Reflections on AI adoption, strategic governance, and building high-impact AI capabilities for SMEs and NGOs.
Running an open-weight model locally can improve cost, latency, privacy, and sovereignty. It also transfers provenance, patching, continuous evaluation, incident response, and audit evidence to your organization.
AI is accelerating while regulation falls further behind. One-time regulation is anachronistic and unsustainable. Explore why continuous evaluation is the only viable response to a regulation in permanent flux.
AI eliminates the PM-as-relay function: status aggregation, workstream translation, stakeholder buffering. What survives is portfolio coherence, held by fewer people with more context. Discover the new competency surface that includes context architecture, seam design, AI-verifiable definitions of done
AI governance includes seven distinct domains, from AI security to agentic failure modes. These responsibilities cannot be absorbed as an add-on to an existing executive mandate. This article maps four structural paths and nine concrete signals that indicate when dedicated leadership stops being optional.
Beyond the Singularity hype: Moltbook reveals that "emergent" AI behavior is actually just human behavior in disguise. Behind 1.5 million bots were 17,000 humans, proving that every "rebellion" was simply a reflection of human configuration choices.
Are hallucinations a bug or a feature? This article examines why natural language protocols like MCP introduce inherent risks into AI architectural design.
On-premise deployments at the end of 2026 will probably have the same performance as end of 2025 cloud models, offering lower costs, data sovereignty and predictable operation costs
Explore how AI's rapid adoption hides structural risks (wealth concentration, unsafe physical deployment, and opaque information control) that will shape society more than capabilities hype.
Follow Alexandru Serbanati on LinkedIn for the latest updates and discussions on AI strategy.
Follow on LinkedIn