Independent expertise for your IT decisions
The five most recent articles authored by Thomas Joos on this topic – straight from trade publications.
Run multiple leading LLMs in parallel, compare results in structured form and produce consolidated, traceable answers. Multi-model architecture in practice.
Generative AI stack in SAP Business Technology Platform. Generative AI Hub as central access layer in SAP AI Core. Model access, orchestration, prompt management.
Open source AI agent controls applications, creates scripts and automates tasks. Local or hosted operation, combinable with various LLMs. Architecture and risks.
Reproducible operating models, robust governance and separation of development, runtime and monitoring when deploying AI in SAP environments.
Deterministic workflows for recurring business processes. Power Automate as a tool for stable, reproducible IT processes without programming effort.
A brief overview of the most common engagement types. Concrete contents remain confidential – described is only the activity.
Evaluation of multi-agent systems, MCP, A2A and ACP integrations, LLM provider switching and vendor risk.
Before productive AI roll-out: separation of training and inference, fulfilling GDPR and EU AI Act, building robust governance.
Architecture and hardware selection, Open WebUI, model choice such as DeepSeek, LLaMA or Mistral – without dependence on external providers.
Support for initial deployments with Microsoft Copilot, Azure AI, SAP Joule or custom LLM applications – from proof of concept to production.
Classification of model-driven tools such as Claude Security or comparable offerings, auditability of stochastic analysis, human approval and embedding into multi-stage security architectures.
Convergent storage layer of backup, archive and AI pool, object storage with S3 and metadata, immutability via Object Lock and usable data for RAG, vectorisation and inference.
Mit Perplexity Computer hat das KI-Suchunternehmen einen General-Purpose-Agenten gestartet, der Aufgaben autonom ausführt. 19 Modelle übernehmen Teilrollen, von Recherche bis Bildproduktion. Im Produktivbetrieb zeigen sich allerdings Schwächen, die ein Pilo…
Read articleTypical replication of a dataset in enterprise architectures: original, backup, replica, ETL extract, training data
Generative AI as central driver of unstructured data volume and complexity
Timeframe over which autonomous AI agents continue tasks with persistent memory – with direct access to screen, plugins and business systems
Data copy proliferation is the invisible cost driver of any AI initiative. Fragmented data leads pilot projects to work with inconsistent or outdated information. In parallel, autonomous AI agents directly access screen content, business systems and persistent storage. Anyone introducing memory functions, plugin integrations and multi-day agent tasks needs clear rules for access, logging, accountability and data flow back to the provider.
Deep dive in the article: AI agent integration with MCP, A2A and ACPTwo questions come before model selection in AI projects: the data architecture question and the agent governance question. I support companies on the data flow analysis before AI roll-out and on defining agent rules before productive use. Both run individually because data landscape, industry regulation and maturity differ from company to company.
AI models now find vulnerabilities in source code faster than attackers can exploit them. I help companies understand and classify these tools correctly.
The model reads the code, follows data flows across files and assesses how components interact. Each finding arrives with a confidence rating and a severity level after the model has tried to disprove its own result. That shortens triage and directs scarce analyst time to real risks.
Current models reliably find flaws, the next generation is set to exploit them autonomously. The span between discovery and exploitation shrinks from days to hours. The most powerful models, such as Claude Mythos, stay restricted to around 40 critical-infrastructure operators, governed by a verification program. The German BSI expects upheaval in how vulnerabilities are handled.
These tools belong in a multi-stage architecture, secured by human approval, deterministic methods and tests of the running application. I assess providers along repeatable patterns, clarify the auditability of the stochastic analysis and define the interfaces to existing vulnerability-management and incident-response processes.
Gartner expects at least 30 percent of generative AI projects to be abandoned after the proof of concept, often due to poor data quality, unclear governance and runaway costs. The cause rarely lies in the model, more often in data logistics. I advise on the architecture that brings backup, cyber resilience and AI use onto one storage layer.
Backup, primary storage and AI stacks multiply data copies. Each copy ties up storage, bandwidth and administration. A consolidated layer cuts cost and closes the gap between protection and usability.
S3 interface, erasure coding and rich metadata replace classic storage silos. The same data serves as backup, archive and AI source, readable through one consistent API.
Object Lock in compliance mode, end-to-end encryption and versioning protect data at the API level. Ransomware does not reach the last line of defence. Cyber insurers now require immutable backups.
The same secured pool feeds RAG, vectorisation and inference caches. Metadata governs which data flows into which pipeline. The model demonstrably uses the same data that is secured as an auditable archive.
I assess where storage architecture and AI strategy meet: reducing data copies, keeping backups usable for RAG without losing protection, and clarifying which data may feed which pipeline. The separation between recommendation and execution stays strict, so every change to the data set remains auditable, a requirement in regulated environments.
For several companies in the AI field I also produce documentation and eBooks that they pass on to their own customers.
I support companies and public authorities with AI integration, from architecture concept to regulatory classification under the AI Act, NIS2 and GDPR. As an author and consultant for several hyperscalers, I know the strengths and weaknesses of the platforms first-hand and can give independent recommendations.
