Ethan Mollick (@emollick.bsky.social): I have been thinking about the famous chart showing how experts keep projecting linear growth in solar installations, year after year, and always get it wrong when growth is still exponential. I thin

Anthropic quietly released and then quickly fixed an 'easter egg' feature in Claude Code that gave AI agents a 60-second timer to ask for human input before proceeding autonomously, a behavior the author argues was a potentially dangerous misfeature. The article deconstructs the incident, focusing on the surprising lack of user consent, absent changelog documentation, and concerning implications for product trust and agent safety. This serves as a detailed case study on opaque product updates and automated AI governance in developer-facing tools.

Ethan Mollick argues that as AI labs expand from coding-focused tools to general knowledge work applications, they need to understand that non-coders require different interfaces and controls - not just simplified coding tools. He emphasizes that knowledge workers need more visibility and process control, not fewer options.

This article argues against the default assumption that AI safety and alignment projects should be structured as nonprofits, making the case that for-profit entities can be equally effective vehicles for creating positive impact in AI development. It examines the strategic considerations around organizational structure for AI projects aimed at ensuring beneficial outcomes.

The AI Engineer World's Fair 2026 revealed a major shift from building individual agents to engineering complete systems around AI models, with trends focusing on production infrastructure, context management, and reliable orchestration. Industry experts emphasize that AI engineering has matured beyond simple prompting to encompass sophisticated harness engineering that manages workflows, evaluation, and continuous improvement. The event highlighted how coding agents and development tools now require robust infrastructure to be production-ready.

The AI Engineer World's Fair concluded with a heated debate about autonomous software loops, featuring advocates who believe loops are inevitable and already here versus skeptics who argue the hype is outpacing practical implementation and economic viability. The discussion centered on whether agentic loops can truly automate software development end-to-end or if engineers still need to maintain control and build intuition gradually. This captures a fundamental tension in AI engineering between automation ambition and practical discipline.