Poolside

Poolside is in Rio this week for ICLR 2026! 🇧🇷 Our research and talent team will be on the ground all week — come say hi. We'd love to meet the community and hear what you're working on at the frontier. We're hiring actively across: → Synthetic Data → Post-training → Scalability → Pre-training / CUDA → Pre-training / Data Research → Evals If you're building in any of these areas and curious about what's next, check out our career page → https://lnkd.in/eQVCWbz4 Poolside on the ground: Robert McHardy George Grigorev Marah Abdin Polina Naimushina Arkadii Bessonov Max Wegman #ICLR2026 #ICLR

Diving into the topics shaping the “next frontier”, our co-founder and co-CEO Jason Warner was in Washington DC this week for the SEMAFOR World Economy Summit. Joining co-panellist Renen Hallak of VAST Data and moderator Morgan Chalfant, the conversation covered a lot of ground - from full-stack AI development and the real-world constraints on building agentic systems, to what the GPU supply chain looks like under geopolitical pressure, and how companies are actually navigating workforce change as AI becomes more capable. The one takeaway? Turning intelligence into action means nothing if your models can't operate where the stakes are highest. Thanks for having us, Semafor!

Training AI models requires storing large amounts of temporary data mid-process; this data is needed later but takes up expensive GPU memory while it waits. The standard workaround has always been to delete it and redo the work when the time comes. How inefficient. NVIDIA's super chips change that. The connection between the GPU and CPU is now fast enough to move that data off the GPU and pull it back when needed, without the GPU ever having to wait. Our team put this to work inside our Model Factory—the training infrastructure we've built to iterate on models as fast as possible—and the result is a meaningful throughput improvement at almost no cost. It's the kind of engineering detail that doesn't make headlines, but compounds significantly over a long training run. We wrote it up in a new blog with NVIDIA (link in comments) worth a read if you work on model training.

Most teams are using AI to write code faster. The more interesting question is whether they're building better software. Eiso Kant is speaking at NVIDIA GTC tomorrow, exploring what it looks like when AI is embedded across the entire development lifecycle, from architecture and design through to deployment and continuous improvement. Find out more and add to your GTC schedule here: https://lnkd.in/e2JYepRT

We’re at DAIC Connect 2026 in London today. If you’re attending, let us know. Vítor Monteiro and Nigel Abbott would love to connect. Government, MOD, and defence industry leaders in one place working through what responsible AI adoption looks like in practice for UK Defence. We’re looking forward to these extremely important and relevant conversations.

We are partnering with Atos to deliver sovereign AI solutions to enterprises across Europe and beyond. This is built for organizations that need frontier AI on their own terms, respecting data sovereignty, integrating deeply with existing operations, and flexing across models, deployment environments and infrastructure. Powerful, private, and purpose-built for the organizations shaping our digital future. Read more here: https://lnkd.in/dbQWaHt9

How do you move fast enough to stay ahead without leaving your people behind? Our co-CEO Eiso Kant sat down with Ericson Chan , Kate Smaje, and Willem Marx at The Wall Street Journal House in Davos to unpack that question. The conversation went beyond the usual AI hype to talk about what real transformation looks like and what it demands from leaders. Here's what stood out: Software developers are the canary in the coal mine. What's happening now in engineering—10x productivity jumps, entire processes collapsing into models—will ripple through every knowledge function. The question isn't whether this affects your organization, but how soon. Hire for high agency, not just skill sets. The organizational structures that work today won't work in five years. The constant? People who can navigate complexity and hit objectives regardless of how the tools around them evolve. We're living in the in-between. AI is useful, but it's not yet autonomous. That changes in the next few years, and leaders need to start preparing now even if they don’t have a clear playbook already. We're seeing this transformation up close with customers in defense and government, where AI isn't just a productivity tool. It's mission-critical infrastructure that needs to work securely with proper safeguards in place. Photo by Maurizio Martorana for Dow Jones

Our Head of People, Beatriz Lourenço, was recently featured on The Breakthrough Hiring Show, sharing the story of scaling Poolside from 4 to 150+ employees in just over two years. Beatriz discusses how we've built a distributed R&D organization that attracts exceptional researchers globally, and why building the right team from day one has been critical to our mission. As we continue building toward AGI through software, we're actively hiring for critical roles, with particular focus on data pre-training and agentic capabilities. Interested in joining the mission? Explore open roles on our careers page: https://lnkd.in/eQVCWbz4 Listen to the full episode: https://lnkd.in/ehvks3h8

We're in Davos this week! Eiso Kant, Noah Smith, Margarida Garcia, and Philip Drury are on the ground at the World Economic Forum's Annual Meeting 2026 connecting with customers and leaders across AI, infrastructure, and public policy—continuing conversations about how frontier AI transforms from compute into intelligence. If you're attending and want to connect, drop a comment below!

Our Co-Founder and Co-CEO Eiso Kant explains our path to AGI and how we're now in AI's "awkward teenage years." Watch the full session at the AI Engineer Code Summit in New York: https://lnkd.in/eRUdsvh8