Building Foundational Models to Generate Actions: Our Partnership with the H company

Over the last 18 months, GenAI has been firmly in the spotlight and its dominance of headlines and conversations shows no signs of waning. Last year, GenAI investment hit record levels, with startups attracting $22.4B. With all of the recent headlines and buzz, it’s easy to forget that AI has been around for decades and we’ve long followed its evolution. In 2011, we launched our big data fund as we started to see early efforts to aggregate data at scale in order to extract value from it. This led to the rapid progress in machine learning from 2014 through to 2018/19 and we’re now in the new wave of AI, shifting from being predictive to generative. We believe the new models now being developed will underpin a transformation as significant as the shift to cloud and mobile and form a part of every piece of software being written.

Over the years, we’ve invested in a number of companies providing the tools to leverage AI and have been fortunate to partner with, support and learn from leaders at the forefront of AI. Ada, AssemblyAI, Celonis, Scale, Synthesia and UiPath are just some examples.

One of the areas we’ve focused on in particular has been the development of specific AI models for enterprise automation: models capable of automating tasks and processes. These models would require larger context windows and some form of reasoning and task prioritization: more than current LLMs can offer. We started to look for a team who wanted to address these challenges and have discovered an exceptional team in Europe. Paris and London have proved very fertile ground for fundamental AI research and these cities have been at the heart of cutting edge AI developments for years.

Thanks to its excellent mathematical, computer science and engineering schools - École Polytechnique and École Normale Supérieure - and research institutions, France in particular is producing a world-class AI talent pool. This is showcased by the fact that Paris has been a magnet for US tech giants for almost a decade. Meta established one of its FAIR (Fundamental AI Research) hubs - led by Yann LeCun - in Paris back in 2015, Google then followed in 2018 when it launched its AI research center and more recently announced the opening of a new AI hub for 300 researchers and engineers. On the London side, DeepMind has been at the center of many AI breakthroughs for more than a decade, including the first deep neural network capable of beating the world champion at Go.<sup>1<sup>

The H Company: Building Agentic Models for Enterprise Automation and Beyond

Today, we’re thrilled to announce our partnership with the H company.

H’s founding team brings together Charles Kantor (Chief Executive Officer) and a world-class team of highly experienced former DeepMind researchers and scientists: Karl Tuyls (Chief Research Operations), Laurent Sifre (Chief Technological Officer), Daan Wierstra (joining soon as Chief Scientist), and Julien Perolat (Chief Multi-Agent Officer) to develop agentic models, expanding the capabilities of enterprise automation and beyond, and scaling towards their ultimate vision of achieving Artificial General Intelligence (AGI).

After just a few deep-dive meetings, we knew we wanted to partner with them. 

• Charles, previously at Stanford, co-founded H towards the end of 2023, developing its vision and strategy
• Karl, a renowned scientist in the multi-agent systems community, led DeepMind’s game theory and multi-agent teams from 2017 onwards. He started and led well- known projects such as DeepNash, an autonomous agent that defeated human expert players at Stratego, and TacticAI, an automated assistant-football coach for corner-kicks, published in Science and Nature
• Laurent spent a decade at DeepMind, contributing to and leading several key research projects in GenAI and deep neural networks. These included AlphaGo, AphaFold, AlphaStar, Chinchilla, Gemini and Gemma
• Joining the H company in a few months, Daan is a founding researcher at DeepMind and a pioneer in reinforcement learning techniques. At DeepMind, Daan led a 100+ team for many years and played a pivotal role in establishing Deepmind’s initial research direction
• Julien, an expert in multi-agent reinforcement learning and game theory, co-led the scientific and technical development of DeepMind’s work on the game of Stratego (DeepNash) and many other foundational works on topics such as mean field games, and population based learning.

We believe that this team possesses a unique body of knowledge well suited to go after these new types of agentic models and further their mission of realizing AGI.

We’re thrilled to be leading this round alongside strategic and institutional partners UiPath, Inc. (“UiPath”) and Daniel Dines, Eric Schmidt, our long-time co-investor BPI, Xavier Niel, Yuri Milner and Amazon. Daniel’s entrepreneurial experience will be invaluable to the team and the close partnership with UiPath should ensure a large distribution channel for the new agentic models that the company will develop. UiPath is already delivering AI models through its enterprise AI and automation platform and this partnership will give its customers even more automation capabilities.  

We’re also looking forward to partnering with a number of other investors and friends for the H company’s journey, including Bernard Arnault’s Aglaé Ventures, Creandum, Eurazeo, FirstMark, Motier Ventures, Samsung Ventures, and Visionaries Club among others.

We’re incredibly excited to partner with the H team and can’t wait to see what this new team will deliver!

The Accel team