← All digests

Dev Digest — May 6, 2026

🔥 HOT RELEASES

Anthropic "Code with Claude" Conference — MASSIVE drops Anthropic's first-ever developer conference happened today in SF. The announcements:

Node.js 26 Released The big one: Temporal API enabled by default. JavaScript's date/time handling just got a proper fix after years of pain. Also ships V8 14.6, Undici 8, and various deprecation removals. LTS expected October 2026. Bonus: Node.js 26.0.1 (tomorrow) brings node:ffi support. 🔗 X: https://x.com/robpalmer2/status/2051754952075125125 🔗 X: https://x.com/NodeSource/status/2051715296960913683 🔗 Blog: https://nodesource.com/blog/nodejs-v26-is-here 🔗 Release: https://nodejs.org/en/blog/release/v26.0.0

SubQ — Sub-Quadratic Attention LLM Miami startup Subquadratic came out of stealth ($29M seed) with SubQ: a model using Sub-quadratic Sparse Attention (SSA). Claims: 12M token context, 52x faster than FlashAttention at 1M tokens, ~1/5th the cost of Opus. Caveat: no independent verification yet. If real, this is a paradigm shift. If not, AI's Theranos moment. 🔗 X: https://x.com/bindureddy/status/2051790195423605236 🔗 Technical blog: https://subq.ai/how-ssa-makes-long-context-practical 🔗 Site: https://subq.ai

Valibot v1.4 New string case transforms (toCamelCase, toKebabCase, etc.), significant TypeScript perf improvements for object/record schemas, and isoDateTimeSecond for local timestamps. 🔗 X: https://x.com/FabianHiller/status/2052011105203941686 🔗 Release notes: https://valibot.dev/blog/valibot-v1.4-release-notes/

GitHub Copilot CLI v1.0.42 Improved MCP server diagnostics, new "rubber-duck agent" for GPT sessions (powered by Claude, experimental), -C flag for working directory, and various UX fixes. 🔗 X: https://x.com/GHCopilotCLILog/status/2052040728428097545 💻 GitHub: https://github.com/github/copilot-cli/releases/tag/v1.0.42

🧪 INTERESTING REPOS

DeepSeek-TUI — Terminal-based coding agent for DeepSeek models. Trending hard: +2,434 stars in 24h, 10.5K total. 🔗 X: https://x.com/trending_repos/status/2051949543886033245 💻 GitHub: https://github.com/Hmbown/DeepSeek-TUI

Authgear — Open-source Auth0/Clerk alternative. Self-hosted authentication server. 🔗 X: https://x.com/tom_doerr/status/2051758674406777201 💻 GitHub: https://github.com/authgear/authgear-server

X-OSINT — CLI tool for real-world OSINT: email/IP/phone lookup, subdomain enum, metadata extraction, CVE search. 🔗 X: https://x.com/VivekIntel/status/2051762879020605842 💻 GitHub: https://github.com/TermuxHackz/X-osint

ProcMon-MCP — MCP server giving AI assistants deep visibility into Windows internals (processes, ETW kernel traces, event logs, drivers, PE analysis). 🔗 X: https://x.com/hackerfren/status/2051743815078137945 💻 GitHub: https://github.com/0xhackerfren/ProcMon-MCP

FlyAI Skill — Search flights and hotels from inside Claude Code. By Alibaba. 🔗 X: https://x.com/tom_doerr/status/2052060544102727779 💻 GitHub: https://github.com/alibaba-flyai/flyai-skill

Anthropic Skills Repo — Official Claude Code skills, including the new /claude-api skill. 🔗 X: https://x.com/ClaudeDevs/status/2052069328435175654 💻 GitHub: https://github.com/anthropics/skills/tree/main/skills/claude-api

Hokou Shorts Bot — Just open-sourced automation for video shorts creation. 🔗 X: https://x.com/amritwt/status/2051932692086993346 💻 GitHub: https://github.com/retrogtx/hokou-shorts-bot

Cocos Engine — Open-source 2D/3D game engine for Web, Mobile, and Desktop. 🔗 X: https://x.com/tom_doerr/status/2051850479894745114 💻 GitHub: https://github.com/cocos/cocos-engine

