The archive

Latest articles

Reporting, explainers, and analysis on data and AI, updated continuously by the newsroom.

Donut chart for multi-agent safety showing AI agents succeeding on 66.3 percent of OSWorld tasks in 2025 and failing on 33.7 percent.

Multi-agent safety gets Google’s $10 million test

Multi-agent safety is the problem of keeping interacting AI agents from amplifying failure. Google’s $10 million bet starts small.

Lars Cornelissen · Jun 11, 2026

Confidential inference in Apple’s AFM 3 lineup: 2 on-device models and 3 server-based models, with AFM 3 Cloud Pro running on NVIDIA GPUs in Google Cloud.

Engineering

Confidential inference makes Apple’s cloud AI real

Confidential inference lets Apple run AFM 3 Cloud Pro on NVIDIA GPUs in Google Cloud while keeping PCC privacy promises.

Lars Cornelissen · Jun 10, 2026

Line chart showing bespoke connectors needed grows to 50 without MCP for 5 clients and 10 tools, versus only 15 with MCP, because the cost goes from clients times tools to clients plus tools.

Engineering

What is MCP? The Model Context Protocol for data engineers

MCP is an open standard that lets AI models call your tools and data through one connector instead of a custom integration per model. Here is what it means for data engineers.

Lars Cornelissen · Jun 10, 2026

Siri AI model tests show AFM 3 Cloud at 64.7 percent preference versus 8.7 percent for the 2025 server baseline, and AFM 3 Core at 45.6 percent versus 23.3 percent.

Siri AI makes Apple’s platform bet depend on Google

Siri AI is Apple’s rebuilt assistant, but its strongest model jump is Gemini-built: AFM 3 Cloud won 64.7 percent of text tests.

Lars Cornelissen · Jun 10, 2026

Donut showing WhatsApp AI assistants enforcement stakes: EU fine ceiling is 10 percent of Meta's 2025 revenue, about $20.1 billion of $201.0 billion.

Business

WhatsApp AI assistants become EU gatekeeping test

WhatsApp AI assistants are now an EU platform access fight: Meta must reopen WhatsApp for rivals for free or risk a fine near $20.1 billion.

Lars Cornelissen · Jun 10, 2026

Two-line chart: a saturated benchmark curve flattening near 95 percent while a frontier-capability curve keeps rising past it, showing the AI plateau is a measurement ceiling rather than a stall.

The AI plateau is mostly a goalpost that keeps moving

An AI plateau is mostly an illusion: old benchmarks maxed out and the AGI goalposts keep moving, even as the frontier capability curve keeps climbing.