today's signal · no scroll
the brief
Agent tooling took a step forward: Claude’s auto mode broadened access, Railway gave agents safe computers, Grok added real connectors, and Claude Code exposed MCP-level costs. Security stayed hot with Perplexity’s Bumblebee scanner and Anthropic’s Glasswing surfacing 10k+ critical vulns. Research pushed on latency with diffusion-style LMs, while hardware constraints loomed large in the background.
Anthropic’s Project Glasswing and partners report 10k+ high/critical flaws found in essential software in a month—expect a surge in coordinated disclosures and patching workload.
Last month we launched Project Glasswing, our collaborative AI cybersecurity initiative. Since then, we and our partners have found more than ten thousand high- or critical-severity vulnerabilities in essential software.
A read-only macOS/Linux scanner for risky packages, extensions, and AI configs, Bumblebee also hooks into Computer to trigger deeper scans as new threats emerge.
Today we're open-sourcing Bumblebee, a read-only scanner for macOS and Linux. It checks developer machines for risky packages, extensions, and AI tool configs. Connected to Computer, it can trigger deeper scans whenever a new supply-chain risk emerges. github.com/perplexityai/b… pic.x.com/wXauD4wDOT
Auto mode is now available to Pro users and supports Sonnet 4.6 alongside Opus 4.7, making hands-off task runs more accessible.
Two updates to auto mode: · Now available on the Pro plan · Sonnet 4.6 is now supported, alongside Opus 4.7 Shift+tab, and let Claude run.
DeepSeek makes its 75% discount permanent, with near-free cache pricing—resetting cost expectations for high-end inference and pressuring competitors.
We are making our discount permanent! 🎉 Enjoy building with DeepSeek-V4-Pro and bring your innovative ideas to life! 🚀 pic.x.com/V8atbTaogH x.com/deepseek_ai/st…
New connectors expand Grok’s automation surface from site building and creative tooling to market data, enabling richer end-to-end agent workflows.
New connectors available Build sites with Vercel, create anything in Canva, design decks in Gamma, and tap into market data from S&P Global. pic.x.com/UbRINA2CLw
Priority Boarding users get sandbox VMs for the chat agent plus HA static egress and org guardrails—bringing safer real-computer control to agents.
We gave the agent a computer Changelog #0291 • Sandbox VMs for the chat agent to Priority Boarding • HA static egress to Priority Boarding • GitHub org guardrails railway.com/changelog/2026…
v2.1.149 surfaces per-category and per‑MCP‑server usage drivers, adds UI polish and enterprise controls—giving teams clearer cost and limit visibility.
v2.1.149 — What's changed /usage now shows a per-category breakdown of what's driving your limits usage — skills, subagents, plugins, and per-MCP-server cost /diff detail view can now be scrolled with the keyboard (arrows, j/k, PgUp/PgDn, Space, Home/End) Markdown output now renders GFM task list checkboxes (- [ ] todo / - [x] done) instead of plain bullets Enterprise: added the allowAllClaudeAiMcps managed setting to load claude.ai cloud MCP connectors alongside managed-mcp.json Fixed a Powe...
Hugging Face details NVIDIA’s Nemotron‑Labs work on diffusion-style LMs targeting near speed‑of‑light text generation via highly parallelizable sampling with competitive quality.
Towards Speed-of-Light Text Generation with Nemotron-Labs Diffusion Language Models
April 2026 CommonCrawl and URL index on Hugging Face Buckets are directly queryable over hf:// with DuckDB, enabling ad‑hoc SQL at web scale without downloads.
You can now run SQL over 2.19 BILLION web pages. Zero download! @CommonCrawl April 2026 crawl + URL index are on @huggingface Storage Buckets. @duckdb reads it straight over hf:// I counted all 2.19B in ~35s. Or point your own agent at it 👇 huggingface.co/spaces/davanst… pic.x.com/3NdWEuS0k2
Simon Willison spotlights David Oks’ argument that AI‑driven DRAM/NAND constraints will raise consumer electronics costs, a planning reality for AI builders.
The memory shortage is causing a repricing of consumer electronics — <p><strong><a href="https://davidoks.blog/p/ai-is-killing-the-cheap-smartphone">The memory shortage is causing a repricing of consumer electronics</a></strong></p> David Oks provides the clearest explanation I've seen yet of why consumer products that use memory are likely to get significantly more expensive over the next few years.</p> <p>The short version is that memory manufacturers - of which there are just three remaini...
Matt Pocock argues teams should maintain an explicit registry of agreed test seams so agents don’t make poor testing choices, improving safety and maintainability.
Another layer of documentation I'm considering (along with CONTEXT.md and ADR's) is a list of all the agreed test seams in the app Agents simply cannot be trusted to make good decisions about what to test, and at what seam. For every small change, they extract out only what