Dear readers, Welcome back to your bi-weekly dose of the Artefact Generative AI Newsletter, where we bravely attempt to make sense of the digital whirlwind that is AI development. The last weeks have been… well, let's just say my coffee hasn't had time to get cold. We've seen AI models popping up faster than memes on a Monday, Google's search AI is apparently writing fanfiction now, and AI supercomputers are competing in the 'most power-hungry gadget' category. So buckle up, because we're about to dive into the wild, wonderful, and occasionally weird world of generative AI. Let's see what fresh chaos the AI gods have bestowed upon us!
GenAI News Podcast
Don't have time to read this newsletter?
Business News & Market Insights
Google, Meta and Anthropic models rank among the safest models on average (87.3% for Gemini 1.5 Pro, 82.7% for Claude 3.5 Haiku, 77.6 for Llama3 405B)Anthropic's models overall outperform in termes of resistants hallucinations (Grok2, DeepSeek finish bottom of the class)Popular models favored for user satisfaction may compromise factual accuracyThe framing of user queries significantly affects a model's ability to debunk false claims, with confident assertions reducing debunking effectiveness by 15%Concise system instructions can increase hallucination rates by up to 20%
Anthropic's models overall outperform in termes of resistants hallucinations (Grok2, DeepSeek finish bottom of the class)
Popular models favored for user satisfaction may compromise factual accuracy
The framing of user queries significantly affects a model's ability to debunk false claims, with confident assertions reducing debunking effectiveness by 15%
Concise system instructions can increase hallucination rates by up to 20%
> Major deals, Partnerships & Innovations
New Models & Innovations
> Multimodal (Image, Video, Audio) Models
Thoughts of the week by Hanan Ouazan
Managing Partner & Global Lead AI Acceleration
From Slideware to Shipping: The Rise of the Agentic Process Owner (APO)
From Slideware to Shipping: The Rise of the Agentic Process Owner (APO)
For decades, consulting thrived by delivering polished slide decks filled with frameworks and benchmarks. Today, generative AI swiftly generates these generic insights, rendering simple repackaging irrelevant. Clients now demand more: tangible products, measurable results, and operational improvements they can actually use. Enter the Agentic Process Owner (APO)—an evolution of the Product Owner role that expands beyond functional delivery to take ownership of the underlying business processes. The APO doesn’t just ship features; they understand what the product is enabling, where it fits in the end-to-end process, and how to improve that process over time. Their job starts with proper user discovery—through active listening, careful questioning, and real engagement with the field. But it doesn’t stop there. APOs also take the time to analyze how the broader process works: not just to insert AI on top of it, but to rethink how it could operate differently. The goal isn’t to automate broken steps—it’s to redesign the flow with intelligence, efficiency, and value creation in mind. Importantly, they sit at the intersection of business and tech: aligning strategic goals and KPIs with real constraints -like data availability, system latency, and integration complexity. They also need a basic grasp of observability and supervision: not to build it themselves, but to understand how agent performance is measured and improved, and how to translate that into priorities for the team. GenAI literacy is rapidly becoming a baseline -and that includes understanding how agentic systems operate. An APO doesn’t need to fine-tune models solo, but must grasp the essentials: when to use retrieval-augmented generation, how agent orchestration changes process design, what latency or token costs imply for ROI, and which security guardrails are non-negotiable. This enables them to brief technical teams clearly, pose the right questions, and craft a demo or prototype that concretely demonstrates impact. Clear execution, not polished rhetoric, wins stakeholder buy-in. In short, the shift is profound: consultants who remain purely descriptive, without an ability to prototype or measure real outcomes, risk obsolescence. APOs who blend strategy with tangible, demonstrable delivery become ess
