Ilya Fushman

Partner @kleinerperkins, Formerly: Partner @indexventures, Product @Dropbox, Startups, Physics.

[Ilya Fushman @ilyaf]

Thrilled to announce KP22, our twenty-second venture fund with $1 billion to back early stage companies, along with $2.5 billion in growth funds to back high-inflection, category-defining businesses — $3.5 billion in total.

AI is reshaping every industry from the ground up, faster than anything seen before. This is the moment to build.

Grateful to our team, and the founders and LPs who’ve trusted us over the years, and looking forward to partnering with the next generation of history-making companies.

https://t.co/FVhpTE0yOy

[Mamoon Hamid @mamoonha]

We couldn’t be more excited to be working with @bneyshabur @HarshMeh1a @shayan_ @tararezaeikh and the superstar team @mirendil to be pushing the frontier of AI R&D. Congrats on an amazing run. The best is yet to come!

[Ilya Fushman @ilyaf]

AI is making @Rippling's compound software strategy more powerful than ever. Great conversation with @parkerconrad on building a compound AI company and where Rippling AI is headed.

[Kleiner Perkins @kleinerperkins]

Rippling is on a roll. The company added more than $150M in net new ARR in Q3, with Rippling AI playing a significant role in that growth.

For the first episode of our new series, Leaders, we sat down with @parkerconrad to talk about what's next for Rippling AI and the future of enterprise software.

Parker shares why @Rippling's compound software approach is particularly well suited for AI, why getting permissions right is critical to enterprise AI, and what it could mean for the way companies operate.

[Aatish Nayak @nayakkayak]

We're excited to back Nidus in their $30M raise to deploy embodied intelligence inside America's factories.

American manufacturing has a capacity crisis. The path to overcoming it is modernizing the swath of existing factories across our country, not only building new ones. Nidus combines frontier models & hardware to enable traditional industrial machines to perform precision assembly tasks.

@jsealynyc @harrisonnguyen @DwivC and @CanePunma are some of the few who have extensive experience actually deploying AI inside the defense industrial base. The only way physical AI has impact on the economy is by doing the hard, gritty work of making it useful in our most critical environments.

And they've already done it, just a few months after starting.

[Ilya Fushman @ilyaf]

RJ brings a unique perspective to physical AI, having spent years building @Rivian and now applying those lessons to @Mind_Robotics. Great discussion with @RJScaringe and Vidya on our Builders pod.

[Kleiner Perkins @kleinerperkins]

Physical AI is moving quickly, with advances in models, hardware, and computing opening up new possibilities for robotics and autonomous systems.

In our latest episode of Builders, @josh_coyne and @LM_Braswell sit down with @RJScaringe, founder and CEO of @Rivian and @Mind_Robotics, and Rivian SVP Vidya Rajagopalan.

RJ shares why he started Mind Robotics to bring more capable robots into manufacturing, building on years of experience scaling Rivian. The conversation covers the challenges of building hardware at scale, the importance of vertical integration, and what it will take for physical AI to deliver real value.

Watch the full conversation ⬇️
https://t.co/ste3IF2JS2

[Dino Mavrookas @MavrookasD]

Another hull in the water! This is Mirage - our 52 ft autonomous vessel. Built in Austin, launched in Galveston, TX this week.

4 years ago our initial test platform was an $800 plastic fishing boat off the internet with $30k of cameras and batteries strapped to it. We’ve come a hell of a long way but lots left to do. Right now China can build 23M gross tons of ships per year. US can do 100K. That's a big gap but with Corsair, Mirage and Marauder all in production, we’re working to close it.

None of this happens without our team who builds these hulls and the crew that gets them safely in the water. Great work by the @Saronic team this week.

[Ilya Fushman @ilyaf]

@jackdent @Roche 🔥

[Dino Mavrookas @MavrookasD]

Proud to report that the 2nd Marauder is in the water and running under its own power out of our yard in Franklin, Louisiana. The team really stepped up across the board - this hull took 8% fewer labor hours than the 1st. That’ll continue to get better, every build needs to be more efficient than the last.

When we took over this shipyard it was shutting down and capacity was trending towards 0.

Today it's is active again + expanding. Fully ramped we’ll do 20 Marauders a year in Franklin.

Thanks to everyone @Saronic for their hard work getting this done.

[Siva Gurumurthy @sgurumur]

It’s quite clear that models have surpassed human intelligence, and that makes it very hard for humans to accept and evaluate the outputs of these models. It’s like an L4 engineer writing feedback for an L8 engineer. In such circumstances, building an organization with low ego, a deeply technical bench, but first-principled humans who put customers ahead of their personal milestones becomes more important, in addition to investing in autoresearch systems, continuously running cloud agent systems, and evals.

