Practice experiences, market observations, and tutorials from our work with production AI systems.
August 3, 2026
LLM judges flip their verdict in up to 30% of cases just from answer order. Here is how we are building our eval framework — openly, including what is still missing.
July 31, 2026
Apertus 1.5 from ETH, EPFL and CSCS was released on 24 July 2026. We assess which use cases the Swiss model suits, the three ways to obtain it, and the operational duties its acceptable use policy creates.
July 27, 2026
Multi-agent orchestration in practice: why concurrent AI agents shift the bottleneck to review, and the worktree rules we adopted after merge conflicts.
July 20, 2026
Local AI models promise privacy and cost savings. We show when running your own LLMs actually makes sense for Swiss SMEs.
July 7, 2026
Twelve ready-to-use prompt templates for email, meetings, presentations and decisions — plus an honest analysis of the seven viral “God Mode” prompts: what actually works, and where hallucinations lurk.
July 7, 2026
98.7 percent versus 31: a spectacular benchmark number is fuelling the claim that vector databases are obsolete. We checked the primary sources — what is real about vectorless RAG, where the marketing begins, and what companies should actually take away from it.
May 25, 2026
The EU AI Act applies to Swiss companies via the market access principle. The May 2026 Omnibus deferred some deadlines — but not the prohibitions already in force since February 2025.
May 23, 2026
Anyone looking at a complete inventory of today's digital advertising protocols sees a spaghetti diagram of eighteen standards. Supply chain, programmatic, video, privacy, identity, mobile, measurement, and now also agentic. Nobody in a marketing team knows all of them — and that is not the ambition either. What you need instead: an orchestrator that hides the protocol complexity behind a single steering logic.
May 23, 2026
An agent scales what it sees — including all distortions in its training and signal data. Classical unit tests do not catch this scaling because they check samples, not all possible inputs. Why the answer to the bias problem in agentic operations must be mathematical — and how we have productively verified sixty-six theorems with Lean 4.
May 23, 2026
Over the past twelve months we have rebuilt the Digital Campaign Manager from a classical multi-channel campaign tool into a cockpit for agentic marketing operations. What this shift means concretely, which four functions we have newly built into DCM, and which agents from the opua brand family DCM coordinates today. A product announcement with architecture and timeline.
May 23, 2026
Marketing operations are shifting on a scale not seen since the programmatic wave fifteen years ago. Anyone still setting up campaigns and interpreting weekly reports in 2026 will have the wrong tool in hand by 2027. What the new steering role actually means — and which tools we have built for it inside the opua brand family.
May 22, 2026
When the CFO asks "which number actually counts?", classic single-source attribution breaks. Four complementary tiers — operational, strategic, agentic, AI-native — are the answer that holds up in 2026.
May 21, 2026
Classical software tests check examples. Formal verification proves properties across all possible inputs. Why this distinction will determine the future of regulated ML applications — and how opua applies the method across MMM, auctions, and equity research.
May 21, 2026
MLPs are universal approximators but black-box. KAN architectures are universal approximators plus interpretable plus convertible into closed-form mathematical expressions. What that means for regulated ML applications — and how opua deploys the architecture as a shadow layer and audit bridge.
May 21, 2026
How can a brand-family network share a unified DNA? Opua's answer: a shared formal-verification stack. Nexbid verifies auctions, MMM-Wizard verifies budget allocation, Mineralis verifies equity recommendations — across all brands the same mathematical foundation.
May 21, 2026
How the Model-Context-Protocol gives Claude and ChatGPT direct access to Bayesian MMM pipelines — and what that means for DACH marketing teams.
May 21, 2026
AI agents learn which phrases convert in the short term and repeat them forever — brands become interchangeable. Specialization, not generalization, is the structural answer.
May 21, 2026
When AI agents call APIs, they need authentication. OAuth browser flow does not work for stdio MCP servers. Three auth paths — when to use which.
May 21, 2026
AWS, OpenAI, Azure dominate AI. Data flows into US data centers. The Swiss FADP and the EU GDPR are obstacles. What does a sovereign EU AI architecture actually look like?
May 21, 2026
Why does opua build four brands instead of one large platform? A strategic answer to AI brand dilution and EU specialization — plus cross-brand SSO as a coherent UX.
