Google rolls out Gemini voice models that speak while reasoning

Relay, a five-year-old alternative to Zapier, has shut down entirely. It used AI agents to automate email and task workflows, putting its closure directly in the path of customers’ routine work.

Lauren Forristal reports in TechCrunch that Relay struggled as OpenAI, Google and other larger platforms built similar automation into their own tools. This is a complete shutdown, rather than a feature being withdrawn from an otherwise continuing service.

Notion Mail has a closing date

Notion Mail is scheduled to shut down on 22 September. Launched in April 2025, the product used AI to organise and automate inboxes. Notion found that users were instead employing separate AI agents to handle their email.

The two closures reflect different pressures. Relay faced automation being built into larger platforms; Notion Mail faced users choosing agents outside its email product. Neither outcome means the underlying jobs have disappeared.

For users, the useful next step is to list the email actions and task workflows that depended on either service, then check their replacements individually. A replacement that handles messages may not reproduce the task automation attached to them.

Customers need to replace a working dependency, not merely choose another chatbot. Relay users face a closed service; Notion Mail users have a stated shutdown date around which to plan changes to inbox routines.

Speaking at Salesforce’s Dreamforce conference, Nvidia chief executive Jensen Huang argued that AI remains a human-built computing system and can be controlled through engineering and existing laws.

Julie Bort’s TechCrunch report describes his rejection of new AI regulation. Huang’s case is that market pressure already gives companies a reason not to release unsafe products, and that speed and safety need not be competing goals.

Companies would decide when to pause

Under the approach Huang advocates, businesses should hold back a product if they lack confidence in its functionality, capabilities or safety. They should move quickly otherwise, pausing when a product appears unsafe or the company loses control.

That puts the initial release decision with the company building the system. Huang is arguing against additional legal requirements, not against safety work or the application of existing law.

His remarks did not establish an industry-wide safety scheme. Nor does confidence before release demonstrate that a product will avoid harm in use. The concrete policy question is how much to rely on companies’ own checks, alongside existing legal remedies, rather than impose new obligations.

For businesses building or buying AI, Huang’s position would leave release checks with suppliers under existing law rather than add AI-specific rules. It is an argument about responsibility, not permission to relax current obligations.