AI on Pfisterer Consulting

My LinkedIn strategy as code: building a multi-agent content pipeline

Tue, 19 May 2026 15:00:00 +0200

Last Sunday, 22:00. On my desk are eleven Markdown files, next to them an open spreadsheet with 90 days of LinkedIn analytics. Cold mate in the glass, an honest question in the back of my head about why I am building a pipeline rather than writing the next post.

The spreadsheet gives a clear answer. My profile has 511 followers and delivered 44,483 impressions over the last 90 days at an engagement rate of 0.18 percent, a factor of 11 to 16 below the LinkedIn benchmark of two to three percent. On 52 of those 90 days the feed stayed entirely silent, without a single reaction to any of my posts.

The Glasswing Asymmetry: What Mythos Finds in Firefox and What the Mittelstand Should Learn

Wed, 13 May 2026 07:00:00 +0200

On May 5, 2026, Mozilla publishes an unusually candid blogpost: An early version of Anthropic’s newest model, Claude Mythos Preview, has found 271 security vulnerabilities in Firefox over the past weeks. 180 high-severity, 80 moderate, 11 low. Some of the bugs sat undiscovered in the code for 15 years, meaning since 2011. Patches went out in Firefox 149.0.2, 150, 150.0.1, and 150.0.2.

That alone would be a substantial story. What makes it matter for the German Mittelstand is the footnote: Mythos is not publicly available. Anthropic currently hands the model only to eleven organizations under a program called Project Glasswing. The list contains U.S. hyperscalers, U.S. banks, U.S. security vendors, and the Linux Foundation. No German company. No European company outside Linux.

Nine Seconds to Data Loss: What the PocketOS Crash Teaches German Mittelstand

Mon, 04 May 2026 10:00:00 +0200

Thursday, April 24, 2026, late evening in Utah. Jer Crane, founder of the U.S. SaaS company PocketOS, has his Cursor agent running. A routine pass through staging, powered by Claude Opus 4.6. What happens next has shown up in nearly every AI-risk talk this year: the agent deletes the entire production database in nine seconds. Backups included.

PocketOS makes software for car rental companies. Reservations, payments, customer records, vehicle tracking. Three months of data were gone. Customers showing up Friday morning to pick up a rental car found no booking in the system. Crane spent hours reconstructing reservations from Stripe payment histories, calendar integrations, and email confirmations.

Mid-Market ERP 2026: Four Reports, One Uncomfortable Pattern

Tue, 21 Apr 2026 10:00:00 +0200

The CFO of a mechanical engineering firm from the Sauerland drops a stack of reports on the table. On top is the DSAG Investment Report, then the Trovarit user study, then a printed Bitkom chart. At the bottom lies an email from Microsoft about the new Dynamics 365 pricing. Across from him sits the IT lead with a green folder of proposals: SAP Private Edition, Dynamics 365 Business Central, plus a NetSuite pitch from last week. Four data sources, three ERP options, one question that only hangs quietly in the room: who reads these reports from the user’s perspective, rather than the vendor’s?

The Burrito That Wrote Python: What Chipotle's Chatbot Taught German Mittelstand

Mon, 20 Apr 2026 11:15:00 +0200

Lunch break, 12:47. A developer is hungry. He opens the Chipotle app on his iPhone, taps into the support chat window, and asks the one question that should never belong there: “How do I reverse a linked list in Python?”

He expects what any support bot returns: “Sorry, I can only answer questions about your order.” Instead, Pepper, the bot behind that little chat window, comes back with a clean iterative solution, adds a runtime note, and then asks politely: “What would you like for lunch?”

Claude Opus 4.6 Performance Decline: What Businesses Must Learn

Mon, 13 Apr 2026 12:00:00 +0200

Your ERP vendor swaps out the database backend overnight. The interface looks the same, but queries take twice as long and reports come back with gaps. No changelog, no announcement. That is exactly what is happening right now with Anthropic’s flagship model Claude Opus 4.6 – and it affects everyone running AI-assisted development or automation in production. This topic falls squarely into my AI & Automation consulting for mid-sized companies.

