Wednesday, 16 September 2026

OpenAI is testing "Sponsored Agents" in ChatGPT, letting a user click an ad and open a clearly labelled conversation with a business-sponsored agent [1][2]. The test runs with a set of selected advertisers in the United States [1][2]. In that side conversation a person can ask follow-up questions — the example given is a shopper who sees a dining table and wants to know whether it fits their room, how many people it seats, or how to care for the finish — then click through to the advertiser's site [1]. OpenAI says the sponsored chat stays separate from ChatGPT's own answers and from the user's original conversation [2].

Alongside the agents, OpenAI is opening tools for advertisers. HubSpot is named as its first CRM partner and Shopify as its first ecommerce partner, both from 16 September; the Shopify app is free and reaches other ChatGPT Ads markets on 23 September [1][2]. An opt-in text feature can rewrite an advertiser's existing headlines to fit a conversation and auto-translate ad copy into the reader's language [1][2].

On data, OpenAI says ads are labelled, kept distinct from its answers, and that advertisers receive only aggregate numbers such as views or clicks, not access to chats [2]. Early advertisers named in the test include Newegg, Best Buy, Lowe's and VistaPrint [1].

Google has rolled out early access to a Model Context Protocol (MCP) server for Google Home, which lets outside AI agents work with the devices and event history in a user's home [1][2]. The named agents include Claude, ChatGPT, Hermes, OpenClaw and Google's own Antigravity [1]. Through them a person can use plain language to monitor activity, review camera summaries, control connected devices and build custom dashboards [1].

Access is gated. The MCP server is limited to subscribers of Google Home Premium Advanced, the $20-a-month tier that also adds longer video history and detailed alerts, and the early build reaches US subscribers over the coming weeks [1][2]. It reaches any device in the Google Home ecosystem, including Nest doorbells and thermostats and Matter gadgets such as light bulbs [1]. Setup is not one click: a user creates a Google Cloud project, configures it for Home MCP, hands the details to their chosen agent, then signs in and grants permissions [1].

The move matters because MCP is becoming the common plug for agents, and Google's own developer centre now carries the Home setup guide [1]. It also means the assistant that answers in your kitchen no longer has to be Gemini.

Google declined to comment on wider rollout plans [1].

Amazon has launched Alexa+, its generative-AI assistant, in India, starting with a free early-access period from 16 September [1][2]. It understands English, Hindi and "Hinglish," and a user can switch between Hindi and English mid-sentence; Amazon says more Indian languages will follow [1][2]. The assistant holds longer, context-aware conversations and handles multi-step requests such as ordering groceries or controlling smart-home devices [1][2].

Pricing kicks in after early access. Alexa+ stays free for Amazon Prime members, while Prime Lite, Prime Shopping Edition and non-Prime customers pay ₹2,000 a month unless they move to a qualifying Prime plan [1][2]. It works across Echo devices, Fire TV and the Alexa app, with browser access noted as coming later [2].

India is a large test for the paid assistant: the country has more than 600 million Hindi speakers, and Amazon has tied Alexa+ to India-based services, from food delivery and travel booking to music streaming [1][2]. Amazon says its smart-home devices in India grew about 20% year over year, the backdrop for pushing a more capable assistant there [1]. The launch follows initial testing that Amazon began earlier in 2026 [1].

For readers the shape is familiar: a capable assistant that is free if you already pay for the platform, and a standalone monthly fee if you do not.

Meta has released a WhatsApp Business Tools MCP server that lets an AI agent handle the setup work most businesses find tedious [1][2]. Instead of clicking through several consoles, an operator describes what they want and the agent runs the steps: checking Terms of Service status, creating a WhatsApp Business account, adding and verifying phone numbers, registering numbers for the Cloud API, creating and editing message templates, and sending test messages and webhooks to confirm the integration works [1][2].

Before this, the same job meant moving by hand between Meta's Developer Console, Business Manager, the API reference and a code editor [1][2]. The MCP server collapses that into a chat: the agent discovers the tools Meta exposes and calls them on the user's behalf [1][2]. It can also flag problems with a business's Terms of Service, payment methods or verification status along the way [2].

The server works with several agents, including Claude, Cursor, Codex and ChatGPT [1][2]. Meta also points to a companion Social Technologies MCP server that an agent can use to discover API endpoints, search documentation and troubleshoot errors in the same session [1].

The launch lands a day before Google's Google Home MCP server, and both use the same protocol — a sign that MCP is settling in as the standard way to let agents act inside a vendor's product rather than only answer questions about it [1]. Meta did not publish figures on how many businesses have used the tool [1].