WeatherNext 3 shipped September 3. It learns from live geostationary satellite mosaics rather than physics runs carrying a six-hour data lag, so it emits a new forecast every hour at 5-kilometer resolution for surface variables like temperature and moisture. The previous generation ran a 25-kilometer grid in 6-hour increments; Google's announcement claims up to 50 percent more accurate precipitation forecasts a day or more out. Query it today in BigQuery, Earth Engine, Cloud Storage, or the Maps Platform Weather API.

Why it matters. Any agent that schedules field work, routes freight, or bids energy now has an hourly, high-resolution grounding signal it can fetch as a tool call, with no model to host and no meteorologist on staff. Price the Maps Platform Weather API calls before you wire an hourly fetch into a loop that runs per user. Read what changed under the hood.

Sanders and Casar announced the Ban Artificial Superintelligence Act on September 3. The release describes a permanent ban on systems that surpass human intelligence or can subvert shutdown commands, a pause on advanced AI development until a new cabinet-level regulator writes review rules, up to 20 years in prison for individuals, and what it calls the corporate death penalty for entities.

Why it matters. This does not pass in this form. What moved is the frame: a sitting senator has now put criminal liability, not licensing or disclosure, at the center of a federal AI bill. That text becomes the template state legislatures copy in the next session, which is where compliance cost actually lands on you. Read the bill breakdown.

Visual automation for the boring half of an agent stack: the triggers, the retries, the handoffs to Slack and Sheets that you should not be writing as bespoke glue code. Good fit if your Claude Code agent is smart but your plumbing is a cron job and hope. Build the first scenario in Make.

Turns a URL into clean markdown an LLM can actually read, with crawl and structured-extract endpoints on top. If the DseWiki story made you look harder at how your agents reach the web, routing retrieval through one scraping API is easier to scope and log than letting every agent hold a browser. Point Firecrawl at your first domain.