
From a year ago. I still think this is underemphasized by coding agent companies....

Partner @upfrontvc investing in AI + dev tools. Sold Hyper Travel to @tradeshift. ex @venmo, @usemacro, YC S2010.

From a year ago. I still think this is underemphasized by coding agent companies....

Our portfolio co @BrightAI_corp announced another $97M in capital to solve our critical infrastructure problems. If we've spoken in the last two years about my investing interests, you've heard me talk about the growing issues in managing critical infrastructure. I've probably rambled on about the observability gaps and how solving them sets the foundation for human and machine interventions in the field. We've seen these opportunities in power distribution, water pipelines, oil and gas, waste management, and many other industries that we depend on. Lack of observability in our critical infrastructure leads to more system outages than necessary. Maintenance takes longer and costs more. And the companies that own and manage the infrastructure rely heavily on deployed field teams to perform simple tasks. Costs, of course, get passed on to consumers in the form of higher utility bills and outages. BrightAI has been supporting some of the world's largest asset owners and service companies with their Stateful platform and autonomous inspection robots. Physical AI can mean many things, but every definition includes a perception layer that takes in novel data, reasoning models that describe and engage with the physical world, and autonomous systems. BrightAI has built a platform to deploy all three. fortune.com/2026/11/18/brigh…

My father’s lifetime dream is to play golf at St Andrew’s at the Old Course. We are headed to Scotland in May. Any chance anyone has any tips for getting a tee time?

Pretty obvious that MacOS is going to be an important platform for agent workloads. This is a big release. Well done @ivanburazin @JukicVedran and team!
Your agents can finally use a Mac. macOS Sandboxes are now live on @daytonaio: M4, M4 Pro and M5 Pro. Spin one up at app.daytona.com

The best models are a function of their data. Yet most ML teams still manage trillion-row video and sensor datasets with tools built for spreadsheets. We’re proud to partner with LanceDB for their Series B.
The next generation of models needs a data stack built for AI. We just raised a $57.5M Series B to help researchers turn petabytes of video, images, and audio into better models, faster. Read the announcement: lancedb.com/blog/lancedb-ser…

no one ever seems to ask "what if we put a billion dollars into simulators instead of data collection"

2 more speakers confirmed for our SF event on Dec 4. Full list of speakers - Bill Poucher (@poucherw) - President of ICPC - Alex Mashrabov (@alexmashrabov) - Founder & CEO of Higgsfield. 2x ACM ICPC World Finalist. - Nick Davidov (@Nick_Davidov) - Founder, DVC. Seed investor in 7 unicorns - Nikunj Kothari (@nikunj) - Partner at FPV Ventures - Peter Zakin (@pzakin) - Partner at Upfront Ventures 100+ IOI/IMO/ICPC medalists, founders, AI researchers and investors in one room. Keynote presentations, "What? Where? When?" (intellectual team game), poker tournament, YPO-style brainstorms. See you all there! Apply to attend: luma.com/tm0qrjtb

It’s been such a privilege for me and @gimmedadeetz to partner with @archiemckenzie_ . Archie is a world class founder. From day 0 he has been unwavering in his conviction that GT could become essential infra company — that he would do for localization what Stripe has done for payments. He’s also a lightning rod for talent. I see this not only in the team he’s assembled, but also in the early set of customers he’s assembled. He attracts the very best in service of the mission of making software international by default. Congrats, Archie and team!
We led @generaltxn seed when it was just @archiemckenzie_ .. These days I get to work out of their office & spend time with the team every day. I’ve watched them build something special, pumped for this team!!



“Model weights and datasets” were stolen? Wow, I wish there was a company protecting them… Join @enclosure_inc to protect AI models from theft, sabotage and escape: enclosure.inc/apply or dm me here on X
Fireworks has been hacked.

I'm always surprised by how mis-read the bitter lesson is. A common misreading I see is that the bitter lesson is that "the naive approach always wins in the end". This is absolutely not the bitter lesson! The bitter lesson is that the *approach that scales most effectively with compute* wins out in the end. Like, imagine you have a bunch of functions and you're trying to figure out which increases the fastest as x -> infinity. One scales linearly with x, another exponentially, and another factorially with x. The bitter lesson is saying- forget how big each function is when x is small, forget how much each function scales with things besides x (e.g. human domain knowledge), the thing that matters most is which function scales hardest with x. The bitter lesson is not necessarily anti- complex approaches. NLP is quite complex, but LLMs are also hilariously complex- especially the modern variants where we've added on many clever but complicated optimizations like DSA.

A rough distillation of my investment appetite: what infra is needed to help agents make better decisions. Sensors and sims.
Very cool simulation use case. Nubank iterated/ released agents faster/better by testing on simulated vs. real customers. All companies will need to do this. Most won't be able to hire the talent. Expect "Simulation as a Service" demand to 🚀

Amazing to see how far @minafahmi and @hokira have come in the last 3 years. Will always be proud that this was the first investment I worked on when I joined Upfront. cc @kobiefuller
3 years ago GPT-4 arrived. Voice was for smart speakers. Wearables were for sleep. And we wanted a different relationship with technology. Today we're a small team in NYC with customers we're excited to serve. Next week we begin shipping Stream, an extension of your thinking

3 years ago GPT-4 arrived. Voice was for smart speakers. Wearables were for sleep. And we wanted a different relationship with technology. Today we're a small team in NYC with customers we're excited to serve. Next week we begin shipping Stream, an extension of your thinking

We just crossed $100M in run rate. Four quarters ago, we were at $1M. Then this year alone, we grew 37.5x. Not long before that, we did a hard pivot- handed our old customers to competitors, rebuilt from scratch, and relaunched with a few months of runway left. The bet was: models think on GPUs but work happens on CPUs. So every agent that writes code, opens a browser, or runs an RL job needs a composable computer that's isolated, stateful, and ready in milliseconds. $1M to $100M in about a year puts us in the same cohort as the likes of Anthropic and Cursor. It’s humbling to share that lane with companies we’ve learned so much from.

Hacktober is going to create a lot of FUD around new SaaS products. We need a new kind of certification that software is safe to use. Huge opportunity to build that right now and displace SOC 2, which has been a bit of a polite fiction for quite some time now. We need whitehat agent swarms with a certification on top. As a buyer living in Black October, I want to be able to answer questions like: "is this vendor going to get hacked by an agent swarm tomorrow?" A SOC 2 report does not remotely answer this question. But I would trust a document from a trusted tester that says: "we threw a sophisticated agent swarm at this vendor for a week, trying everything from hacking their infra to phishing their employees, here's how they did." Give me a pentest for your whole org. One that simulates the threat we're waking up to, and evolves as the threats evolve.