Autonomous workflows that make decisions and get work done.
In short: I design, build, and operate agentic AI systems for Swiss companies. I currently lead agentic workflows for SAP and database migrations as well as historization cases at Data Migrations Solutions AG. I bring the same engineering to SME and enterprise projects in Switzerland — from the first discovery workshop to live operation in production.
An agent is an AI system that receives a goal, breaks it down into steps, uses tools, evaluates intermediate results, and escalates when uncertain. Unlike a chatbot, an agent doesn't just react — it plans and acts.
That makes agents suitable for the work that today ties humans up in routine loops: data extraction with case distinctions, migration preparation, compliance reviews, triage. Where classical RPA fails because of "too many edge cases", a well-designed agent gets to the goal.
Live in production: agentic workflows for SAP migrations and historization.
At Data Migrations Solutions AG I work on agentic workflows for the JiVS platform — the Swiss market standard for enterprise data migration and historization. Concrete areas:
Detailed use cases for each of these areas are in the use case library (in preparation).
Planner/executor patterns, tool use, memory, human-in-the-loop escalation, multi-agent orchestration. With Claude, GPT-4/5, or open-source models.
Experience with SAP S/4HANA migrations, the JiVS platform, Z-table reconciliation, customizing analysis, and historization compliance.
Input/output filters, eval datasets, regression tests, cost caps. So you know when the agent is good — and when it isn't.
Where classical RPA / Power Automate fails on edge cases, agents take over. Power Platform stays as a delivery option (not as the headline).
A chatbot reacts to user input and executes predefined actions. An agent is given a goal, plans steps on its own, calls tools, evaluates intermediate results, and decides when it's done or when to escalate. That makes agents suitable for open-ended tasks with many edge cases — for example preparing a SAP migration.
Instead of manually analyzing every migration object, an agent takes over the preparatory work: it reads schema definitions, looks for customer-specific Z-fields, checks data quality, suggests mappings, and produces a structured plan. The human migration expert only reviews escalations and final sign-offs. A detailed use-case page is coming in the use case library.
Pragmatically chosen per use case: Claude (Anthropic) for longer reasoning chains, GPT-4/5 for multi-tool use, open-source models (Qwen, Llama) where data residency or cost are decisive. Hosting: Azure OpenAI Switzerland, AWS Bedrock EU, or on-premises via vLLM/Ollama.
A typical PoC sprint costs between CHF 15,000 and CHF 35,000 (two weeks, one concrete use case, delivered end-to-end). Production implementations start at CHF 50,000 for tightly scoped use cases. Precise costing follows a non-binding discovery call.
Three layers: (1) an eval set with curated test cases before every deployment; (2) guardrails for input/output (cost caps, output schema validation, dangerous tool calls as human-in-the-loop); (3) observability in production (logs, escalation rates, drift monitoring). No agent goes live without these three.
More detail on the concrete application of agents in SAP migrations and historization cases.
View SAP & data migrationBook a non-binding 30-minute discovery call. I give honest feedback on whether agentic AI makes sense here — or whether classical automation would be the better path.
Book a discovery call