Prof. Dr. Kay Rottmann

Service · AI Agents

AI Agent Development for Companies

We build AI agents that don't just work in a demo but in the real world. From the first use-case workshop to a production-ready, measured agent - based on 15+ years of AI engineering at Meta, Bosch, and Amazon.

Development and content: Prof. Dr. Kay Rottmann

Professor of Applied AI · HdM Stuttgart · ex-Meta, Bosch, Amazon

Last updated:

Request an agent project

What is an AI agent?

An AI agent is an LLM-based system that solves multi-step tasks autonomously - it plans, calls tools (databases, APIs, other systems), reads results, corrects itself, and delivers an outcome. Unlike a simple chatbot that just responds, an agent acts. Examples: a sales agent that qualifies leads and updates CRM entries; a support agent that classifies tickets and resolves standard requests; a research agent that compares multiple sources and writes a report.

Why AI agents work in 2026 - and where they fail

With current LLMs (GPT-5, Claude 4.6, Gemini 2.5), AI agents have actually become production-ready for many tasks. They work when the task is clearly bounded, the available tools are unambiguously defined, and there's a clean evaluation mechanism. They fail when the task is poorly specified, tools behave unexpectedly, or no one measures whether the agent actually does its job. That's exactly where we focus: we build agents to be measurably good, not to look good in a demo.

How we develop an AI agent for your company

In four steps. Use-case scoping: we define exactly what the agent should do, with which data and tools, and how we measure whether it works. Prototype: a working agent in 2 to 4 weeks that solves the core task on real data. Evaluation: we build an eval framework that automatically checks whether the agent works in representative scenarios. Production: integration into your systems, monitoring, fail-safe mechanisms, handover to your team.

Which AI agents are worth building - and which aren't

Worth building: classification and routing agents (tickets, emails, requests), internal research assistants (PDF reports, knowledge bases), data-maintenance agents (CRM/ERP updates), structured sales qualification. Often not worth it: fully automated customer-service bots without human fallback (reputation risk), agents in regulated areas without clear audit trails, agents replacing already-working deterministic workflows. In the workshop phase we check exactly which category your use case falls into.

Multi-agent systems - when sensible, when not

Multi-agent systems orchestrate several specialized agents that solve a larger problem together. They make sense when different steps need different models, tools, or permissions - for example a research agent feeding a writing agent. They don't make sense when a single well-specified agent could do the same job. Our standard advice: start with a single agent. Move to multi-agent only when you concretely hit a problem one agent can't solve.

Format and investment

Typical engagement: 6 to 10 weeks from use case to production agent, depending on complexity and integrations. Investment in the low to mid five-figure euro range for a first production agent - significantly less if a prototype is enough. Includes eval framework, documentation, and handover to your team. On request we provide a concrete quote.

Frequently asked questions

What's the difference between an AI agent and a chatbot?
A chatbot answers questions. An AI agent acts: it plans, calls tools, reads results, corrects itself, and delivers an outcome. An agent can be a chatbot, but a simple chatbot is not yet an agent.
Which LLM do you use for AI agents?
It depends on the use case. For complex reasoning tasks, currently most often Claude or GPT. For high volume with simpler tasks, often open-source models (Gemma, Mistral, Qwen) for significantly lower cost. We don't earn anything on tokens, so we help with honest model selection - not every use case needs the most expensive model.
How do you measure whether an AI agent actually works?
With an eval framework that automatically checks: does the agent achieve its goal? Does it follow policies? Is it efficient? Does it behave consistently? Lessons from real production agents flow directly into this approach.
How much does developing an AI agent cost?
Typically in the low to mid five-figure euro range for a first production agent including eval framework. A pure prototype is significantly cheaper. On request we provide a concrete quote.
How long does it take to develop an AI agent?
From use-case workshop to production agent typically 6 to 10 weeks, depending on complexity and number of integrations. A first prototype on real data often within 2 to 4 weeks.
Can we maintain and extend the agent ourselves afterwards?
Yes - that's the goal. Handover to your team is part of every engagement: code walkthrough, documentation, an eval framework you can extend yourself. You're not locked in.

Let's talk.

Drop me a short note about what you're working on - I'll get back to you within a few days.

Request an agent project