Confirmed
Databricks ships FILE, a native column type for storing unstructured data inside lakehouse tables
Databricks released FILE, a beta column type that stores documents, images, audio, and video as governed rows inside standard lakehouse tables instead of external object storage. Deleting a row deletes the underlying file in storage, and the same access controls and security policies that apply to structured tables now apply to raw files. Databricks says it is building FILE support directly into Parquet, Delta Lake, and Iceberg, plus Lakeflow connectors for SharePoint and Google Drive, across AWS, Azure, and GCP.
In plain termsA 'lakehouse table' mixes traditional structured data, rows and columns, with big unstructured files in one governed system, and this update means an actual PDF or image can now live inside a table row instead of just a link pointing out to separate storage.
Why it mattersTeams building AI features on top of Databricks now get row-level governance over unstructured files instead of tracking loose object-storage permissions by hand.
Our takeThis is Databricks extending table-level governance to the exact kind of unstructured data AI pipelines already depend on, rather than leaving file access controls as a separate, easy-to-forget layer. Expect other data platforms to follow with their own native file types once teams start asking why row-level security stops at structured data.
Confirmed
Google confirms a Merchant Center outage, still hasn't shared a cause five days in
Google's Merchant Center status dashboard shows an active incident that began August 6 at 13:06 UTC, with a single update at 17:46 UTC the same day saying only that the issue is under investigation. As of August 11, Google has not disclosed the cause, which features are affected, or an expected resolution time. The incident could disrupt merchants managing product listings for Google Shopping, Performance Max, and free listings.
In plain termsA 'status dashboard' is Google's own uptime page for a product, separate from the tool itself, that shows whether Google has publicly acknowledged something is broken.
Why it mattersMerchants troubleshooting feed or listing problems this week should check Merchant Center's status dashboard before assuming the mistake is in their own feed.
Our takeFive days on a single 'we're investigating' update is a long silence for a platform merchants depend on to manage live product listings, and it says as much about Google's incident-communication habits as the outage itself. Watch whether Google posts a resolution notice or lets the incident quietly age off the dashboard without one, since that pattern would be the more consequential story.
Confirmed
Yelp brings restaurant reservations and waitlists directly into ChatGPT
Yelp launched an integration letting ChatGPT users book a table or join a waitlist at thousands of U.S. and Canada restaurants without leaving the chat. The feature runs through Yelp Guest Manager and builds on Yelp's existing OpenAI partnership. Yelp frames itself as the reservation and waitlist layer behind multiple AI assistants at once, citing similar integrations with Apple Maps, Amazon's Alexa+, Microsoft Bing, DuckDuckGo, and Yahoo.
In plain terms'Yelp Guest Manager' is the backend software restaurants use to manage tables and waitlists, and this update means that same backend now also handles bookings a customer starts by typing into ChatGPT instead of opening Yelp.
Why it mattersA restaurant's booking availability now needs to be correct inside Yelp's system specifically, since that data now surfaces transactions across at least six AI assistants instead of just Yelp's own app.
Our takeYelp is betting on becoming the fulfillment layer behind conversational interfaces rather than a destination competing with them, the posture aggregators tend to take toward any dominant new front end. The list of six assistant partners matters more than the ChatGPT integration alone, since the strategy is building infrastructure once and plugging it into whichever assistant wins distribution.
Confirmed
Meta releases Muse Glimmer, a 30B open-weights model built for local agentic tasks
Meta released Muse Glimmer, a 30-billion parameter open-weights vision model under an Apache 2.0 license, a step up from Meta's earlier, more restrictive Llama licenses. Meta optimized it for agentic task completion, tool calling, and multi-step reasoning, citing benchmarks including SWE-Bench and tau-Bench. A quantized 18.16 GB version runs on a single consumer GPU through LM Studio, small enough to leave headroom on a machine with 32 GB of RAM or more.
In plain terms'Open-weights' means Meta published the actual trained model file for anyone to download and run on their own hardware, unlike a closed model such as GPT or Gemini that only runs on the provider's servers.
Why it mattersA capable open model that runs locally on consumer hardware lowers the cost of running SEO or content agents that need tool use and multi-step reasoning without a per-token API bill.
Our takeA 30B model that fits on a single consumer GPU and handles multi-step tool calling is exactly the size class that makes local agents practical for individual practitioners, not just labs with GPU clusters. Watch whether Meta's cleaner Apache license, versus the more restrictive old Llama terms, changes how freely this actually gets built into commercial agent tooling.
Observed
Search Console's generative AI performance report looks broadly live, though Google hasn't confirmed a full rollout
Barry Schwartz of Search Engine Roundtable reports the Generative AI performance report, introduced in June 2026 for a subset of properties, now shows up across roughly 25 of his own. Google has not announced a full rollout; its help documentation still says it is 'rolling out this report to a subset of website owners, allowing for thorough testing before rolling it further.' The report covers impressions, pages, countries, and devices for AI Overviews and AI Mode, minus clicks and queries.
In plain terms'Impressions' here just means how often a page showed up inside an AI-generated answer, which is different from a click, so this report can only tell you that you were seen, not whether anyone acted on it.
Why it mattersAnyone not yet seeing the report in their own properties should check back rather than assume something is broken, since Google's own documentation still frames this as a staged rollout.
Our takeA practitioner's spot-check across two dozen properties is a reasonable signal, but it isn't the same as Google confirming general availability, and Search Console's own help page hasn't caught up either way. Treat this as directionally true and worth checking your own properties, not as a confirmed platform change yet.