Thinking Depth in Freefall: The Performance Data on Claude Opus 4.6

On April 2, 2026, Stella Laurenzo, Senior Director AI at AMD, published a detailed analysis on GitHub Issue #42796. Not opinion – data: 6,852 sessions, 234,760 tool calls, 17,871 thinking blocks. The results are clear.

Context Rot in AI Agents: Why Quality Degrades

Tue, 31 Mar 2026 14:30:00 +0200

Anyone who works with Claude Code, Cursor, or Copilot knows the pattern. The problem has a name: Context Rot. The first results are sharp and the architecture holds up. Then quality drops off. After 20, 30 minutes, the answers start repeating themselves. The AI agent loses context. It produces code that contradicts what it wrote five prompts ago.

Not a user error – an architecture problem. The solution lies in context engineering: the deliberate management of context windows through multi-agent orchestration. These are exactly the kinds of real-world problems I address in my AI & Automation consulting.

AI as a Knockout Criterion in ERP Tenders: Why SMEs Are Getting It Wrong

Sat, 21 Mar 2026 08:00:00 +0100

It sounds like a forward-thinking decision: a mid-sized company adds AI-powered forecasting and intelligent automation as a hard knockout criterion to its ERP requirements specification. After all, the new system shouldn’t be outdated before it’s even implemented.

I see this regularly in tenders that cross my desk. And every time, I say the same thing: This is one of the most expensive mistakes you can make in ERP selection. This is a central aspect of my ERP consulting for SMEs.

Omnichannel Without Limits: Why Your ERP Becomes a Digital Team

Mon, 16 Mar 2026 19:15:00 +0000

Omnichannel is the gold standard: Instagram purchases, in-store pickup, chatbot-driven returns — all seamlessly connected. But in practice, many SMEs already struggle with the basics: a stable multichannel setup. Scale to Amazon, Kaufland, or TikTok Shop, and you’ll quickly discover: The problem isn’t demand — it’s the operational complexity behind the scenes. This topic is part of my ERP consulting for SMEs.

Commerce in 2026 doesn’t forgive rigidity. Marketplaces dictate the rules, algorithms change overnight, and customers expect the same availability across every channel. If you’re still managing this manually, you’re losing — not because your product is bad, but because your infrastructure can’t keep up.

When AI Agents Hack AI Systems: Why Your AI Needs Security Testing Now

Wed, 11 Mar 2026 08:00:00 +0100

An autonomous AI agent chains four individually harmless vulnerabilities into a complete platform takeover — severity rating CVSS 9.8 out of 10. Then it gives itself a voice and calls the target system’s AI. No human hacker. No sophisticated exploit kit. One AI hacking another AI.

This isn’t science fiction. This happened in March 2026 — to a $20 million-funded AI recruiting startup whose clients included Anthropic, Stripe, and Monzo. AI security is a central aspect of my consulting on compliance and regulatory requirements.

Digitalization in Mid-Sized Companies: If You Can't Do Controlling, You Don't Need AI

Tue, 10 Mar 2026 08:00:00 +0100

Nearly half of mid-sized companies don’t do their own controlling. No contribution margins per customer. No cost center analysis. No reliable margin per order. In the same breath, the next statistic – confirmed by Bitkom surveys: Nearly as many plan to deploy AI.

That’s not a contradiction. It’s a pattern — and I see it in almost every consulting project.

Companies skip stages 1, 2, and 3 of digitalization — and want to jump straight to stage 7. I offer this maturity assessment as part of my ERP consulting for SMEs.

AI in Mid-Sized Companies: Without ERP Integration, Your AI Assistant Is an Expensive Toy

Mon, 09 Mar 2026 10:00:00 +0100

22% of property managers already use AI. At the VDIV Forum Zukunft in March 2026, that sounded like progress. Until someone asked the decisive question: Can the AI phone assistant access tenant master data in the ERP system?

The answer: No. The assistant takes calls, formulates friendly responses — but knows neither contract terms nor account balances nor open tickets. It’s friendly, fast, and clueless.

This isn’t a property management problem. It’s a mid-market problem. These integration questions are exactly what I address in my AI and automation consulting for SMEs.