🎥 WORTH WATCHING

Boris Cherny (creator of Claude Code) at AI Ascent 24-min interview with Sequoia. Hasn't written a single line of code in 2026. Says coding is "effectively solved," works entirely from phone + PR reviews. Predicts Claude Code will be ~100 lines of code soon. 🔗 X: https://x.com/DataChaz/status/2051770066912379309

Tom Crawshaw's Claude Code Content System (YouTube) 9-year automation veteran walks through his Claude Code setup: Skills > Projects, auto-updating voice profile from X API, a copywriting framework distilled into markdown, and the hidden /insights command. Practical stuff. 🔗 X: https://x.com/coreyganim/status/2052007472010076295 🎬 YouTube: https://youtu.be/HYHbVWGJ23U

Code with Claude Keynote Livestream Full conference stream from today. Managed Agents, dreaming, GitHub at scale. 🔗 Livestream: https://twitter.com/i/broadcasts/1qGoNegbnRNKv

Simon Willison's Live Blog Real-time notes from the Code with Claude keynote. 🔗 X: https://x.com/simonw/status/2052055655230706032 🔗 Blog: https://simonwillison.net/2026/May/6/code-w-claude-2026/

💡 TECHNIQUES & IDEAS

The Anatomy of a Claude Code Project Solid thread on structuring repos for AI-native development: CLAUDE.md as repo memory (keep it short), .claude/skills/ for reusable workflows, .claude/hooks/ for guardrails, progressive context in docs/, and local CLAUDE.md files near risky modules. 🔗 X: https://x.com/BharukaShraddha/status/2051920978809401584

7-Agent Solo Agency ($18.8K/mo) Viral thread about a dev running a solo web design agency with 7 Claude Code agents: Scout (Google Maps scraper), Diagnoser, Builder (Lovable), Filmer (Higgsfield), Pitcher, Checker, and Mobile. $480/mo API cost, 47 clients at $400 each. Whether real or exaggerated, the architecture pattern is interesting. 🔗 X: https://x.com/eng_khairallah1/status/2051995792840544496

GitHub Secret Scanning via MCP — Now GA Secret scanning through the GitHub MCP server is generally available for repos with Secret Protection. Works with Copilot CLI, VS Code, and any MCP-compatible agent. 🔗 X: https://x.com/GHchangelog/status/2051795340936843629 🔗 Blog: https://github.blog/changelog/2026-05-05-secret-scanning-with-github-mcp-server-is-now-generally-available

Personal AI Agent Comparison Deep Dive Honest comparison of OpenClaw, Hermes, Claude Code, Codex, and Gemini as personal agents. Nobody has won the race yet. 🔗 X: https://x.com/petergyang/status/2052030213861879894 🔗 Article: https://creatoreconomy.so/p/the-race-to-build-a-personal-ai-agent-openclaw-hermes-claude-codex-gemini

🔮 EMERGING TRENDS

Agent Dreaming & Self-Improvement — Anthropic's "dreaming" feature lets agents review past sessions and learn from them. This is a step toward persistent, continuously improving AI systems. The outcomes loop (rubric + grader sub-agent) is particularly interesting for production agent reliability.

Sub-Quadratic Attention — SubQ's SSA architecture claims to break the quadratic scaling wall. If validated, this changes the economics of long-context AI dramatically. Multiple researchers already connecting it to their own sparse attention work.

MCP Everywhere — MCP servers are proliferating fast: Windows process monitoring, GitHub secret scanning, flight booking, payment integration. The protocol is becoming the standard interface between AI and everything.

Code → Configuration — Boris Cherny's "coding is solved" take and the rise of agent orchestration systems suggest we're moving from "writing code" to "configuring agent systems." The repo structure thread about CLAUDE.md reflects this shift.

Compute Partnerships — Anthropic-SpaceX deal signals compute is still the bottleneck. Rate limit increases immediately followed. Expect more unusual compute partnerships.

Generated May 6, 2026 at 18:00 UTC by 99 Cooking

Full digest: https://digest.99.cooking