Today’s lead · techcrunch.com
Claude becomes a persistent Slack teammate
Anthropic launched Claude Tag in beta for Claude Team and Enterprise customers, giving Slack channels a shared Claude identity with scoped memory, admin-controlled channel/tool access, task delegation, and an ambient mode that can surface updates without being explicitly prompted. The Claude Code team says it has been using the pattern internally all year and now attributes 65% of its product-team code to Claude Tag-assisted workflows. This caught my attention because it is the clearest product move yet from chatbot-as-destination to agent-as-org-member with permissions, memory, and public work traces.
Top signals
5 more
Agent skills have a supply-chain problem
Security firm AIR says it pushed a fake agent skill through a popular skill marketplace, got clean scanner results, bought Instagram distribution, and reached roughly 26,000 agents by hiding the risky instructions behind an external URL it could later change. The uncomfortable lesson is that skills are executable supply-chain artifacts, not just prompt snippets, so scanners that only inspect the submitted package are looking in the wrong place.
Model routing is becoming an online learning loop
A new Agent-as-a-Router paper frames coding-model selection as a Context → Action → Feedback → Context loop instead of a one-shot classifier, with a 15.3% relative gain from adding task-dimension performance stats and a CodeRouterBench environment covering about 10,000 tasks across 8 frontier LLMs. I like this because it matches the production reality of routing: the best router should learn from verified execution outcomes, not just pick a model from stale benchmark priors.
Executor compresses the MCP tool swamp
Executor, an open-source MCP gateway joining YC S26, now has a self-hostable Docker build, desktop app, chat-based setup, multi-account support, and 2,000 GitHub stars. The technical hook is context efficiency: it claims to expose 1,640 connected tools as one executable interface, cutting tool context from about 278,800 tokens to roughly 1,044 tokens until the agent actually needs a schema.
Legal evals expose the all-or-nothing gap
Vals AI released Legal Research Bench, an eight-area U.S. law benchmark graded by practicing lawyers with a strict all-pass rubric rather than forgiving partial credit. Claude Opus 4.8 leads at 43.8% all-pass accuracy, followed by GPT-5.5 at 40.4% and Claude Sonnet 4.6 at 38.5%, while top models sit around 80% under partial-credit scoring, which is a useful warning for anyone designing evals for high-stakes agent work.
Claude’s kernel demo raises the bar for coding-agent evidence
Security startup Tolmo published what it says is a transcript-level account of Claude Fable 5 creating a booting NT-shaped Rust kernel from an empty directory in 38 minutes of active model work, then extending it over eight days with Claude Opus 4.8 until it could load unmodified Windows drivers and run real Windows binaries. I’m treating the claim cautiously, but the interesting part is the evidence format: long-horizon systems work needs boot logs, self-tests, traces, and bug root-cause transcripts, not just a flashy screenshot.
Tools & repos
3 selectedcalesthio/OpenMontage
World's first open-source, agentic video production system. 12 pipelines, 52 tools, 500+ agent skills. Turn your AI coding assistant into a full video production studio.
ZhuLinsen/daily_stock_analysis
LLM 驱动的多市场股票智能分析系统：多源行情、实时新闻、决策看板与自动推送，支持零成本定时运行。 LLM-powered multi-market stock analysis system with multi-source market data, real-time news, decision dashboard, automated notifications, and cost-free scheduled runs.
mukul975/Anthropic-Cybersecurity-Skills
817 structured cybersecurity skills for AI agents · Mapped to 6 frameworks: MITRE ATT&CK, NIST CSF 2.0, MITRE ATLAS, D3FEND, NIST AI RMF & MITRE F3 (Fight Fraud) · agentskills.io standard · Works with Claude Code, GitHub Copilot, Codex CLI, Cursor, Gemini CLI & 20+ platforms · 29 security domains · Apache 2.0
Blogs worth your time
1 readsMidjourney wants to delete 30% of all death...
From youtube.com.
Funding & acquisitions
5 movesBluerails Discovery
The rails AI agents use to find and pay you
OpenArt Director
Direct cinematic videos through chat
Cotypist
Local AI Autocomplete in your voice, anywhere on your Mac
Latitude
Fix what's breaking in your AI agent
Thumbmagic
AI thumbnail generator trained on top-performing thumbnails
Bengaluru radar
0 eventsThere are no relevant Bengaluru events to highlight today.