Google connects Search, Vids adds avatars: July 17 brief

Google Search can now send a shopper from a question to an action inside another service, a shift that matters to any business relying on online discovery. Google also brought AI presenters and plain-English editing to Vids. Meanwhile, its research notebook gained serious data skills, and Kimi put a huge new AI model on the market.

Google Search starts doing the next step

Google began connecting AI Mode to outside services in the US on July 16. AI Mode is the conversational part of Search that answers a question in a chat-style screen. According to Google's announcement, people can now add ingredients to an Instacart cart, pull up Canva templates, or save a YouTube Music playlist from their search session.

That is a bigger change than three new shortcuts. Google is reducing the distance between a recommendation and the action that follows it. A customer may soon go from asking for ideas to choosing a product or creating something without visiting several websites along the way. Google says more partners are coming.

For owners, the immediate job is to study the information these systems can read. Clean product details, current business profiles, and consistent names give an AI answer less room to misunderstand you. Our guide to how AI systems recommend businesses covers that groundwork, while our AI search coverage tracks the wider change.

Google Vids creates a presenter from one selfie

Google Vids added two production features on July 16: video generation and editing through written instructions, plus a digital presenter made from a selfie and a short voice recording. Google says the editor can change a background, correct lighting, or add effects without rebuilding the clip.

The features are available to Google AI Pro and Ultra subscribers and Google Workspace business customers. Personal presenters are restricted to the account holder's likeness, require the user to be at least 18, and are limited by region. Generated clips carry SynthID, Google's invisible mark for AI-made media.

This lowers the effort needed to make repeatable training updates, product explanations, and internal announcements. It also makes generic presenter videos easier for every competitor to produce. Use the time saving on scripts and examples that only your company can supply. Our comparison of AI video tools for small brands explains where presenter tools fit beside editors for real footage.

Gemini Notebook moves from summaries to analysis

Google renamed NotebookLM to Gemini Notebook on July 16 and gave eligible notebooks a secure online computer that can write and run code. In plain words, it can now calculate from source material instead of only reading and summarizing it. The company announcement says the feature is available to Ultra and certain Workspace business customers, with Pro access coming on the web over the next few weeks.

The product remains separate, but notebooks now sync with the Gemini app and are due to appear inside AI Mode in Search. Google reports more than 30 million users and 600,000 organizations.

The useful business case is narrow and real: load approved reports, price sheets, or research, then ask for calculations tied to those files. Still check the math and the cited source before a number reaches a customer. Sensitive records also need the same access rules you would apply to any other cloud service.

Kimi puts a giant open model on the price sheet

Kimi released K3 on July 16, calling it a 2.8-trillion-parameter model. Parameters are the internal values an AI learns during training. The number signals scale, though it does not prove that the model is right for a business task. Kimi's launch post says the model can handle text, images, and long documents, while its downloadable model files are due by July 27.

The hosted interface costs $3 per million input tokens when earlier information cannot be reused, and $15 per million output tokens. Tokens are the small pieces of text an AI reads and writes. Those prices matter mostly to companies paying a developer to build AI into a product or internal system.

Small teams should treat K3 as pricing pressure, not a migration order. A cheaper model saves little if staff must recheck weak answers or rebuild a working process. Test it against a fixed set of your own documents and tasks before comparing bills.

Worth watching

• Google says more connected services will enter AI Mode, but it has not named the next partners.
• Gemini Notebook is coming to AI Mode in Search, which could turn saved company research into part of everyday searching.
• Kimi plans to release K3's model files by July 27, when independent testing should become easier.