3% approved, 90%+ fail post-animal. Let's get new oncology drugs through trials!
So how do your questions turn into a decision?
Instead of asking a model to describe a photo, you give it a checklist of yes/no, multiple-choice, and rating questions so it knows what to check and what action to take.
Here, Gemini Decision 1 scans 100 delivery photos in 4.5 seconds (nearly 4x faster than Gemini 3.5 Flash-Lite), answering 16 questions per photo at once:
📦 Is the box damaged? (yes/no)
🌧️ Is it under a porch? (yes/no)
⭐ What's the delivery quality? (0 to 3 rating)
⚡ What happens next? (routes to a claim review)
Gemini 4 Carbon is good at pelican SVGs, but that's nothing new for Gemini
"Make a three js animation where yonkers stands up as a big mech (cool stand up motion) and stomps around manhattan and queens and Brooklyn, use openstreetmap mapping data"
simonwillison.net/2026/Nov/10/ll…
To see how fast it works in practice, check out this interactive zookeeper game built in @GoogleAIStudio where you race the model to spot a hidden zookeeper 👀
Gemini Decision 1 takes one snapshot of the board, checks all 16 squares at once, and rapidly pinpoints the right square.
Introducing @hyperbenchAI : AI agents for hardware/software co-development.
Coding agents often lack the context to understand how a software change affects a physical system.
Hyperbench maps repositories, hardware components, interfaces, protocols, and test evidence into one system graph, giving agents the context to understand how the pieces connect.
Sign up for the beta: hyperbench [dot] ai/# private-beta
Today we are launching @Sintropix, backed by @ycombinator: The AI-native ERP that closes your books every day and files your taxes in every country you operate in.
We connect to your banks, invoicing and payroll, and handle the accounting end-to-end. Founders get one place to do their finance work: track KPIs, see consolidated P&L and ARR reports, and forecast runway.
Gemini 4 Argon greatly accelerated the development of Gemini Robotics 2.5.
It acted as a simulation engine, teaching robotics new skills that are usually dependent on massive amounts of physical data.
Gemini Robotics 2.5 is now rolling out to our robotics partners. Learn more.
An afternoon filming rubber bands, buckets, and glasses around the house (starring his cat) turned into a playable instrument when @alexanderchen used Carbon in @Antigravity to find the pitch of every clip and map them to a MIDI synth 🎹
Claude Fable 5.2 is now available on Google Cloud, bringing frontier intelligence to technical and scientific workloads, agentic coding, and multi-agent orchestration.
Try it on Google Cloud today → goo.gle/4zFp6IZ
Alessandro has an incredible story. He was originally going to be part of the P26 batch to help bring cancer drugs to market faster, when he was diagnosed with cancer himself.
He's since beat it, and is doing his life's work building NxtCure, an AI CRO using digital twins of humans to test cancer drugs before they reach patients.
Congrats on the launch @alessandrohammd!
NxtCure Labs (backed by @ycombinator) just launched! We are an AI CRO with digital twins that test cancer drugs before they reach patients. We are 28% better than even the best frontier AI models. Oncology Trials take 10+ yrs, ~3% approved, 90%+ fail post-animal. Let's get new oncology drugs through trials!
The model also adapts to lab experiments on the fly with no retraining needed.
Researchers can plug in data from just a few dozen cells to see how DNA changes behave in specific moments, like:
🔘 Immune cells responding to a vaccine
🔘 Specific cells in the gut driving inflammatory disease
🔘 Stem cells maturing into blood cells
In testing Gemini Robotics 2.5, we evaluated the model across 200 distinct tasks — such as serving food with tongs, tying a bow on a gift box, and pipetting a liquid. These tasks represent a step toward human-level dexterity by combining a number of skills, including advanced tool use and handling delicate objects with ease.
Our evaluations also show significant increases in task precision, robustness, and speed. Compared to previous baselines, Gemini Robotics 2.5 achieves a 30% to 100% higher success rate depending on the type of task, while completing tasks 20% to 30% faster.
We’re proud to announce the partnership:
@Kalshi x AnyAlpha
AnyAlpha is the agentic trading platform that turns trading ideas into AI agents.
Kalshi traders can now connect their account to AnyAlpha and let their agents trade 24/7.
Centuries of art and history come together in this interactive guide to tarot crafted by @soumyadesign with Carbon in @Antigravity.
See the cards rendered as tapestries with realistic fabric physics, hear the shuffling of the deck, and then draw three cards ✨
🔄 Move to the Copilot harness before the Local harness is removed in February 2027.
Copilot is already the default for new sessions. Look for Copilot in the harness picker, then follow the transition guide to move workflows.
