Latest / UpNext AI / Private Social AI, Banking Software Bets, and Retail Humanoids | UpNext AI – July 23, 2026
Transcript
- Welcome to the UpNext AI podcast. It's Thursday, July 23rd, 2026, and here's what matters in AI today. First up, Yope has raised $12.3 million in seed funding, according to TechCrunch, for a very specific alternative to the dominant social model. The pitch is simple: no algorithmic feed, no ads, and no public-content treadmill. Instead, Yope is built around small private groups of friends and family, with accounts private by default. The app is betting that people still want social products, just not necessarily the version optimized for creators, recommendation loops, and public performance. TechCrunch reports that Yope says it has nearly 15 million registered users, and that users are sharing between 10 million and 20 million pieces of content weekly. The company says more than 50 percent of users open the app at least five days a week, and around 20 percent of active users have invited an older family member onto the platform. What makes this relevant for the AI beat is how the company describes the role of AI. Yope’s founder told TechCrunch the company does not see AI as a tool for generating content. Instead, it wants AI to support interaction: things like recap features, mini-games for groups, and tools that could help friends meet up in real life. That’s a notable framing shift. A lot of consumer AI has been aimed at making more stuff. This is aimed at making social products feel smaller, more private, and more useful between people who already know each other. The funding round was led by Northzone, and it brings Yope’s total funding to $20 million. The company says it will use the money to keep building the product, grow its team, and open a U.S. office. So the bigger signal here is not just one startup raise. It’s that at least some founders and investors think the next AI-native consumer hit may come from reducing the feed, not maximizing it. Next, ServiceNow is making a more enterprise-flavored AI bet. TechCrunch reports the company is investing $40 million in BusinessNext, an Indian banking software specialist, at a $700 million valuation. That gives ServiceNow roughly a 5 percent stake and deepens a partnership aimed at selling more AI-powered software into financial services. BusinessNext is based in Noida, is 24 years old, and generated about $32 million in revenue in its latest financial year, according to the report. TechCrunch also says the company serves more than 70 banks across India, Southeast Asia, the Middle East, and the U.S., and that about half of its revenue already comes from outside India. The strategic logic is pretty clear. BusinessNext focuses on customer-facing banking workflows, while ServiceNow is strong in workflow automation and back-office systems. Together, they plan to sell that combination to financial institutions. BusinessNext’s founder described the arrangement to TechCrunch as a strategic partnership cemented with funding, and said ServiceNow’s global sales reach should help the company expand in markets where it has limited presence today. The other important detail is architectural. TechCrunch reports that BusinessNext has been building what it calls an autonomous banking platform, using AI agents to automate banking workflows while keeping sensitive customer data on private AI infrastructure for regulatory and privacy reasons. That matters because it reflects where a lot of practical enterprise AI work is heading now: not generic copilots, but domain-specific systems tied to regulated workflows, existing software stacks, and strict data boundaries. So while this is a single investment, it’s also a useful snapshot of the next phase of enterprise AI adoption: vertical, compliance-aware, and sold through distribution partnerships rather than hype alone. Now to the research section. A paper posted earlier this week on arXiv is called "Closing the Lab-to-Store Gap: A Data-Efficient Post-Training and Experience-Driven Learning VLA Framework for Retail Humanoids." The core problem is easy to understand. A robot can look impressive on benchmarks or in a tightly controlled demo, but still struggle in a real store where shelves vary, lighting changes, mistakes happen, and the environment is messy. The paper focuses on vision-language-action, or VLA, humanoid robots, meaning systems that combine visual input, language-like instruction handling, and physical action. The researchers present a framework they call DEED, short for Data-Efficient Post-Training and Experience-Driven Learning. They evaluated it on a supermarket chip-restocking task using a Unitree G1-Edu humanoid robot and the GR00T N1.6 foundation model. According to the abstract, the framework has three main parts: a data-efficient post-training pipeline, a real-world experience-driven refinement loop, and a tool for analyzing how the system behaves when conditions shift away from what it saw in training. The most interesting claim is not that they found a magical new robot architecture. It’s almost the opposite. The paper says the lab-to-store gap is primarily a systems-integration challenge rather than an architectural one. In other words, careful data design, targeted post-training, and learning from real deployment experience may matter more than just swapping in a flashier base model. The abstract also says that this approach turned a policy that failed under naive fine-tuning into a competent real-world system using only a single GPU. We do not have quantitative results in the supplied material, so that claim still needs the full paper for context. But even at a high level, the direction is clear. Bottom line: for humanoid robots, real progress may come less from bigger demos and more from the unglamorous work of post-training, error recovery, and learning on the job. ...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. ... Cisco says its small open-source cybersecurity models can detect about 150 times more vulnerabilities per dollar than larger AI agents, according to The Decoder. The notable part is not just the performance claim, but the cost framing: more security work from smaller specialized models instead of always reaching for the biggest general-purpose system. And the Financial Times reports Google burned through $6 billion in cash as AI spending climbed again, while saying it will commit up to $205 billion to AI investments in 2026. With the details we have here, the safe takeaway is simply that the hyperscale AI buildout is still a hard-dollar infrastructure race. 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 tomorrow with what's up next!