Coding agent costs, fragility and workflow limits
This week's trends
Coding agent costs, fragility and workflow limits

This was the week's most covered thread, though it drew less coverage than the week before. Evanxmerz.com argues heavy AI use yields fragile codebases, while deepmind.google describes AlphaProof Code verifying Signal's cryptographic protocol code in Lean. Commandlinesystems.com looks at AI tools in Emacs culture, and sergeykuhne.com lays out a ticket queue to fight AI brain fry.
Read: evanxmerz.com, deepmind.google, commandlinesystems.com and sergeykuhne.com
Vibe-coded games built with AI agents

Coverage was far greater than the week before. Notesbylex.com reports Claude Fable 5.2 built a 12-level beat 'em up from a nephew's drawings in one overnight prompt, and fable52.games collects browser games made by the model. Sandromaglione.com describes shipping indie games with agents, while jpain.io lists five vibe-coded projects its author abandoned.
Read: notesbylex.com, fable52.games, sandromaglione.com and jpain.io
Gemini 4 and Claude Fable 5.2 launch

Model launches were new this week. Anthropic announced Claude Fable 5.2 and Mythos 5.2, priced at $8 per million input tokens and $40 per million output tokens. Blog.google and antigravity.google announced Gemini 4 Hydrogen and the Orchestrator option in Antigravity. Edge0.ai released open-source Edge0-145B and Edge0-120B models for local use.
Read: antigravity.google, anthropic.com, blog.google and edge0.ai
How AI agents change engineering work and architecture

Coverage was far greater than the week before. Youtube.com carries a Jonathan Blow talk with Wolf Tivy on software complexity and why today's AI cannot produce better code than people. Hrih.si argues that architectures bet on scarce resources, and that AI made code cheap while attention, review and context stay scarce.
Read: youtube.com and hrih.si
Chrome ships web guidance skills for coding agents

This thread drew far more coverage than the week before. Developer.chrome.com and github.com describe Modern Web Guidance 1.0, with Firefox and Edge joining, which steers agents toward native browser APIs; evals show pass rates rising from about 50% to about 90%. Ui-sdk.dev offers a library of UI behaviors for agents to assemble components..
Read: github.com, developer.chrome.com, ui-sdk.dev and flawless-ui-ux.vercel.app
Compiled, edited and written with AI. Published by Rapid Review, Inc.