Three things jumped out today. First, the tooling is getting serious — Claude Code now schedules tasks that keep running on Anthropic's infrastructure. This isn't a demo feature anymore. This is "I can actually automate my weekend PR reviews" territory.
Second, the capability overhang piece nails something I've been thinking about: coding agents work because codebases are self-contained. Your enterprise knowledge work is scattered across Slack threads, email chains, legacy systems, and whatever's in Jim's head from 2019. Code lives in a repo. No wonder agents crush coding first.
Third, Willison's economic argument about code quality is fascinating but wrong. Sure, competition might favor maintainable code generation — if the people buying AI coding tools could tell good code from garbage. Most can't. They just know it compiles.
New permission deferral system lets headless sessions pause on tool calls and resume interactively, plus opt-in flicker-free rendering.
Production-ready Java SDK for AI agents with Google Maps integration, URL fetching, and standardized Agent2Agent protocol for multi-framework workflows.
GitHub engineer shares hands-on experience using coding agents to build automation agents for their own workflow.
Part 2 of PM series shows how AI eliminates traditional handoff delays and translation loss in prototyping workflows.
Schedule recurring tasks like PR reviews and CI analysis that keep running on Anthropic's servers even when your device is off.
Open-source agent uses type schemas, compiler verification, and structured feedback to transform qwen3-coder from 6.75% to 99.8% function calling success.
Fashion company shipped creative generation system 2x faster using Vercel's AI stack to orchestrate 50+ image models for seasonal campaigns.
New datasette-llm releases enable purpose-specific API key configuration and tool call loop tracking for LLM workflows.
Progress on making current AI systems friendly doesn't solve the technical problem of building godlike superintelligence safely — we need to track this crucial distinction.
Frontier AI models achieve tasks at 3% of human costs with no upward cost trend, suggesting capability advances can continue under strict economic constraints.
Coding agents outperform other domains because codebases provide self-contained context, unlike fragmented enterprise knowledge work across calls and legacy systems.
Massive funding round signals unprecedented investment in next-generation compute and global AI expansion as demand for enterprise AI accelerates.
Review of documentary that captures the real human experience of grappling with AI timelines and existential risk through marriage, parenthood, and apocaloptimism.
Leaked details on 'Mythos' — Anthropic's larger, more intelligent successor to Opus with dramatic improvements in coding and reasoning, but expensive to serve.
Competition between AI models will favor those generating maintainable code because good code is cheaper to generate and maintain in the long run.