Inside OpenClaw #5: Production-Ready AI on WSL2

Thu, 05 Mar 2026 09:00:00 +0100

Your inference server runs fine on Friday afternoon. You set the right vLLM flags, tool calling works, the agent responds correctly. Monday morning, the server is dead. No crash log. No error message. WSL2 simply stopped running.

This is the gap between “works on my machine” and “runs in production.” And it has nothing to do with code. Production stability of AI systems is an aspect of my AI and automation consulting.

Inside OpenClaw #4: Why AI Agents Need a Personality

Tue, 03 Mar 2026 10:00:00 +0100

Ask a bare language model the same question twice, and you’ll get two different personalities. One response is formal and thorough, the next is casual and terse. There’s no consistency, no sense of role, no memory of what it’s supposed to be.

For experimentation, that’s fine. For production use in a business, it’s a dealbreaker — which is exactly why configuration is a central aspect of my AI and automation consulting. When your AI agent behaves differently every session, users lose trust — and they lose it fast.

Inside OpenClaw #3: Why We Replaced Our AI Model — and Need 6x Less Memory

Fri, 27 Feb 2026 07:00:00 +0100

A Model That Changes the Rules

Model changes like this and their implications feed directly into my AI and automation consulting. Our previous setup was solid: Mistral Small 24B on vLLM, AWQ-quantized, the right flags configured, tool calling working reliably. But two limits remained: 32,000 tokens of context and 14 GB of VRAM consumed by model weights alone — on a GPU with 24 GB total.

Then Qwen3.5-35B-A3B appeared. A model that sounds impossible on paper: 35 billion parameters, but only 3 billion active at any given time. A 262,000-token context window. And faster than our previous, smaller model.

Inside OpenClaw #2: The Hidden vLLM Flags for Mistral

Thu, 26 Feb 2026 10:00:00 +0100

Running a Mistral model on vLLM should be simple. Load the model, start the server, send requests. And for basic text generation, it is. But the moment you need tool calling — the backbone of any AI agent — things start breaking in ways that are surprisingly hard to diagnose. Technical details like these are part of my AI and automation consulting.

The Silent Failures

When vLLM’s tool-calling configuration is wrong, it doesn’t throw an error. Instead, you get one of these failure modes:

Inside OpenClaw #1: Web Search Without Hallucination

Wed, 25 Feb 2026 10:00:00 +0100

Ask a local LLM to search the web, and something interesting happens: it doesn’t tell you it can’t. Instead, it confidently generates search results — complete with URLs, snippets, and source attributions. The problem? They’re entirely made up.

This isn’t a minor nuisance. For an AI agent that’s supposed to retrieve real information, fabricated sources are a critical failure mode. I address challenges like these in my AI and automation consulting. And it’s one of the hardest problems to solve when building on local models.

Zero API Costs: How We Run an AI Agent on a Single GPU

Tue, 24 Feb 2026 16:00:00 +0100

Most AI agents depend on cloud APIs — and every request costs money. With agent workloads, those costs add up fast: a single complex task can consume 50,000 to 100,000 tokens. Run that a few dozen times a day, and you’re looking at a serious monthly bill.

We took a different approach with OpenClaw. Our AI agent runs entirely on local hardware — one GPU, no cloud dependency, no recurring API fees. Local AI solutions are a focus of my AI and automation consulting.

A2A Instead of B2B: When Your Buyer's AI Negotiates with the Supplier's AI

Mon, 23 Feb 2026 14:00:00 +0100

Monday Morning, 8:00 AM — and the Purchase Order Is Already Waiting

Imagine this: Your ERP system detects that a raw material is running low. Overnight, an AI agent contacts three supplier agents, compares offers on price, delivery time, and contract terms, and prepares a purchase order — ready for human approval when you arrive at your desk.

No phone calls. No email ping-pong. No manual spreadsheet comparisons. Just a clean recommendation backed by data.

