Behind the Engine: How Carl.AI Scales Knowledge Access at Mercedes-Benz.io

At Mercedes-Benz.io, not every product we build is visible to customers. Some of the most important ones work quietly behind the scenes, shaping how teams collaborate, share knowledge, and build digital products more effectively, Carl.AI is one of those products.

Carl.AI is part of our internal tooling landscape, positioned earlier in the value chain. Its impact, however, reaches far beyond that, by helping teams find the right knowledge faster and reducing the time spent chasing answers, improving how work gets done across the organisation and that improvement flows downstream into everything we build.

This tool did not begin as an "AI for AI's sake" initiative but it emerged from a very real operational challenge: recurring questions, fragmented knowledge, and too much reliance on interrupting the same people for answers that already existed somewhere.

The problem it set out to solve

Across teams, a familiar pattern kept repeating. Support questions around platform topics, internal processes, and domain‑specific knowledge were being asked again and again. In many cases, the answers already existed, buried in documentation, old message threads, or previous discussions, but they were difficult to locate when needed.

This created friction on both sides. People asking questions lost time searching or waiting for responses, while subject‑matter experts were repeatedly interrupted with the same requests. Knowledge existed, but access to it did not scale.

Carl.AI was created to address exactly that gap.

How Carl.AI started

The first version of Carl.AI was intentionally small and practical. It was not designed to be a large platform from day one. The initial goal was simple: reduce the pressure on people answering the same questions repeatedly and make existing knowledge easier to access.

Once this approach proved effective, it became clear that the problem was not limited to a single team. Many teams faced similar challenges with support load and scattered knowledge within their own domains, and that insight marked a shift in direction.

Carl.AI evolved from a helpful internal bot into a broader internal AI platform, becoming a foundation teams could build on themselves.

What Carl.AI is today

Today, Carl.AI is much closer to a self‑service workspace than a single assistant.

Teams can create assistants for specific domains, connect their own knowledge bases, and bring in data from the tools they already use. The platform is accessible not only through conversational interfaces, but also via APIs and more developer‑oriented workflows that integrate naturally into existing ways of working.

This adjustment allows teams to shape Carl.AI around their own context, personalising their use according to their needs.

Why Carl.AI matters

At its core, Carl.AI exists because support does not scale when knowledge access depends on knowing exactly who to ask. When reliable answers require messaging the right expert and waiting for context, bottlenecks form without intention. Work slows down, interruptions increase, and valuable knowledge becomes underutilised.

The purpose of Carl.AI is not to "add AI" to the organisation but to remove friction. By shortening the distance between a question and a reliable answer, Carl.AI helps teams stay focused, reduces context switching, and enables greater autonomy. In that sense, it is as much an enablement product as it is an AI product because when it works well, people spend less time hunting for information and more time building.

What makes Carl.AI technically interesting

Carl.AI does not rely on a single standout feature. What makes it interesting is the combination of accessibility and integration.

Non‑experts can create useful and knowledge‑aware assistants without writing code or building custom pipelines. Self‑service is a key principle: teams define an assistant, connect relevant knowledge, and put something practical into use quickly.

At the same time, Carl.AI is designed to feel natural for developers. With knowledge bases, data-source integrations, and an OpenAI-compatible API, it connects to existing tools and workflows instead of operating as a separate system.

The platform is also moving beyond answering questions. Code sandbox workflows and artifact style outputs support more hands-on use cases, helping teams generate prototypes, structured assets, and working outputs directly from conversations and the move towards actionable outcomes is a central part of Carl.AI's evolution.

What's next

The next chapter for Carl.AI focuses on widening the platform surface without losing the simplicity that made it useful in the first place.

If the first phase was about reducing repetitive support, the next phase is about scaling autonomy. The north star remains unchanged: fewer interruptions, fewer context switches, and more time spent building. By continuing to make internal knowledge accessible, practical, and actionable, Carl.AI aims to quietly improve how work gets done across the organisation.