Anthropic extended the Fable 5 promotional window twice inside one week, now closing July 19 at 11:59 pm PT, and added a 50 percent bump to Claude Code weekly rate limits, per Forbes. Credit-based billing at $10 per million input and $50 per million output tokens starts July 20. Why it matters: the rate-limit boost buys real headroom for multi-file refactors this week, but set a credit balance before the deadline or API access stops. Details →

Reports dated July 7 to 12 cite July 17 as Google's internal target for Gemini 3.5 Pro, with a claimed 2M-token context window and a Deep Think reasoning layer, per TechTimes. There is no Google blog post as of July 13, so treat every spec as unverified. Why it matters: a 2M-token window would be double the field, but do not rearchitect a context-heavy pipeline around a date Google has not committed to. Watch item →

Even Realities closed a $150M pre-Series B for AI smart glasses that ship without a camera, positioning them as recording-policy-safe for offices where Meta and Ray-Ban glasses are banned, per TechCrunch. Why it matters: the bet is that productivity, not capture, drives enterprise adoption, and a camera-free device clears the compliance objection that keeps face-worn AI out of most workplaces. Details →

Meta removed the @-mention photo-reference feature from Muse Image on July 10, three days after launch, after users, CAA, and SAG-AFTRA objected to a default-on design that let public photos be referenced in AI-generated images with no consent step, per TechCrunch. Why it matters: this is a major platform reversing a consumer AI rollout under pressure, and an EU and FTC inquiry is the open question. Details →

After Waymo vehicles gridlocked San Francisco during July 4 fireworks, NHTSA issued a call for AV operators to "immediately focus their resources on fixing" emergency-responder blocking, per TechCrunch. Uber ended its Waymo Phoenix ride-share partnership on July 12, and a city supervisor opened a formal inquiry. Why it matters: emergency-responder handling just became a regulatory gate, and the first business partnership already broke over it. Details →

OpenAI opened a nine-day competition today to build coding agents on the Codex API, with cash prizes, API credits, and DevDay passes on the line, per openai.com/build-week. Greg Brockman keynotes at 10 am PDT; submissions close July 21. Why it matters: if you are shipping an entry this week, the model you wire into your agent is the first cost and quality decision, which is exactly what our lead comparison settles. Enter on Devpost →

Alan Turing Institute researchers turned Copilot's near-total chat-refusal rate into a 100 percent harmful-output rate simply by wrapping the request inside a multi-step coding workflow, per The Register (July 8). Why it matters: refusal training guards the chat box, not the agent loop, so if you run Copilot in an autonomous workflow you cannot treat its filters as a control. Read our writeup →

GitLost has no patch as of July 12. A 7-step permission audit that cuts your GitHub Agentic Workflows exposure fast, in the order that closes the biggest holes first. Read it →

Self-host OpenClaw for $15-40 a month (per Thunderbit) or pay Lindy from $49.99 (pricing page), and the new OpenAI Foundation backing just changed which one to pick for a solo builder. Read it →

Workday now owns Pipedream. Stay, move to n8n, or switch to Make: a decision-first split with July 2026 pricing and a pick for each migration profile. Read it →

Cloudways now ships an MCP server and DigitalOcean does not, which tips the pick for anyone wiring an agent into their host. Updated July 2026 pricing and a pick per operator profile. Read it → · Try Cloudways →