Dev Digest — April 10, 2026

🔥 HOT RELEASES

Bun v1.3.12 Massive release with 120 bug fixes. New Bun.WebView for headless browser automation, in-process Bun.cron() for scheduled tasks, bun ./file.md pretty-prints Markdown, and better async stack traces. 🔗 X: https://x.com/bunjavascript/status/2042441934439137453 📖 Blog: https://bun.com/blog/bun-v1.3.12

Meta Muse Spark Meta Superintelligence Lab's new multimodal reasoning model. 10x less compute than Llama 4 Maverick. Key innovation: "thought compression" — model learns to reason in fewer tokens. Contemplating mode runs parallel agents instead of longer chains. Beats Claude Opus 4.6 Max on HealthBench Hard (42.8 vs 14.8) but trails on ARC AGI 2 and GPQA. 🔗 X: https://x.com/Marktechpost/status/2042382842152567005 📖 Blog: https://ai.meta.com/blog/introducing-muse-spark-msl/

Anthropic Advisor Tool New API feature: Sonnet/Haiku can consult Opus mid-task, only when needed. Sonnet + Opus advisor gained 2.7 points on SWE-bench Multilingual while costing 11.9% less. Haiku + Opus doubled BrowseComp score (19.7% → 41.2%). One-line API change. Berkeley published a parallel paper validating the same "advisor models" concept. 🔗 X: https://x.com/akshay_pachaar/status/2042479258682212689 💻 GitHub: https://github.com/az1326/advisor-models 📄 Paper: https://arxiv.org/abs/2510.02453

Hermes Agent by NousResearch Trending hard — 6,485 stars in 24h (47,947 total). New /fast command for GPT-5.4 provider. 🔗 X: https://x.com/trending_repos/status/2042527463046287509 💻 GitHub: https://github.com/NousResearch/hermes-agent

Omarchy 3.5 — Full Linux on Dell XPS Panther Lake DHH's Omarchy distro is first to ship complete Linux compatibility for new XPS Panther Lake laptops. Dell made a custom unit with Omarchy keys replacing Windows and Copilot buttons. 🔗 X: https://x.com/dhh/status/2042565957097328945

🧪 INTERESTING REPOS

Multica — Open-Source Clone of Claude Managed Agents Self-hosted agent orchestration platform. Works with Claude Code, OpenAI Codex, OpenClaw, OpenCode. Isolated workspaces, reusable skills, shared human+AI board. 4,000+ stars in days. 🔗 X: https://x.com/DataChaz/status/2042505721292726393 💻 GitHub: https://github.com/multica-ai/multica

AI-Trader 2.0 — Agent-Native Trading Platform Multi-agent trading where AI agents collaborate, debate strategies, and form consensus. Agents develop specializations over time (technical analysis, sentiment, risk). 🔗 X: https://x.com/huang_chao4969/status/2042634193990226010 💻 GitHub: https://github.com/HKUDS/AI-Trader

OpenDataLoader PDF — High-Accuracy PDF Parser AI-ready data extraction from PDFs. 🔗 X: https://x.com/tom_doerr/status/2042462511937491137 💻 GitHub: https://github.com/opendataloader-project/opendataloader-pdf

Dexter — Autonomous Deep Financial Research Agent 🔗 X: https://x.com/tom_doerr/status/2042512186233245815 💻 GitHub: https://github.com/virattt/dexter

WorldMonitor — AI-Powered Global Situational Awareness Dashboard 🔗 X: https://x.com/tom_doerr/status/2042554504768360729 💻 GitHub: https://github.com/koala73/worldmonitor

LLMFit — Match LLM Models to Your Hardware Automatically recommends which models will run well on your specific hardware. 🔗 X: https://x.com/tom_doerr/status/2042648466967384565 💻 GitHub: https://github.com/AlexsJones/llmfit

DeepTutor — Turns Textbooks into Personalized Learning Systems Multi-agent reasoning transforms any textbook into an adaptive learning platform. 🔗 X: https://x.com/RoundtableSpace/status/2042572432704442583 💻 GitHub: https://github.com/HKUDS/DeepTutor

Strix — AI Agents for Autonomous Penetration Testing 🔗 X: https://x.com/tom_doerr/status/2042584901212356962 💻 GitHub: https://github.com/usestrix/strix

