The last day in AI pushed agents into advertising, productivity software, homes and scientific research. OpenAI began testing business-sponsored conversations inside ChatGPT, while Anthropic removed the boundary between chat and autonomous work. Google opened smart-home controls to outside agents, and Stanford researchers turned papers into callable tools. A transatlantic merger and new polling also showed how competition, sovereignty and public concern are shaping the market around those products.
OpenAI turns ads into sponsored conversations
OpenAI introduced new tools for ChatGPT Ads, led by a US test of Sponsored Agents. After clicking an ad, a person can start a clearly labeled conversation with an agent representing the advertiser. OpenAI says the sponsored exchange remains separate from ChatGPT's independent answers and from the conversation that produced the ad.
The company also added natural-language campaign management, creative suggestions, optional text customization and integrations with HubSpot and Shopify. That moves advertising beyond placement and targeting toward a guided sales interaction. The central trust issue is whether users can consistently distinguish a paid agent's objectives from the assistant they originally asked for help.
Anthropic folds Cowork into one Claude interface
TechCrunch reported that Anthropic is merging Claude chat and Cowork into one interface. Claude will route requests across chat, autonomous work, Artifacts and Claude Design without asking users to select a separate mode. The company is also adding collaborative documents and presentations that can be shared, edited and exported.
The redesign makes agentic behavior less of a specialist feature and more of a default product capability. That can reduce friction, but it also raises the importance of showing when Claude is merely drafting, when it is using files or tools, and when work continues in the background. Clear progress, permissions and stopping controls become part of the interface, not only the safety layer.
Google Home opens physical context to outside agents
Google announced early access to Home MCP, a Model Context Protocol server that lets compatible agents interact with devices and event history in the Google Home ecosystem. Supported uses include summarizing activity across cameras, analyzing device history, controlling equipment and creating custom dashboards. Access starts with Google Home Premium Advanced users in US English.
Home MCP adds useful real-world context, but it also expands the consequences of a mistaken action or compromised agent. Google says the service applies rate limits and blocks sensitive operations such as unlocking doors. Those protections are a baseline. Users will also need understandable scopes, revocation tools and logs that show which agent accessed which household data.
Cohere and Aleph Alpha formalize a sovereign AI company
Cohere and Aleph Alpha signed a definitive business combination agreement following their planned partnership announced in April. The combined company will operate as Cohere, with headquarters in Toronto and Berlin, more than 1,000 employees, and Aleph Alpha's Heidelberg office retained as a research center. Closing remains subject to regulatory approval.
The strategy is aimed at governments and regulated enterprises that want capable models with local control over data, infrastructure and compliance. Combining Canadian and German operations gives the company a larger research base and public-sector footprint. It must still show that sovereignty translates into competitive products, reliable deployment and governance that works across two legal systems.
Stanford turns scientific papers into callable agents
Stanford Medicine researchers introduced Paper2Agent, a system that converts a manuscript, its code and associated data into an interactive agent. Worker agents attempt to reproduce the research, capture the method in a Model Context Protocol server and let other agents apply it to new data. The team has created more than 100 paper agents.
In one demonstration, agents based on separate genomics papers combined their methods and identified a potential ADHD-related molecular variant that the researchers say had not been reported. The result is an early proof, not a replacement for replication or peer review. Attribution, human context and controlled collaboration will be essential if scientific knowledge becomes executable rather than merely readable.
US concern about AI jobs shifts across party lines
Pew Research Center found that Democrats are now more concerned about AI's role in daily life than Republicans, at 56% versus 49%. In the June survey of 3,488 US adults, 75% of Democrats and 68% of Republicans said AI will lead to fewer jobs over the next 20 years.
The partisan shift matters because AI policy is becoming an election issue tied to data centers, labor and regulation. Concern is rising even as consumer and workplace products gain more agency. Companies will need evidence about job redesign, oversight and distribution of benefits, not only demonstrations of what new systems can do.
What matters today
Agents are moving from optional tools into the default structure of major products. OpenAI wants paid conversations to continue after an ad, Anthropic is routing ordinary requests into agentic work, Google is connecting assistants to homes, and Stanford is making research methods callable. Cohere's expansion and Pew's polling show the pressure around that shift. Distribution is accelerating, but durable adoption will depend on transparent incentives, narrow permissions, strong audit trails and proof that people retain meaningful control.