[Ilya Fushman @ilyaf]

🔥 @krandiash @_albertgu and team!

[Cartesia @cartesia]

Introducing Cartesia Managed Agents: our end-to-end voice agent platform for enterprise.

#1 on @ArtificialAnlys Speech Agent Arena and @ServiceNow EVA-Bench, built by co-designing our voice models and harness as one system.

Check it out → https://t.co/7kJhCYqjQo

[Harmonic @HarmonicMath]

Fifty years of C and we're still shipping buffer overflows.

Aristotle proves your C does what you said it should. When it doesn't, you get the input that breaks it.

New sign-ups get $500 in credits. Get started with the most effective way to protect your enterprise:

https://t.co/WqAcppy0Hh

[Gabe Pereyra @gabepereyra]

Excited to welcome @piyprah @atishdassarma and the rest of the @get_aida team to Harvey.

Piyush co-founded Glean and we met a couple years ago when he was starting Aida.

At Aida, they were building Harvey for sales and a lot of the product and infrastructure they were building was things we had been planning.

He became someone I regularly turned to for advice when I was thinking through enterprise search and related technical strategies.

A couple months ago when we decided to focus more on these efforts they were the first team we turned to to help lead these.

Together we're going to build each firm its own ontology: a private context layer that connects all of its data and turns it into IP the firm owns.

That context makes every agent and model and product the firm uses better and stays private and portable through our APIs and will be able to run on-prem for firms that want full control.

A firm's context is its edge and the firms that own theirs will be the ones that stand apart.

[Karan Goel @krandiash]

We’ve seen incredible results in applying new ideas that bring tokenization into the main model, a culmination of 2 years of research — up to 10x compute and data gains on language models that we’ve trained!

What’s most exciting is that these hierarchical networks open up a new design space for building models that can be optimized end to end on raw data, something we don’t yet do very much of in AI.

This work is particularly exciting for us @cartesia, and a core part of our long term research agenda in building models that are fundamentally more intelligent at understanding diverse and multimodal data.

[Ilya Fushman @ilyaf]

@KaranKunjur 👏

[Ilya Fushman @ilyaf]

🔥

[Neel Kunjur @neelkunjur]

Our propulsion team is cooking with gas! (Krypton gas in fact)

20.6kW achieved on orbit on our Gravitas mission. This concludes demonstration of the full operational envelope of our in house hall thruster system on orbit - the highest power ever fired in space by 4x! The Gravitas mission continues to add to the record books.

Every propulsion system we build from here will be even better than the last, starting with our new 2-axis gimbaled system shown below launching on our next mission. These systems will eventually propel our spacecraft to Mars and beyond.

[Harmonic @HarmonicMath]

Tensordyne gave Aristotle its hardest block to verify: 12.2 million gate equivalents. 185 properties proved, 56 refuted.

Jonathan Ko, their verification engineer, calls it "unthinkable with traditional tools."

Read the the full @TensordyneInc case study: https://t.co/oXvSz32e69

[Gabe Pereyra @gabepereyra]

Proud to share @harvey has achieved FedRAMP Class D certification.

FedRAMP Class D is one of the most rigorous tiers of the program used to vet cloud software and required to support federal agencies.

Getting there meant rebuilding Harvey to run entirely inside an isolated environment where every vendor touching federal data must be FedRAMP certified or self-hosted.

We rebuilt auth, moved to FIPS-hardened images and stood up a separate gov CI/CD and observability stack. Over 1k PRs from nearly every eng team to get all nine product suites passing end to end.

That work makes Harvey stronger for the most regulated organizations we serve both in how we secure their work and in how fully they can own their own intelligence.

Amazing work by our CISO, Brooks, and our VP of Infrastructure @sridhartreddy and both of their teams.

[Satnam Singh @satnam6502]

Formally verify SystemVerilog and SVA (or autoforamlized properties).
Prove memory safety about your C programs.
Prove a C model is equivalent to its SystemVerilog implementation.
Our AI theorem prover is now publicly available.
Try it out at https://t.co/MnPxSjXZnK

[Ilya Fushman @ilyaf]

The future of software is verifiable software

[Harmonic @HarmonicMath]

We’re bringing Aristotle to code.

AI is writing more code than anyone can review. We're building the AI that verifies it.

Starting today with C and Verilog.

[Tudor Achim @tachim]

Starting today, Aristotle can verify C code including ~~memory safety~~, and control- and data-path properties of Verilog designs! We’ve started with bytes and electrons, and this is just the first step on our path to making all software and hardware verified by default.

[Harmonic @HarmonicMath]

We’re bringing Aristotle to code.

AI is writing more code than anyone can review. We're building the AI that verifies it.

Starting today with C and Verilog.