Remanence Secures $30 Million to Train AI in Business

On November 17, 2026, Remanence announced that it had raised $30 million in seed funding, approximately €25.5 million, in a round led jointly by EQT Ventures and FirstMark. This Paris-based company aims to utilize data generated by AI agents to train models that align with the internal processes of its professional clients.
Adaptive ML is already on the field
Remanence enters a competitive landscape with Adaptive ML, whose Adaptive Engine integrates evaluation, post-training, and improvement through production usage feedback. Adaptive ML claims that AT&T has selected it to create models tailored to its data and tools. Veris AI focuses on simulating enterprise systems and converting operational traces during testing. An internal team can also perform part of this work using open tools, including ART, the training framework offered by OpenPipe.
For the buyer, the relevant measure remains the total cost of a task completed correctly. A specialized model may be less expensive to run, but the savings depend on volumes, training expenses, updates, and cases still assigned to a human or a generalist model. A documented deployment would allow for a comparison, under the same procedure, of the success rate, the cost of each successful task, and the proportion of cases requiring human intervention.
The timeframe to obtain a usable model from traces and the reusable work from one client to another will serve to evaluate the commercial offer. If specialists need to reconstruct each procedure with its exceptions, their involvement will increase the product's cost. Conversely, reusable tools for a comparable procedure would allow for the sale of a platform that the client retains as their agents evolve.
Clients will be able to choose their infrastructure
Remanence promises clients a version of the model tailored to their own procedures. They will be able to determine where to train and execute it, without having to extract their data from their infrastructure. However, the export of the model and the possibility of continuing its training will depend, among other factors, on its initial license. FirstMark envisions a hybrid organization, combining internal models for certain uses with continued access to models from major labs.
Targeted companies are already having their agents perform numerous repetitive tasks with verifiable outcomes. Support is a suitable case when precise rules govern the work. The legal field imposes more distinctions: verifying an internal procedure does not require the same criteria as assessing complex advice. Remanence is currently working with European and American design partners in these areas as well as in enterprise software.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Traces must become verifiable exercises
The post-training envisioned by Remanence aims to transform execution traces into exercises suitable for internal procedures. In support, an agent can consult an order, check the conditions for a return, and then propose a refund, leaving a history of its operations. However, simply noting that a refund has occurred does not allow for a judgment of its correctness: it is also necessary to know the applicable rule, the amount due, and the exceptions. The platform will therefore need to reproduce situations and establish validation criteria; clients could then retrain their model when their procedures change.
Co-founders Pierre-Louis Cedoz, Hamza Benchekroun, and Axel Moyal worked together at H Company on Holo, a technical experience that underpins the project. Learning to use software involves repeating tasks and controlling the outcome. For Holo, the agent had to identify relevant elements of an interface, decide on an action, and then examine its effect before proceeding. H Company had built interactive environments for this training and to measure the capabilities of its models.
Pierre-Louis Cedoz and Hamza Benchekroun are among the signatories of the technical work on Holo3, dedicated to actions within computer interfaces. Remanence also notes that the trio's backgrounds include experiences at Google DeepMind, InstaDeep, and Meta.
Remanence plans to recruit mainly in Paris
The funds will finance a dedicated set of graphics processors, or GPUs, as well as the necessary training environments and evaluation devices. The planned recruitments will focus on Paris, with initial presences anticipated in London and San Francisco. The SAS was established in Paris in September 2026. Remanence presents Pierre-Louis Cedoz as its CEO, while the legal announcement of its formation names Hamza Benchekroun as president and Axel Moyal as general director.
Alongside the funds co-leading the operation are Zero Prime, Emblem, Drysdale, Global Founders Capital, and Yellow. FirstMark had previously invested in the $220 million round announced at the launch of H Company in May 2024. Remanence also cites individual investors including Olivier Pomel from Datadog, Philippe Corrot from Mirakl, Romain Niccoli from Pigment, Victor Riparbelli and Steffen Tjerrild from Synthesia, as well as Mehdi Ghissassi from AI71.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.