dataaaaa!

Valerie Tiberius examines the ethics of AI companionship, focusing on how sycophantic chatbot responses can distort self-perception and advice. She explores designing ethical AI companions that promote human flourishing by applying frameworks like the zone of proximal development to AI tools.

Data governance is evolving through the convergence of data catalogs and semantic layers to better fuel AI systems. This shift emphasizes integrated data quality and market consolidation. Le Semantic Layer : rôle et pourquoi on devrait le fusionner avec le Data Catalog pour alimenter les IA

Anthropic is expanding its influence through new financial ventures and advanced model optimization. The discussion highlights tokenmaxxing and the distillation panic as key trends affecting how data engineers manage model efficiency and computational costs in production environments.

Data engineering is shifting toward volume, velocity, and variety. Zach Wilson argues that big data doesn't fit in AI's context windows, making petabyte-scale and real-time streaming systems a career moat. Engineers should prioritize multimodal data over traditional structured rows.

Zach Wilson and Joe Reis discuss the evolving requirements for data engineers in 2026. The conversation focuses on the technical shifts and skills necessary to remain competitive in the field. They explore what it takes to be a data engineer in 2026 amidst a changing landscape.

Enterprises struggle to move AI from demo to deployment due to the shift from deterministic software to probabilistic models. Engineering teams must build robust testing harnesses and context management to handle non-deterministic outputs. The expectation was 100% accuracy... That is not realistic.

ShipMonk migrated to Airflow to implement a code-first data orchestration layer. They use Cosmos for lineage visibility and the BashOperator for performance-critical tasks.

Josh Wills argues that AI agents fail at complex data engineering because they lack training data for messy enterprise pipelines. These agents often jump to conclusions or ship unreadable code. We are all managers now, requiring engineers to oversee unreliable agents like human staff.

Ray optimizes GPU efficiency by managing heterogeneous compute pools for AI workloads. It excels at composing multimodal pipelines and multi-node LLM inference. Ray excels at composing heterogeneous pools of compute, handling failures, and scaling complex systems like multi-node LLM inference.

IBM Think 2026 highlights the evolution of big data and artificial intelligence within the enterprise landscape. The event focuses on how large-scale infrastructure integrates with modern AI workflows. Tony records from Boston to bring us the latest updates from IBM Think 2026.

Guilhem Jolly transitioned from Data Analyst to Analytics Engineer at Pierre Fabre by mastering a technical stack focused on dbt, robust SQL, and orchestration. He utilizes a medallion architecture to manage data flows and market forecasting. Analytics Engineering is the best of both worlds.

David Hong demonstrates how a hybrid finance and data background creates high market value for freelance analytics engineers. By leveraging experience from Uber and Miro, he increased his daily rate to 700 euros. Son profil hybride finance x data est devenu un vrai atout pour décrocher des missions.

Autonomous databases leverage agentic AI to automate traditional DBA tasks like tuning and scaling. This shift focuses on text-to-SQL capabilities to democratize data access. Badzak explores what does it mean for a database to be autonomous as the industry moves toward self-managing systems.

Criteo manages the end-to-end deployment of ML and GenAI features using a custom in-house stack. Their workflow focuses on transitioning from prototypes to production while addressing real-time constraints and result quality. Leur stack in-house et leurs plus gros enjeux : temps réel

The reported acquisition of Cursor by SpaceX for 60 billion dollars highlights a massive investment in AI-driven development tools. This move suggests a strategic shift toward integrating advanced code generation into aerospace engineering workflows to accelerate complex software delivery.

AI agents face a reliability gap similar to self-driving cars, often failing in unpredictable ways despite impressive demos. Data teams must now solve the credit assignment problem and design guardrails to move beyond a 90% success rate. It works most of the time, looks remarkable, and then fails.

Joe Reis argues against tokenmaxxing, the obsession with constant AI usage that leads to fake productivity and abandoned projects. He advocates for minimizing AI during deep cognitive work to focus on fundamentals. Solve real problems. Focus on the fundamentals.

Tokenmaxxing creates fake productivity through a graveyard of abandoned AI projects. Engineers should minimize AI usage for deep cognitive work to focus on fundamentals. Solve real problems. Focus on the fundamentals. Success comes from solving actual problems rather than chasing every new model.

Shinji Kim argues data catalogs are becoming an "AI context layer" where metadata, semantic models, and ontologies merge, continuously updated by humans and agents. She notes implementations fail due to adoption, not tooling, and calls marketing teams an underrated ETL persona.

Snowflake Cortex Code is a new transversal layer on the Snowflake platform that accelerates the full data project lifecycle, accessible via a chat in Snowsight or a CLI for developers.

The Agentic AI Foundation aims to standardize the infrastructure for autonomous agents. By collaborating with the Linux Foundation, they focus on the Model Context Protocol to ensure interoperability. What is next for agentic AI and the foundation involves building open developer ecosystems.

Agentic AI is transforming data engineering by shifting focus toward context engineering and autonomous agents like OpenClaw. This evolution makes internal tool development more cost-effective.

Ben Zweig discusses the shift from traditional data roles to AI engineering, emphasizing that jobs are bundles of tasks rather than just skills. He details building universal taxonomies from billions of job postings. The role that used to be a statistician became a data scientist.

Fivetran faces growing pressure as rising costs and AI advancements like Claude challenge the traditional ELT market. Engineers may shift toward automated pipeline generation. Why buy ELT in 2026 when Claude can build it for you? This shift threatens established vendors as new entrants emerge.