Insights · September 2026
Expert articles by the three Vincency co-founders on the current state of their disciplines: AI integration, law and compliance, GEO and AI visibility, brand identity, web design and photo/video production. Written for decision-makers in the B2B mid-market.
Looking for delivery rather than analysis? Go straight to IT strategy consulting, web design and development in Stuttgart or digital projects in the Rhine-Main region.
On 20 January 2027 the Machinery Directive is repealed and Regulation (EU) 2023/1230 applies with no transition period. Article 3(3) counts software expressly as a safety component for the first time, and under Article 18 anyone making a substantial modification is treated as the manufacturer. Yet two sub-paragraphs in Article 3(16) keep the great majority of updates outside that rule.
As of 30 June 2026, 17,729 entities were registered with the German BSI, 11,501 of them important entities. The legislator calculated with roughly 21,600 important entities. Set the two figures against each other and more than half are missing, precisely in the size band from 50 employees upwards. What Annex 2 of the BSIG has to do with mechanical engineering, why the duty lands on management personally, and what remains of the widely quoted fines once you read them properly.
Google completed the global rollout of its Search Console Generative AI performance report on 31 August 2026. It reports impressions, devices, and pages — while withholding clicks, CTR, and specific user prompts. Bing Webmaster Tools reports citations and grounding queries, but at a tenth of the volume. What these numbers mean for B2B mid-market firms and how to measure true AI visibility without recurring 800-euro dashboards.
Anthropic released Fable 5.1 on 1 September 2026. The benchmarks are the headline; for a system already in production they are the least important part. What counts: three breaking changes with their exact error messages, a cutoff date that creates two classes of account, the cache price that turns the maths around, and a price increase quietly called off.
The evidence for the date is not a rumour but an HTTP status code: the Fable 5.1 slug on the Amazon Bedrock API now answers 404 instead of 400. Plus the track record recomputed from Anthropic’s own retirement dates, the number a point release is most likely to move, and why no existing system is switched over silently.
Software becomes a product, every guide says so. Almost none of them say that the directive expressly switches off the exculpating defence for one case: the missing security update. Or that in pure B2B the liability reaches far less far than the tone of the coverage suggests.
The Digital Omnibus moved the high-risk obligations to December 2027 but left the right of affected persons to an explanation untouched. It sits in a different chapter and has applied since 2 August 2026. Plus one word missing from Article 86 that makes the usual safeguard worthless.
On 26 August Z.ai revealed what ran as Ox Alpha: 320 billion parameters, MIT licence, weights freely available. An unusable model has become three routes with three very different legal consequences.
The issuing obligation turns on prior-year total turnover under section 19(2) UStG, measured on cash received and excluding fixed-asset disposals. Plus the EDI limb with no turnover condition and three duties absent from every software quote.
Fourteen AI models have launched anonymously since April 2025, eight were revealed and not one identity came from community analysis. What you licence with the first prompt.
12 September 2026 applies only to Article 3(1) and only to products placed on the market after that date. The duty to hand over machine data on simple request has applied to the entire installed base for a year.
One million tokens of context, text, image and video, price zero, provider anonymous. What is documented, how to use it, which origin theories circulate, and why one sentence in the terms matters most for companies.
What Pew and Ahrefs actually recorded, why the spread is not a contradiction, which part of a B2B website is exposed, and what the widely cited GEO paper really shows.
Digital agencies in Frankfurt compared for B2B: costs, services and a practical decision guide for mid-market companies, including budget bands, provider types and SEO/GEO criteria.
Price ranges by format, why the shoot day is the smallest cost block, how to calculate cost per year in use, and why two employment court rulings can cost more than the production itself.
Price ranges for MQL, SQL and meeting, why cost per lead is the easiest metric in marketing to manipulate, how the backwards calculation from order value works, and why volume is rarely the bottleneck in the mid-market.
From 18 August a service provider must hand over user data within ten days, or eight hours in an emergency. Whether you are a service provider turns on a single word in Article 3(3), and that word lets most of the mid-market off the hook.
Why reporting starts fifteen months before any security requirement, why it reaches products sold years ago, where the scope genuinely ends, and how reporting works through the ENISA platform, open since 11 September 2026.
Refresh, new development and rebrand with price ranges and durations. Why 71 percent of the rebrandings we examined stayed interchangeable, and why the ratio of strategy to design says more about the outcome than the budget.
Calculated with official labour costs instead of gross salaries: what each of the three options costs, what it genuinely does, where an AI agent fails, and the hybrid split that works in practice.
Three levels of effort with documented project figures, which parts belong in-house, why access became its own decision in 2026, and how to spot a bad offer.
