Bessent has reportedly asked Nvidia to explain within 10 days whether it facilitated “seemingly illegal” chip sales to China; two institutional investors with combined holdings of about $2.13 billion have urged the company not to rush a leadership change. Researchers examined roughly 159,000 Lovable projects and found that more than 200,000 private messages on one forum could be read without logging in; the analysis was limited to apps using Supabase. Gemini is beginning to offer all app users in the US a connection to Google Wallet to analyze spending and subscription price increases, but it can only read account information, not take action with funds.
According to Bloomberg, Bessent urged Nvidia to hire an outside CEO and asked it to explain within 10 days whether it facilitated “seemingly illegal” chip sales to China to boost revenue. Separately, CNBC reported that two institutional investors with combined holdings of about $2.13 billion wrote to the board, calling for a careful CEO selection process that leaves open the possibility of Jensen Huang’s return. They did not demand his immediate reinstatement, instead arguing against a hasty leadership change before the facts are established.
Read original →🔍 Analysis: Bessent is pushing for an outside CEO, while two institutional investors with combined holdings of about $2.13 billion want to leave open the possibility of Jensen Huang’s return. The disagreement centers on whether to pursue an irreversible leadership decision before the facts are established. For Nvidia, its explanation of sales within the 10-day deadline will address not only compliance allegations but also provide an important basis for the board’s handling of succession.
Once sensitive data is connected, what can be read—and by whom—determines risk before what an app can generate does.
Researchers examined roughly 159,000 Lovable projects and found that more than 200,000 private messages on a German forum could be read without logging in. A service provider also exposed more than 1,000 records containing passport and other identity document information. Lovable advertises website building without technical knowledge, but its terms place responsibility for apps and data processing on customers. The analysis focused on apps using the Supabase database; the sample does not represent all apps on the platform.
Read original →Google announced on November 13 that Gemini was beginning to offer all app users in the US a connection to Google Wallet. Google says the connection can generate spending charts, identify subscription price increases, and create holiday budgets based on actual transactions from the previous year. Users can also link bank and credit card accounts through Plaid. The connection requires users to opt in and can be disabled at any time; Gemini can only read account information and cannot transfer, manage, or invest funds.
Read original →🔍 Analysis: Gemini’s Google Wallet connection is opt-in, can be disabled, and is read-only. In the Lovable case, however, the risk to more than 200,000 private messages arose precisely because they could be read without logging in. The contrast shows that read-only access can limit consequences such as transactions involving funds, but it is no substitute for access authentication. For apps handling financial or identity data, the ability to read information is itself a core capability that needs to be controlled.
Beyond funding, revenue, orders, and business metrics offer different levels of commercial validation.
According to FinTech Global, market data API company Databento completed a $100 million Series C at a valuation of $1.1 billion. The company says it has signed enterprise agreements with five of the six largest frontier AI labs, providing real-time and historical market data through a unified API. Just months after its $97 million Series B, the company has more than doubled revenue while remaining profitable. The new funding will support data infrastructure and broader asset-class coverage.
Read original →According to Upstarts Media, Nidus raised a combined $30 million across pre-seed and seed rounds to develop AI for manufacturing robots serving US defense industry customers. The company is starting with wire harness manufacturing—assembling cables, wires, and connectors into units that are easy to install—with the aim of increasing capacity without rebuilding existing processes. Nidus says it secured more than $1 million in preorders within four months. The funding will support hiring, additional data collection, and model training.
Read original →According to Tech.eu, Nebuly completed a $21 million Series A for technology that helps enterprises measure the business value of AI agents. By analyzing agents’ effects on productivity and revenue, it grounds evaluation in concrete business metrics. The technology focuses on the returns enterprises get after deploying agents, rather than the tasks the agents themselves can perform.
Read original →🔍 Analysis: Databento raised $100 million while remaining profitable and more than doubling revenue, whereas Nidus’s $30 million in funding comes alongside more than $1 million in preorders. Their commercial validation cannot be compared directly on the basis of funding totals. By evaluating agents in terms of productivity and revenue, Nebuly addresses the missing link between technical deployment and measuring business returns. For enterprise buyers, data agreements, preorders, and actual business gains should be measured separately rather than treated as interchangeable.
Embedding AI in development means putting exploration, constraints, and acceptance checks in the right places.
Meta VP of Product Mona Sarantakos discussed Muse’s development and launch on Lenny’s Podcast. She said that although Muse got started earlier than most competitors, it waited seven months before launching. The team also used “agent swarms,” with multiple AI agents collaborating to uncover user insights at scale.
Read original →Y Combinator released a Lightcone interview in which Lauren Tan and Roman Ugarte discussed building and deploying Grok Bot. Topics included a workflow handling 4,000 PRs (pull requests) per month, using automated tests and repository guardrails to help agents check their own work, and gradually expanding the bot’s permissions.
Read original →In an interview with Peter Yang, Anthropic designer Nate Parrott shared five ways to make Claude’s designs feel less “AI-generated.” He recommends first providing references such as screenshots, studying interaction patterns familiar to users, using files to constrain style preferences, then validating flows with wireframes, and finally generating multiple versions and mixing elements from them. He argues that models do not recognize their own biases, so users can explicitly ask them to avoid familiar tropes such as all-serif typography and off-white backgrounds.
Read original →🔍 Analysis: Meta uses “agent swarms” to scale user insights; Grok Bot relies on tests and repository guardrails to support a workflow of 4,000 PRs a month; and Nate Parrott puts reference materials, style files, and wireframes before design generation. What they share is an emphasis on organizing models through workflows rather than expecting a finished product from a single prompt. Muse’s seven-month wait before launch also shows that starting earlier does not automatically mean delivering earlier. The reusable practice for teams is to delegate exploration to agents while embedding constraints and acceptance checks into the process, rather than judging efficiency solely by output volume.