Specialist in computer use agents, H Company aims to automate complex workflows for businesses. Interview with its CEO to understand the strategy and ambitions of the French scale-up.
JDN. H Company – or H – claims to specialize in computer use agents. What concretely differentiates this approach from current AI agents? ?
There was a first generation of AI with LLMs, used in the form of chatbots. Then a second generation with integrations via API, RAG or data, which many companies today call agents. These approaches are effective for connecting a few applications together, but a vast majority of Fortune 500 companies’ software lacks comprehensive APIs. With the computer use models that we develop at H Company, as soon as a human can use an interface, our agents can operate on it. They can interact with the graphical interfaces of web, desktop or even legacy tools from the 90s, and execute complete workflows like a human. It is therefore a new generation of agents, designed for the complex environments of large companies, allowing them to operate on tools such as SAP, Salesforce or even old internal software.
You joined the French scale up H in June 2025. What have been the important strategic decisions taken since your arrival? ?
H was founded in 2024. The first phase was very research oriented with the development of the first Holo models by the founding team. Laurent Siffre, co-founder of H and former researcher at DeepMind, where he notably contributed to AlphaGo, developed the Holo 1 and Holo 1.5 models with his team. There was no marketing or real commercial product then. The priority was to move from a research project to a company capable of deploying its solutions to major accounts. We recruited experienced profiles, particularly from Palantir, where I spent ten years. We have structured a product team and a go-to-market strategy while adopting a forward-deployed engineers model, allowing us to work directly with client companies.
What is your main target today and for what needs?
Our goal is to address real business issues, whether growth or cost reduction, and to demonstrate tangible ROI. This is also why we appreciate being able to discuss directly with CEOs about their challenges and objectives. Our promise is focused on ROI. We want to show that, used well, AI can produce concrete results. We mainly target large companies. For our agents to be truly efficient, it is essential to understand in depth the business context, the internal tools and the vocabulary specific to each organization. This is why we adopted this forward deployed approach, with H engineers directly involved with customers during the first phases.
How does this forward deployed engineer approach constitute a differentiating lever in the deployment of your agents? ?
This allows you to iterate more quickly and resolve operational issues in hours rather than weeks. We support our clients over several months to ensure that agents operate correctly and produce a measurable impact, whether it is increased revenue or the development of a new market for example. Last year, we chose to support a limited number of 2 to 3 clients. This year we should gradually accelerate our deployment capacity as the team grows from 80 to 120 employees. This workforce should allow us to support between five and ten clients in 2026.
Last November you presented your new Holo2 model. What does it bring compared to previous generations ?
The objective with Holo2 was to combine a high level of performance with real economic efficiency for use in production. Holo2 marks an important milestone as it is our first generation of models capable of working not only on the web but also on desktop and mobile environments. We offer three versions, namely 4B, 8B and 30B-A3B, which use a mixture-of-experts architecture. This means that the model only activates a subset of its parameters, making it much more efficient, while still delivering performance comparable to a full-size model. With Holo2, our Surfer 2 agent outperforms all open source models on the main computer use benchmarks.
How does H Platform work and what role does it play in managing and deploying agents to your enterprise customers?
H Platform is designed to provide visibility into what agents are doing, define their level of autonomy and validate certain actions when necessary. A company can, for example, configure an agent so that it analyzes prices, stocks or histories on several tools, but also define that an action, for example a purchase, is proposed then validated by a human. These agents can also be used via existing interfaces like Slack. Ultimately, any employee could create and deploy agents.
What is your billing model and how do you actually measure ROI for your clients?
We offer a global license that varies depending on the size of the company, which allows all employees to access it. By subscribing to an unlimited annual license, the company can deploy as many agents as it wishes. Concerning hosting, we offer European cloud via Scaleway or American cloud via AWS. Companies that have their own servers and GPUs can also manage hosting themselves. With ROI at the heart of our promise, our goal is for you to return at least two to three times your investment, with a goal of 10 times for certain use cases, particularly commercial ones.
How is it possible to achieve such high ROI levels?
Because our agents allow you to do much more, without necessarily increasing your workforce. For example, you can use this freed up time to manage more clients or expand into new markets. Companies often seek to reproduce with an agent what they are already doing currently, while the value is mainly found in the creation of new workflows for tasks that they could not previously carry out due to lack of time or resources.
Do you have an example ?
Let’s take the example of a large company in the advertising industry that we work with. Their salespeople spent 70% of their time receiving emails, checking in Salesforce if the contact exists, creating it if necessary, writing a proposal, sending it back by email, then transferring it to an internal planning system if it is accepted, etc. We have reduced this time to 5%, essentially corresponding to the one-off validation time. Not only are these salespeople more productive, but this frees up their time to address markets that they were not targeting before, for example by canvassing small local players and no longer just large accounts.
Which international markets are you targeting as a priority?
We started in France but quickly opened up internationally. The United States represents a key market for H Company and we already have a small team in New York. At the same time, we are also targeting Europe, but also Asia and the Middle East, where we already have exchanges with several companies.
H experienced a turbulent period in the summer of 2024 with the departure of three co-founders, then that of Charles Kantor in June 2025 during VivaTech. What was the impact of these departures? ?
Our investors, such as Eric Schmidt or the Accel fund, were above all convinced by the presence of Laurent Siffre, CTO and co-founder, and by the quality of the research work carried out from the start. There remains a very solid core of engineers and researchers who have worked together since the creation of the company. From this point of view, these departures have not weakened the internal dynamic. Many talents also want to join H Company. For my part, I was working at that time in NYC at Palantir. I already knew Laurent before joining H and I was impressed by his research work. However, it was necessary to be able to market this technology and develop a go-to-market strategy. It is this challenge that motivated me to join H as CEO.
You raised $220 million in May 2024. Is a new fundraising in the pipeline? Do you plan, in the long term, to develop into more general public uses? ?
We still have good financial visibility and a new fundraising is not our priority in the short term, although we will remain open should the opportunity arise. Our model does not require the colossal amounts raised by general model providers, because we develop more specialized and lightweight models. Today, our focus is clearly on large companies and complex business processes. Ultimately, if the platform becomes sufficiently mature, we will be able to consider opening it to smaller structures such as SMEs, entrepreneurs or to more general public uses. Our agents already work very well on certain consumer use cases such as flight reservations for example. However, this is not our current priority.
Are you in discussions with the French State to develop your solution within public services?
We have exchanges with the French State, but the processes are longer and recent political changes have not made things easier. In the short term, I do not think that the State will be part of our first contracts. On the other hand, there is real potential for impact in certain sectors, notably health. Today, caregivers spend a lot of time on administrative tasks, which takes them away from their core work. Being able to automate part of these processes would have a real impact and it is a subject that is particularly close to my heart, even if we were to generate less revenue on this type of project.
Gautier Cloix is CEO of H, after ten years at Palantir, where he supported the company’s growth in Europe and then internationally. A graduate of Centrale and X-HEC Entrepreneurs, he began his career at JPMorgan in London before engaging in social entrepreneurship via the On Purpose program and Big Society Capital. He joined Palantir in 2015, founded and managed the French office, where he managed strategic projects in the public sector and with large CAC40 groups such as Airbus.