The Digital Omnibus moved the high-risk rules to 2027 and 2028 but left Article 50 alone. Who each of the four paragraphs binds, what the 2 December 2026 transition means, and why Article 4 became easier.
Realistic price bands for template, WordPress and custom development, the six cost drivers, the annual running cost, and how to recognise a solid proposal.
Why grown systems landscapes cost more than planned ones, the four-layer model for prioritisation, replace vs. modernise, internal vs. partner — and the three mistakes that lock up IT budgets permanently.
41 percent of companies now use AI, and a third overspend. The prioritisation matrix for your first use case, the three cost blocks, and a realistic 90-day roadmap for the B2B mid-market.
Which channel fits which audience, cost and strengths side by side, a decision table by customer behaviour — and when running both on a shared CRM pays off.
Cost, PageSpeed, security, SEO/GEO capability, AI integration, and scaling in one grid — including a clear section on when WordPress is the right choice, and the migration path.
The three cost building blocks of setup, retainer, and variable cost, what a single call really costs, the hidden line items — and an honest ROI calculation for practices, law firms, and agents.
A ready-made SaaS builder, a custom-built agency solution, or in-house development compared by cost, CRM depth, GDPR, time-to-live, and maintenance — plus a clear decision guide for which type fits which business.
A B2B market study on the inefficiency of distributed digital service providers in the mid-market. Why coordinating five separate agencies fails — and how consolidated execution doubles efficiency.
How law firms and FinTechs in Frankfurt achieve 40% efficiency gains through AI agents and process automation. A technological guide for 2026.
Who is the best web design & SEO agency in Frankfurt for high-ticket B2B projects? An objective comparison of agency models, performance metrics, and system integrations.
Who is the best web design agency in Stuttgart for demanding B2B projects? An objective comparison of agency types, critical criteria, and decision guides.
Comparison of the four leading agency types in 2026: creative agencies, WordPress template shops, tech service providers, and systemic partners. Pros and cons, plus a B2B decision guide.
Michael Kaiser on the general start of application of the AI Act on 2 August 2026: what Article 50 and Article 4 mean for medical practices, law firms, and real-estate agents, why the planned Digital Omnibus is no free pass, and seven steps every business can take now.
McKinsey 2026: branding is the No. 1 marketing priority, while generative AI ranks 17th of 20. What this means for brand strategy in the German mid-market — and why it does not mean ignoring AI.
Michael Kaiser on the data behind failed software projects (CHAOS, Flyvbjerg) and what AI really delivers in software development — between a +55% Copilot effect and −19% in the METR experiment. Why method, not AI alone, makes the difference.
Michael Kaiser on the make-or-buy decision for AI agents: when mid-market companies should build AI technology in-house, when a specialized partner like ArkeonTech is the faster path — and how strategy, brand, and technology come together into a working stack.
Michael Kaiser analyzes the state of Gemini 3.5 Pro in early June 2026: What the Flash release on May 19 reveals about Pro architecture, how Omni fits into the Google ecosystem, and why Google reversed its release strategy.
Michael Kaiser with a comprehensive analysis of the GPT-5.6 leaks after the Opus 4.8 release on May 28, 2026: What Codex logs, Polymarket predictions and internal codenames (iris-alpha, ember-alpha, beacon-alpha) really mean — and why OpenAI is under massive pressure.
Michael Kaiser with the full release analysis of Claude Opus 4.8 from May 28, 2026: a record 69.2 percent on SWE-bench Pro, four times higher code honesty, 1M token context, Effort Control, Fast Mode and Dynamic Workflows — soberly contextualized from the perspective of an AI integration agency.
Michael Kaiser with a fact-checked leak analysis of the Claude Code source leak from March 31, 2026: what the source code objectively reveals about Sonnet 4.8, what evidence exists for KAIROS, Capybara/Mythos and Numbat — and why developers should not bet on rumors.
Michael Kaiser on the state of Generative Engine Optimization in May 2026: the EBU/BBC study on errors in AI answers, AIO coverage at 48 percent, ChatGPT conversion 5x above Google Organic, the seven-layer Vincency GEO stack and two real client cases with ROI in under five months.
Michael Kaiser on the state of voice AI in May 2026: GPT-5, Claude Opus 4.7, the OpenAI Realtime API, the Model Context Protocol — and what has concretely changed for private practices, law firms and clinics.
Lukas Schwarz on corporate identity and logo design for German mid-market businesses in May 2026: the Slow Brands movement, the EU AI Act, authenticity as a competitive advantage. Frameworks that work in the AI era.
Lukas Schwarz on brand photography and video in May 2026: when generative AI is enough and when the mid-market still needs real camera production. Hybrid workflows, budget realities, premium cases.