DeepL was for years that atypical German challenger: an inconspicuous office in Cologne, almost no marketing, one function understood by every CIO and every location-based company on earth – translations better than Google. On this one pillar, it has managed to build a company worth $2bn. (May 2024 round valuation), more than 200,000 B2B customers and a stock market candidate position.
Now DeepL is trying a second trick: it wants to go beyond text and into the core workflows of companies. This is where the real ring with Microsoft, Google and OpenAI begins. DeepL showed a new product in Berlin: DeepL Agent. In a nutshell, it’s an autonomous AI agent that’s supposed to act not like an API, but like a virtual user supporting business applications, including those that aren’t in standard integration catalogues. In the company’s material: exporting insights for sales, OCR/invoice automation in finance, connecting niche applications and proprietary systems.
This is not just another chatbot for the enterprise. It’s an attempt to leap into an area that has so far been dominated by Microsoft Copilot, OpenAI with integrations in ChatGPT Team/Enterprise and SAP, which has ambitions to build its own agent ecosystem around S/4HANA.
DeepL has one advantage: product discipline. Over the years they have not been tempted by everything for everyone and have built selectively. Customisation Hub was the first step towards bringing AI into existing corporate language processes – glossaries, style guides, corporate context. Agent is an extension of this thinking: more automation close to the processes.
But the question is: can DeepL monetise automation as effectively as it monetises high-quality machine translation? The LLM/AI agent market is only just building, but it is already clear that margins will not be as high as in enterprise MT – because here Microsoft sells the same promise in Copilots and has a gigantic distribution advantage (M365). On the other hand – companies want supplier diversity. Especially in Europe – where the argument of jurisdiction and data sovereignty is still a real selling point.
If DeepL wants to do an IPO in 2025-2026, it needs a new growth story. Translations may be a growing, but not a pioneering segment. Agents – they will be at the peak of popularity in 2026 and then probably be in the practical phase (implementations into processes, not in marketing demos).
DeepL has the brand, it has the technology, it has the customers. Now it needs to prove that its agent is not just a nice demo from an event in Berlin, but a tool that can realistically take work from people – without rewriting the whole organisation along the lines of Microsoft or SAP. If it succeeds – DeepL will indeed enter the league of those with whom it has so far only struggled for translation quality. If not – it will become a premium translator-as-a-service.