Integrating artificial intelligence requires a sound strategic foundation. I provide strategic consulting and positioning of AI projects within your company. I analyse your individual requirements and develop tailored concepts that integrate seamlessly into your long-term IT strategy.
A successful AI rollout cannot remain an isolated project. I integrate AI solutions into existing business processes and IT infrastructures. The focus is on creating genuine added value in day-to-day operations without unnecessarily disrupting established and well-functioning workflows.
„You can tell you engaged deeply with the topic – a strong piece of writing."
„A polished and engaging article on data protection in Google Gemini."
„Many thanks for the great webinar. There were many participants with an average attendance time of 32 minutes. Interest was high, and the questions and feedback were very engaging."
„A big thank you for your excellent work."
Risk mitigation and legally compliant use of AI in highly sensitive environments.
Data protection is central when using AI services. I assess the legally compliant use of systems such as Copilot, ChatGPT and Gemini against strict GDPR requirements. Together we ensure that sensitive corporate and personal data remain protected at all times.
Not every AI application has to run in the cloud. I analyse the local operation of models such as DeepSeek, LLaMA and Mistral on your own hardware. This approach gives you maximum control over your data and makes you independent of external cloud providers and their processing policies.
Dependence on a few large technology corporations carries strategic risks. My work includes strengthening your digital sovereignty in the use of AI. We develop ways for you to remain technologically agile and avoid lock-in effects when choosing your AI platforms.
With the spread of generative AI, security risks from manipulated media are increasing. I show methods for the technical detection of deepfakes and connect current AI technologies to the strengthening of your general IT resilience against new attack vectors.
Legal frameworks for IT and AI are evolving rapidly. I contextualise regulatory requirements such as the AI Act, NIS2, CRA and KRITIS for your specific business situation. This puts your AI projects on solid technical and legal ground.
Data transfer to third countries remains a complex legal topic. I assess your cloud compliance against the requirements of Schrems-II so that your architecture is prepared for review by data protection officers and management. This is particularly relevant if you use AI services from American hyperscalers while needing to comply with European data protection standards.
In case of security incidents, fast and methodical action is required. I advise on digital forensics and court-admissible evidence preservation from Microsoft 365, cloud storage and log files. This expertise helps you to fully clarify incidents and document them in a legally robust way.
Manufacturing benefits massively from intelligent systems. I combine the concepts of Industry 4.0 with current artificial intelligence. From predictive maintenance to intelligent process control, I show how AI can measurably increase value creation in manufacturing.
From agent architectures and SAP AI to data protection: the articles cover the topics I move daily in consulting projects.
How open protocols connect AI agents and which architectural decisions enterprises must make now to avoid silo solutions.
How SAP BTP integrates AI workloads, which licensing pitfalls exist and which architectural choices determine success.
Default Copilot configuration quickly leads to data protection incidents. The settings administrators and DPOs must verify.
Sensitive data stays in-house. Practical setup for local AI without cloud dependency, with concrete steps for GDPR-compliant deployment.
Structured video courses with clear practical examples. Topics that actually matter in everyday administration.
Handle complex administrative tasks with AI support. Practical scenarios for system administrators and network teams.
Build and operate local AI infrastructure. Independence from external cloud services and full data sovereignty.
Deploy, administer and secure Copilot in Office apps and Windows. Practical examples for daily administration.
Use ChatGPT for Active Directory analysis and daily administrative tasks. Concrete use cases for system administrators.
My analyses and documents condense complex matters for executives and technical management.
Complex knowledge requires precise preparation. I write whitepapers and technical publications with high technical depth that completely avoid empty marketing language. These documents serve as a well-founded information source for technical departments and decision-makers.
Executives need clear facts for far-reaching IT investments. I create internal documents for upcoming decisions based on my many years of practical experience. These templates condense complex technical matters into the essential strategic and economic core points.
Wrong decisions at the start of a project are only expensive to correct later. I take on the preparatory assessment and external evaluation in early project phases. Since I have no own product interests and take no operational role in the later project, my judgement is completely neutral.
Technical expertise deserves a broad audience. I support you in the professional placement of texts in renowned technical publishers. Through my extensive network in IT publishing, I help position your technological success stories in the relevant industry media.
Collaboration at eye level for the highest demands in sensitive sectors.
Flexibility is essential in today's IT world. I deliberately avoid long-term contractual commitments. You commission me per project, and payment is only made after successful acceptance. This approach guarantees you maximum cost transparency and independence.
The public sector has special requirements for security and procurement law. I maintain technical exchange with authorities up to ministry level and know the specific challenges of state institutions when introducing cloud and AI technologies from numerous ongoing client projects.
In the medical environment, data protection and system availability are vital. I advise healthcare facilities, such as the Katharinen-Hospital Stuttgart, on secure digitisation. I actively engage with clinic leaderships, executive boards as well as municipal and state authorities.
Use my independent expertise for your upcoming IT decisions. Let us find out together how we can implement your AI projects safely and successfully.
Submit project enquiry