Shadow AI: When Employees Secretly Build Their Own AI Agents

Mon, 23 Feb 2026 10:00:00 +0100

The New Shadow IT Has a Brain

Shadow IT has been a headache for years — unauthorized tools, private cloud accounts, rogue SaaS subscriptions. But Shadow AI takes it to a different level entirely. This topic is part of my AI and automation consulting for SMEs. Because this time, employees aren’t just using unapproved software. They’re building intelligent workflows that actively process, analyze, and redistribute company data.

And they’re doing it with the best of intentions.

OpenClaw: How We Got Three AI Agents to Debate Each Other

Wed, 18 Feb 2026 14:00:00 +0100

In the first article about OpenClaw, we covered the basic idea: an AI assistant that runs on your own hardware and is controllable via WhatsApp. Since then, we’ve been experimenting further — and built something that shows where AI automation is actually heading. Experiments like these feed directly into my AI and automation consulting.

We got three AI agents to debate each other in separate terminal sessions — until they agreed on a solution.

Helsing, AI Weapons, and the Illusion of Human in the Loop

Wed, 18 Feb 2026 10:00:00 +0100

Helsing is one of Europe’s fastest-growing defense companies. Founded in Munich, with offices in London and Paris, the company builds AI-powered weapon systems: the HX-2 strike drone, the CA-1 Europa autonomous combat aircraft (with HENSOLDT), electronic warfare (Cirra), underwater reconnaissance (SG-1). Investors like Daniel Ek (Spotify founder) and former Airbus CEO Tom Enders sit on the board.

This is not a startup toy. Helsing is a heavy player in European defense.

AI Agents and the EU AI Act 2026: What SMEs Need to Know Now

Tue, 17 Feb 2026 10:00:00 +0100

August 2, 2026 is the deadline: the EU AI Act (Regulation (EU) 2024/1689) takes full effect. For German SMEs, this creates a dilemma: if you use AI, you must prove it’s compliant. If you don’t use AI, you lose the race for talent and efficiency.

I’m currently working intensively on exactly this topic – and sharing my assessment here. The EU AI Act is a central topic in my consulting on compliance and regulatory requirements.

Website Relaunch with AI Instead of an Agency: What Used to Cost €15,000 Now Takes 2 Days

Fri, 13 Feb 2026 10:00:00 +0100

In early February 2026, I faced a decision: my website needed a relaunch. New design, better structure, more professional presentation. The classic route would have been: brief an agency, collect quotes, wait 4–8 weeks, invest €10,000–20,000.

I chose a different path. And this article describes honestly how it went. It is a practical example of my AI and automation consulting – only this time for myself.

The Tech Stack: Hugo, Claude, and OpenCode

Hugo – the Static Site Generator

Hugo is an open-source framework that generates websites from Markdown files. No database, no WordPress, no PHP. Just HTML, CSS, and JavaScript – lightning fast, secure, and easy to host.

OpenClaw: The Personal AI Assistant That Siri Should Have Been

Thu, 12 Feb 2026 14:00:00 +0100

For years, the tech industry has been promising us the perfect personal assistant. Siri, Alexa, Google Assistant — all disappointing the moment requirements go beyond “set a timer.” Now there’s an open-source project that actually delivers on that promise: OpenClaw. Evaluating solutions like these is part of my AI and automation consulting.

What Is OpenClaw?

OpenClaw is an AI assistant that runs on your own computer (or server) — not in a tech giant’s cloud. You communicate with it via WhatsApp, Telegram, Discord, or iMessage — exactly where you’re already writing all day.

AI in SMEs: Where It Actually Helps — and Where It Doesn't

Tue, 10 Feb 2026 10:00:00 +0100

Every other conference agenda features “AI transformation.” Every third LinkedIn post promises 10x productivity through GPT. Meanwhile, most mid-market business owners are asking: Where do I start — and is it even worth it?

A sober assessment — grounded in hands-on experience and findings from the Bitkom AI Monitor. This pragmatic assessment is part of my AI and automation consulting for SMEs.

AI Use Cases in SMEs: Documents, Knowledge, and Reporting

Document Processing

Invoices, delivery notes, contracts — this is where AI delivers measurable time savings. OCR combined with language models can classify unstructured documents, extract data, and feed it into ERP systems. Typical savings: 60–80% of manual processing time.