Latest / UpNext AI / ChatGPT Ads Go Global, OpenAI’s Revenue Push, and Long-Horizon AI Agents | UpNext AI – August 14, 2026
Transcript
- Welcome to the UpNext AI podcast. It's Friday, August 14th, 2026, and here's what matters in AI today. OpenAI’s experiment with advertising in ChatGPT is no longer confined to an initial U.S. test. The company says ChatGPT Ads have now launched in the United Kingdom, Mexico, Brazil, Japan, and South Korea, with further expansion planned this year. OpenAI frames the program as a way to fund broader access to ChatGPT, especially its Free and Go tiers, while keeping higher subscription tiers ad-free. The company says ads are clearly labeled and visually separated from answers, and that advertising does not influence what ChatGPT tells users. There is an important privacy distinction in the design. OpenAI says advertisers do not receive users’ chats, chat histories, memories, or personal details; they receive aggregate performance information such as views and clicks. But during the test, ad selection can be informed by the current conversation, past chats, and past ad interactions. Users can dismiss ads, manage personalization, and delete ad data. OpenAI also says it will not show ads to accounts where it believes the user is under 18, or alongside sensitive and regulated subjects including health, mental health, and politics. The strategic shift is clear: conversational AI is becoming both a subscription product and an advertising surface. The real test will be whether OpenAI can preserve trust as it scales relevance and revenue at the same time. That commercial pressure also helps explain OpenAI’s next move. The company has appointed Dali Rajic as chief revenue officer to lead its global revenue organization. OpenAI says its products now reach more than one billion weekly active users and more than two million businesses, double the business count from a year ago. Rajic arrives as the company tries to turn that adoption into a repeatable global go-to-market operation. He most recently served as president and chief operating officer at Wiz, the cybersecurity company recently acquired by Google. Before that, he held senior operating and revenue roles at Zscaler and AppDynamics. OpenAI says Rajic’s remit is to build the revenue operating system for its next phase, selling to both large enterprises and technical customers. Denise Dresser, who helped establish OpenAI’s commercial foundation, will depart after a transition period. The significance is less about an executive title than the timing: OpenAI is putting more structure around enterprise deployment just as it argues AI will become embedded in everyday workflows. For customers, that could mean a more disciplined commercial organization; for OpenAI, it is an effort to make a rapidly growing business more scalable. For the research note, a new paper argues that judging autonomous agents by their final result misses the most useful question: how did they get there, and do they improve from experience? The researchers evaluated seven frontier models across 36 long-horizon tasks involving improvements to models, systems, and other technical artifacts. Rather than relying only on end scores, their framework tracked three parts of an agent’s work: how it frames a solution, how it executes, and how it uses feedback. They also tested whether experience from one run helps on later decisions, within and across tasks. Their finding is a useful reality check. Today’s agents look more like engineering optimizers than fully autonomous researchers. They can formulate and implement practical solutions, but performance varies substantially between runs. Their strongest outputs mainly adapt or combine established techniques, while genuinely new methods remain rare. The study also finds that similar final outcomes can hide very different process bottlenecks, and that accumulated experience can either help or mislead later choices. The caveat is that this is one evaluation across a defined set of tasks and harnesses. Still, the takeaway for teams building research agents is clear: measure the process, not just the final score. ...Are you building apps with voice? Elevate your app's voice capabilities with ElevenLabs. Their API is a game changer for embedding dynamic, responsive voice interactions in your applications, providing unprecedented realism, flexibility and latency. In fact, you're listening to one of their voices - right - now. If you are a developer looking to elevate user experience with natural voice interfaces, this is your solution. Visit up next dot fm slash eleven to check out their latest offerings. ... Developer Simon Willison released version 0.33 of his llm-gemini plug-in, adding support for Gemini 3.7 Flash, earlier Flash models, and two Gemini embedding models. The update also adds access to reasoning traces and server-side tools, giving developers a simpler path to test those capabilities from the command line. Apple has reportedly trained a custom AI model for China with Alibaba’s support. Reuters, as cited by The Verge, says the China-focused model would give Apple more control over its products in the country’s competitive smartphone market, while marking an unusual cross-border partnership amid tensions between Beijing and Washington. And TechCrunch reports that OpenAI is previewing an Ultrafast mode for GPT-5.6 Sol, which it says runs at 14 times the speed. OpenAI is positioning the faster version toward enterprise users, though the report does not specify the speed metric or any capability tradeoffs. Before we wrap up, a quick note: this podcast is generated with the assistance of AI and is intended for informational purposes only. All referenced articles, research, and commentary remain the property of their original authors and publishers. If you enjoyed this episode, don't forget to subscribe, rate, and leave us a review! And that's your briefing for today. Full source links are in the episode notes, and we'll be back Monday with what's up next!