GitHub Spec-Kit — Kill Vibe Coding Describe what you want → AI writes specs → plans → builds. Works with every major AI coding agent. 🔗 X: https://x.com/AIGuideHQ/status/2042485307291877763 💻 GitHub: https://github.com/github/spec-kit

domo-actors v1.2.3 — Actor Model for TypeScript Plus domo-tactical for DDD patterns (Event Sourcing, CQRS) with PostgreSQL/Cloudflare D1 backends. 🔗 X: https://x.com/VaughnVernon/status/2042380866056569160 💻 GitHub: https://github.com/VaughnVernon

🎥 WORTH WATCHING

Gemma 4 31B — Autonomous Agent Demo by Google Zero-shot code generation, tool usage, multi-step debugging and recovery, learned multimodal output. Impressive for a 31B model. 🔗 X: https://x.com/googleaidevs/status/2042590030367973468

Ace Step 1.5 XL — Open Source AI Music Generator Reportedly beats Suno. Full tutorial available. 🔗 X: https://x.com/aisearchio/status/2042446695678750882 🎬 Video: https://youtu.be/UAlLD5fS7-c

Replit Vibe Coding Course (Free) 1-2 hour free course on building and deploying apps with AI-assisted coding. 🔗 X: https://x.com/RoundtableSpace/status/2042335489961898195 📖 Course: http://deeplearning.ai/short-courses/vibe-coding-101-with-replit

OpenAI Builders Unscripted — Agentic Workflows with Codex Ashe talks about moving fast with Codex, building in public, and agentic workflows. 🔗 X: https://x.com/OpenAIDevs/status/2042657797419262175

💡 TECHNIQUES & IDEAS

Rust-to-TypeScript Port = 3x Faster P99 Boris Cherny (@bcherny) had Opus port a Rust+NAPI file indexer to native TypeScript. Result: P99 latency got 3x faster. The prompt was beautifully simple: "compile our rust+napi to native ts, pass original tests, profile it, put up a PR." 🔗 X: https://x.com/bcherny/status/2042352725481177315

6,000+ Files: ESLint → oxlint Migration Detailed write-up on migrating a large TypeScript codebase from ESLint to oxlint (the Rust-based linter from the OXC project). 🔗 X: https://x.com/boshen_c/status/2042398792289566786 📖 Blog: https://www.ikorason.dev/why-we-migrated-6000-typescript-files-from-eslint-to-oxlint/

Encore's 67k-Line Rust Runtime for TypeScript Multi-language support, multi-threaded infra (HTTP, DB, pub/sub, tracing) in one process with Node.js. 9x higher throughput than Express. 🔗 X: https://x.com/rustaceans_rs/status/2042452548527772060

Claude + Obsidian as a Knowledge System Pattern: Obsidian as the data layer, Claude as the intelligence layer. Auto-maintained wiki, morning briefings, call transcript processing, contradiction flagging. Gets smarter over time. 🔗 X: https://x.com/alphabatcher/status/2042568855864639577

🔮 EMERGING TRENDS

Advisor/Routing Models Are the New Paradigm Both Anthropic (production API) and UC Berkeley (research paper) independently validated the same idea: small models route to big models only when needed. This is probably the most important cost/quality architecture pattern right now.

MirrorCode: AI Can Now Do Weeks of Engineering Epoch AI + METR's new benchmark tests AI on reimplementing entire software projects. Opus 4.6 reimplemented a 16,000-line bioinformatics toolkit — a task estimated at weeks of human work. The benchmark creators say it's "likely already saturated." 🔗 X: https://x.com/EpochAIResearch/status/2042624189421752346

The OpenCode vs Bun Drama OpenCode is moving away from Bun to Node despite Bun being "improved daily for their exact use-case." Community reaction is mixed — some see it as political (Anthropic-related), others as pragmatic. 🔗 X: https://x.com/rawkode/status/2042574454811693341

React Server Components CVE-2026-23869 (CVSS 7.5) New vulnerability in Next.js App Router — crafted HTTP requests cause CPU spike via deserialization. Affects Next.js 13.x-16.x with App Router. Patch: Next.js 15.5.15+ or 16.2.3+. Related to the devastating React2Shell (CVE-2025-55182) from late 2025. 🔗 X: https://x.com/imaimai17468/status/2042240483435639110

GPT-5.5 Launch Imminent Polymarket at 84% chance of GPT-5.5 launching this month. Meanwhile Mythos at 24% by June 30. 🔗 X: https://x.com/Polymarket/status/2042348640476303734

Google Skills Platform Google launched a free AI learning platform covering coding, Gemini, data analysis, AI security. 🔗 X: https://x.com/aiedge_/status/2042452560133701684 📖 Link: https://www.skills.google/

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