Dev Digest — March 11, 2026
🔥 HOT RELEASES
Claude Code /btw — Side Conversations While Claude Works
Anthropic shipped /btw for Claude Code, letting you have side-chain conversations while your main task keeps running. Ask questions, get context, without interrupting the workflow. 23k+ likes — the most talked-about dev feature today.
🔗 X: https://x.com/trq212/status/2031506296697131352
Gemini CLI Plan Mode Google launched Plan Mode in Gemini CLI — a read-only mode that lets you safely explore and plan before implementation. Supports MCP tools, Agent Skills, and is fully extensible. 🔗 X: https://x.com/geminicli/status/2031761736232050707
NVIDIA Nemotron 3 Super (120B MoE) NVIDIA dropped Nemotron 3 Super — a 120B parameter MoE model with only 12B active params, 1M token context, multi-agent reasoning, and royalty-free weights. Open-weight, deployable locally. OpenHands already tested it. 🔗 X: https://x.com/OpenHandsDev/status/2031764879019839577 🔗 HuggingFace: https://huggingface.co/collections/nvidia/nvidia-nemotron-v3
Cursor Marketplace Launch Cursor just launched a marketplace — AI agents can now access the entire dev stack through extensions. 🔗 X: https://x.com/Abhinavstwt/status/2031786459980800502
Replit Agent 4 Replit released Agent 4 with canvas mode, parallel agents, real-time collaboration, and one-click app deployment. Moving from "AI assists coding" to "AI collaborates on the whole product." 🔗 X: https://x.com/iamfakhrealam/status/2031772738692731075
Ramp Agent Cards — Credit Cards for AI Agents Ramp launched "Agent Cards" — virtual credit cards designed for AI agents with spend limits, merchant controls, no exposed card numbers, and native API/MCP/CLI support. 🔗 X: https://x.com/sytaylor/status/2031804012886917519
🧪 INTERESTING REPOS
OpenPencil — Open-Source Figma Alternative AI-native design editor that opens .fig files, has built-in AI chat (90+ tools), real-time P2P collaboration via WebRTC, headless CLI, MCP server for AI agents, auto layout, Tailwind CSS export. MIT licensed, ~7MB. 🔗 X: https://x.com/CamilleRoux/status/2031448731086733815 💻 GitHub: https://github.com/open-pencil/open-pencil
Hermes Agent by Nous Research Self-improving AI agent with a built-in learning loop — creates skills from experience, searches past conversations, builds a model of you across sessions. Runs on any model (OpenRouter, OpenAI, etc.), works on a $5 VPS. Telegram/Discord/Slack/WhatsApp native. 🔗 X: https://x.com/alexdolbun/status/2031692872572178665 💻 GitHub: https://github.com/NousResearch/hermes-agent
GitHub SecurityLab Taskflow Agent GitHub's open-source AI-powered vulnerability scanning framework. Uses task-based agent workflows for security analysis. 🔗 X: https://x.com/AISecHub/status/2031746508958085580 💻 GitHub: https://github.com/GitHubSecurityLab/seclab-taskflow-agent
Claude Terminal Free, open-source terminal client for Claude Code. Windows/macOS/Linux. 🔗 X: https://x.com/ClaudeTerminal_/status/2031750721763221824 💻 GitHub: https://github.com/Sterll/claude-terminal
InsForge — Backend for Agentic Development Open-source backend (~2K stars) providing databases, auth, storage, model gateway, and edge functions — all through a semantic layer that AI agents can understand and operate end-to-end. 🔗 X: https://x.com/RitikaAgrawal08/status/2031750718777119012 💻 GitHub: https://github.com/InsForge/InsForge
AlphaClaw — OpenClaw Deployment Harness Graphical harness on top of OpenClaw with UI for install/config/maintenance, Google Workspace OAuth, auto GitHub backups, crash detection, token analytics. MIT licensed. 🔗 X: https://x.com/czverse/status/2031505964667646122
Multi-Agent LLM Hallucination Detector Multiple agents critique model output to detect hallucinations. HuggingFace demo available. 💻 GitHub: https://github.com/Shreya831/multi-agent-llm-hallucination-detector
🎥 WORTH WATCHING
Claude Code + Ollama = Free AI Coding Forever (Tutorial) Full tutorial on running Claude Code with local Ollama models — zero API costs. 🔗 X: https://x.com/intheworldofai/status/2031613640852304374 🎬 YouTube: https://youtu.be/mN2VUw5Fb3E
4 AI Models You Should Pay Attention To (Turing Post) Covers Phi-4-reasoning-vision-15B, Olmo Hybrid (transformers + linear RNNs), DC-DiT (AMD diffusion transformer), and Helios (real-time long video generation). 🔗 X: https://x.com/TheTuringPost/status/2031662991004033170 📰 Article: https://www.turingpost.com/p/fod143
CopilotKit — Add AI to Your App in 5 Minutes Tutorial on building AI coworkers inside your product — app state awareness, user actions, generative UI. 🔗 X: https://x.com/CopilotKit/status/2031752980605935656 📖 Docs: https://docs.copilotkit.ai/built-in-agent/quickstart
💡 TECHNIQUES & IDEAS
Karpathy: Stop Coding, Start Managing Agents The man who taught a generation to build neural nets from scratch now says the future is managing AI agents on a tmux grid. "Software engineer" = "person who watches 12 AI agents on 6 monitors." He also casually mentioned forking entire companies, not just code. 🔗 X: https://x.com/TukiFromKL/status/2031784108301472030
The Amazon AI Outage Amazon pushed "AI-first" coding after 30k layoffs. On March 5, AI-generated code slipped through review and crashed checkout, login, and pricing for ~6 hours. 21k+ Downdetector reports. Mandatory engineering meeting followed. "Vibe coding" just broke one of the biggest websites on earth. 🔗 X: https://x.com/abhijitwt/status/2031748465458643258
Figma Make → Manual Tweak → Vibe Code Workflow New workflow pattern gaining traction: Figma Make for base designs, manual iteration, import into AI tool, vibe code frontend/backend, publish. 🔗 X: https://x.com/jiayun_studio/status/2031545541398605886
🔮 EMERGING TRENDS
MCP Roadmap Published The MCP team published their official roadmap with priority areas: transport evolution (stateless scaling, session migration, Server Cards), agent communication (Tasks primitive), and keeping the protocol small. SEPs aligned with the roadmap get expedited review. 🔗 X: https://x.com/dsp_/status/2031421005093044662 📖 Roadmap: https://modelcontextprotocol.io/development/roadmap
Agent Payment Infrastructure Ramp's Agent Cards signal a new category: financial infrastructure purpose-built for AI agents. Virtual cards with spend limits, merchant controls, API-native. The thesis: agents will use traditional payment rails (not crypto) today, with stablecoin settlement eventually coming underneath.
The "Agent Swarm" Pattern Goes Mainstream
Between Karpathy's vision, Replit Agent 4's parallel agents, Claude Code's /btw multitasking, and Cursor's marketplace — the pattern of managing multiple AI workers simultaneously is becoming the default development paradigm, not an experiment.
Open-Weight Models Getting Competitive NVIDIA Nemotron 3 Super, Phi-4-reasoning-vision, Penguin VL from Tencent — the open-weight space is shipping production-quality models weekly. The gap with frontier closed models keeps shrinking.
Full digest: https://digest.99.cooking