The new /automate skill takes a goal in plain English and wires up the triggers, instructions, and tools itself; v3.8 also adds five GitHub trigger types (issue comments, inline PR-review comments, review submissions, thread-status changes, workflow-run completions). If you live in PR review, the "auto-fix PR review comments" Marketplace template is the one to steal first. Cursor changelog.

xAI announced the integration at the 2026 Data + AI Summit, putting Grok in front of enterprise data teams building agents on Databricks. The read: xAI is going straight for the enterprise build-and-deploy lane Anthropic and OpenAI already contest, where the model lives next to the warehouse. xAI's announcement.

AI agents now run inside Photoshop, Premiere, Illustrator, and InDesign, with Anthropic, OpenAI, and Microsoft models all selectable. This is the first multi-lab agent rollout across a full production suite, which means the model becomes a swappable component inside the tools designers already open every day. Adobe's announcement.

OpenAI tied its updated health intelligence to a new Global Physician Network of 262 doctors and a HealthBench eval. The shift from general assistant to clinician-validated health tool is a deliberate liability posture, and it raises the bar any competitor's health feature now has to clear. OpenAI's writeup.

The LLM-ready scraping API just shipped a Keyless tier: 1,000 scrape credits a month with no API key and no signup, the lowest-friction way to feed live web data into a Claude Code or Cursor workflow. Paid plans start at $16/mo for 5,000 credits, per Firecrawl's pricing page. Try Firecrawl.

Kit's free Newsletter plan covers up to 10,000 subscribers, not the 1,000 the slider default implies, per Kit's pricing page. Creator at $33/mo (annual) unlocks Kit MCP: 76 read/write endpoints that connect Claude, ChatGPT, and Cursor straight to your list. Try Kit.

NVIDIA's open-source SkillSpector scans the CLAUDE.md, AGENTS.md, and SKILL.md files an agent executes with OS-level privileges, flagging 26.1% of skills as vulnerable and 5.2% as likely malicious, per the SkillSpector README. Run skillspector scan ./CLAUDE.md --score for a 0-100 risk score across 64 detection patterns in 16 categories. If you run agents in production, read this first. Read the guide.

Keyless gives you 1,000 credits a month with no account; Hobby ($16/mo) jumps to 5,000 and Standard ($83/mo) to 100,000, at one credit per scraped page. Start Keyless to prototype, move to Hobby the day you script it. Read it · try it.

Kit MCP left beta on June 11 with 76 endpoints and 3,000-plus creators already on it. If you want your agents writing to your list, the call is simple: stay free to 10,000 subscribers, upgrade to Creator only when you need the MCP. Read it · try it.

Scale unlocks the Ad Network, paid subscriptions at a 0% platform fee, and automations. At the documented $5 CPM per beehiiv's Ad Network page, it breaks even around 8,600 unique opens per issue (43 divided by 0.005). Below that, stay on Launch; above it, Scale prints. Read it · try it.

Per a Zero Click by Profound session: Moving Traffic Media's Jonathan Clark found Claude pulls Brave's top 10 directly, searches on only 36.6% of prompts (versus about 90% for ChatGPT), and overlaps Google rankings 64% of the time, per Search Engine Land's report. The lever: rank in Brave for "best X" and "X vs Y" queries, where Claude searches most. Read it.