Centerbit insights
Blog
Practical articles on automation, AI agents, workflow design, and maintainable systems.
Trades 2026: What the DIHK digitalization survey reveals about AI in SMEs
The 2026 DIHK digitalization survey shows AI has reached German SMEs, but trades, construction and industry lag behind in productive use. What the numbers mean in practice, where the typical friction points sit, and why human-in-the-loop is the decisive difference between pilot and production.
Read articleAI agents and prompt injection: What enterprises can do today against indirect attacks
Indirect prompt injection is the biggest security threat to AI agents in 2026. OpenAI's GPT-5 system card shows 56.8% attack success rate on hardest-tier benchmarks, other frontier models exceed 70%. Three real incidents (GrafanaGhost, ForcedLeak, GeminiJack) illustrate the attack class. Why model guardrails no longer suffice, which five concrete safeguards enterprises should pull now, and what an audit-ready security concept must deliver.
Read articleAgent Loop Engineering: How One-Off Prompts Become Running Systems
Loop engineering shifts the focus from prompt craft to system architecture. Five building blocks, memory, self-improvement, and the risks that sharpen with every loop.
Read articleClaude Fable 5 Is Here: Why the Strongest Model Is Still the Wrong One for Companies
On 9 June 2026 Anthropic released Claude Fable 5, its strongest model to date, at double the price and with 30 days of mandatory data retention. What that means for cloud-AI architectures in companies and why the case for local AI grows stronger.
Read articleEnterprise Knowledge Management 2026: When the Master Leaves, the Knowledge Goes
When the master retires, the knowledge goes with them. Why classical RAG alone is too imprecise, what hybrid search delivers, and how centerbit combines several methods to make implicit knowledge productive.
Read articleCloud Infrastructure for AI in 2026: What Edge Computing, SLMs, and Spec-Driven Deployment Mean for SMEs
AI agents are reshaping what cloud infrastructure must deliver. What edge computing, small language models, and spec-driven deployment mean for midmarket cloud architectures in 2026.
Read articleLocal AI agents on 16 GB: What Gemma 4 QAT means for European businesses
Google's Gemma 4 QAT models now run on 16 GB laptops and 1 GB mobile devices. What Quantization-Aware Training changes technically, which model fits which hardware, and how European businesses get into production in ten minutes — including the gotcha and GDPR-compliant deployment scenarios.
Read articleDocument Automation 2026: Quotes, Reports, and Templates from Structured Data
Quotes, reports, templates: AI-powered document automation cuts processing time by 70 to 90 percent. Three document types, their specifics, their limits, and why the template matters more than the prompt.
Read article