“Our customers said, couldn’t you offer something at the same level of product quality that we’ve seen from you on language? We just can’t not answer that,” DeepL CEO Jarek Kutylowski told CNBC yesterday, explaining why the company is now building a general-purpose AI agent.
In most cases, listening to customers is sound practice. It’s how companies refine products and build loyalty. However, these customers aren’t asking DeepL to improve translation or writing: the areas where it has built real trust. They’re asking for an entirely new kind of product, plunging the company into a more chaotic market. Sometimes the right answer to a customer is “no.”
DeepL’s strength has always been clarity of focus. In a crowded AI field, it carved out an enviable niche: enterprise-grade translation with unmatched accuracy. It serves over 100,000 business customers, including Coursera, Deutsche Bahn, and Nikkei. A commissioned study by Forrester Consulting found DeepL delivered a 345% ROI, reduced translation time by 90%, and halved workloads over three years. That kind of impact comes from obsessive specialization.
By moving into the broad, still-messy market for “agentic AI,” DeepL risks diluting that which made it valuable.
Kutylowski insists this isn’t reactive. “I think it’s more of a reflection on the fact that we can do more and have more impact,” he told CNBC, describing the new agent as a way to expand DeepL’s value beyond language.
General-purpose agents, software that mimics a human using a computer, are suddenly everywhere. Anthropic has “Claude with computer use”. Microsoft, Copilot across Office and Windows. OpenAI’s GPTs can navigate browsers and APIs. These incumbents already have distribution and deep pockets. Entering that fight as a $2 billion German unicorn is brave, to say the least.
Kutylowski acknowledges the risk himself. “It’s a very, very dynamically evolving market, and to some extent, maybe even scarily so,” he says. “Everybody in this market will have to stay very, very sharp”. Translation: the moat that once protected DeepL doesn’t exist in general-purpose AI. What matters there is scale, partnerships, and speed: all areas where Microsoft, OpenAI, and Anthropic already dominate.
There’s also the adoption problem. Enterprises are still struggling to find ROI from broad AI deployments, with many projects remaining experimental and results uneven. Agents sound promising, but they are error-prone and hard to trust unsupervised, especially for a company that built its reputation on reliability.
Which brings us back to the missed opportunity. If any company should be building a narrow agent, it’s DeepL. The enterprise spends billions on localization, multilingual customer support, and compliance-heavy documentation. These are workflows where accuracy matters more than novelty, and where DeepL already has credibility. Tools like DeepL Write and DeepL Voice show the path: own the “language layer” of the enterprise. That could have been extended into a “language-ops agent”, that translates, rewrites, and routes content across global systems.
Instead, DeepL is now pitching an assistant that can automate invoices, research accounts, or tidy up marketing slides. And it’s hard to see why a procurement officer would pick DeepL over Microsoft Copilot bundled with Office, or Anthropic, raising funds at staggering valuations.
The irony is that DeepL already has what its rivals don’t: deep customer trust in a mission-critical niche. According to ALC, DeepL is the most-used machine translation provider among language service companies, with usage at 82% compared to 46% for Google and 32% for Microsoft.
To be fair, Kutylowski argues DeepL can handle both. “Our margins are great. I’m not so super worried about that,” he said, pointing to the company’s ability to make compute more efficient over time. And perhaps customers will welcome a general-purpose agent that inherits the company’s translation DNA.
History in tech is full of cautionary tales. Slack tried to go broad as a “work OS,” only to get boxed in by Microsoft Teams. Zoom branched into contact centers and collaboration suites, but the world still sees it as a video app. Focus is both branding and strategy.
DeepL is one of Europe’s most respected AI companies. Kutylowski wants DeepL to be seen as a general AI innovator. The danger is that by going broad, DeepL could end up being seen as just another player in an already bloated space.
Counterpoint
There’s another way to read this pivot. Perhaps DeepL isn’t so much chasing opportunity as defending against encroachment. Translation is no longer a walled garden: ChatGPT, Google Gemini, and others already handle multilingual text as part of broader platforms. If translation is destined to be absorbed into the toolkits of giants, then DeepL’s expansion into agents may be more about survival, hedging against losing the one moat it has relied on.
