South Korea, a nation celebrated globally for its lightning-fast internet, pervasive broadband, and its standing as a digital innovation powerhouse nurturing international tech giants like Hyundai, LG, and Samsung, paradoxically finds its very success making it a prime target for cyberattacks. This vulnerability has starkly revealed the inherent fragility of its cybersecurity defenses. The country is currently grappling with the aftermath of numerous high-profile cyber incidents, which have impacted credit card companies, telecommunications providers, burgeoning tech startups, and government entities, thereby affecting a significant portion of its populace. In the wake of each attack, ministries and regulatory bodies often appear to react in an uncoordinated manner, frequently deferring to one another rather than presenting a unified front. Brian Pak, CEO of the Seoul-based cybersecurity firm Theori and an advisor to SK Telecom’s parent company’s special committee on cybersecurity innovations, observed to TechCrunch that the government’s approach to cybersecurity remains largely reactive. He argues it is treated as a crisis management issue rather than being recognized as critical national infrastructure. Pak further elaborated that the siloed operations of government agencies responsible for cybersecurity often lead to the neglect of developing robust digital defenses and training a skilled workforce. Indeed, South Korea is facing a critical deficit of skilled cybersecurity professionals. According to Pak, this is predominantly due to an approach that has stifled workforce development. He explains, "This lack of talent creates a vicious cycle. Without enough expertise, it’s impossible to build and maintain the proactive defenses needed to stay ahead of threats." Political gridlock, Pak notes, has fostered a tendency towards seeking swift, superficial "quick fixes" following each crisis, while the more demanding, long-term endeavor of building genuine digital resilience continues to be relegated to the sidelines. This year alone, South Korea has experienced a major cybersecurity breach almost monthly, intensifying anxieties over the robustness of its digital infrastructure. However, the existing fragmented system in South Korea weakens accountability. Pak cautions that consolidating all authority into a presidential "control tower" could introduce risks of "politicization" and potential overreach. In response to inquiries, a spokesperson for South Korea’s Ministry of Science and ICT affirmed that the ministry, in collaboration with KISA and other pertinent agencies, is "committed to addressing increasingly sophisticated and advanced cyber threats." The spokesperson added, "We continue to work diligently to minimize potential harm to Korean businesses and the general public."

Drivers for ride-hailing platforms such as Uber and Lyft in California are poised to gain the right to unionize while retaining their classification as independent contractors. This significant development comes after Governor Gavin Newsom signed a new bill into law on Friday.

Fresh data from PitchBook underscores the profound extent to which artificial intelligence is seizing control of startup investment. Current projections suggest that by 2025, AI-focused ventures are set to attract over half of all venture capital deployed, a historic first. Kyle Sanford, PitchBook’s director of research, described this evolving landscape to Bloomberg as increasingly “bifurcated.” He highlighted a stark emerging dichotomy within the market, asserting that “you’re in AI, or you’re not,” and equally, “you’re a big firm, or you’re not.”

The relationship between U.S. startups and the federal government has undergone a significant transformation in recent years, becoming both closer and more complex. This burgeoning alliance is largely driven by a governmental push to leverage innovations in artificial intelligence, automation, space technology, robotics, and climate tech for defense purposes. While this has opened up new avenues for capital acquisition for emerging companies, it has simultaneously introduced new layers of entanglement. An increasing number of startups now count the U.S. government among their primary customers, or are actively pursuing essential permits and lucrative defense-related contracts. When federal operations run smoothly, these connections can provide invaluable momentum and revenue streams. However, as demonstrated by the government shutdown initiated on October 1st, such deep reliance can quickly turn into a liability, potentially stifling or even halting progress for these closely aligned ventures. As observed during the Equity podcast, this evolution reflects a broader shift in the startup ecosystem over the past decade. Previously dominated by consumer internet ventures, the landscape now features a strong emphasis on defense tech and deep tech, sectors inherently requiring various forms of regulatory approval and direct government engagement. Consequently, a much wider spectrum of the startup world now finds itself dependent on governmental functions in ways that were far less prevalent a decade ago. The government's expanding influence is not limited to its role as a customer or regulator. The Trump Administration, in particular, has demonstrated an increasing tendency to extend its direct reach and even ownership into the tech industry. This trend was recently underscored by the renegotiation of a third federal loan in recent months, following similar arrangements with Intel and the rare earth miner MP Materials. As part of these newly structured deals, the administration has often secured an equity stake, further solidifying its direct involvement in key technological enterprises.

Earlier this year, Anker, the Chinese company behind Eufy security cameras, launched a program that compensated users for submitting videos of package and car thefts. The firm explicitly sought footage of both genuine and staged incidents, asserting on its website that such data was crucial "to help train the Al what to be on the lookout for." Notably, Eufy's instructions explicitly encouraged users to actively create these scenarios. "You can even create events by pretending to be a thief and donate those events," the company stated, suggesting efficiency by capturing "one act" with multiple outdoor cameras simultaneously. It even quantified potential earnings, noting that staging a car door theft "might earn $80." Eufy underscored that "the data collected from these staged events is used solely for training our Al algorithms and not for any other purposes," attempting to assuage potential concerns. This initiative sheds light on a broader trend: companies' increasing willingness to compensate users for data considered valuable in training AI models. While this offers some individuals a means to monetize their own data, it simultaneously raises significant security and privacy implications that warrant careful consideration. The initial campaign, offering $2 per theft video, ran from December 18, 2024, to February 25, 2025. According to user comments on the campaign's announcement page, over 120 individuals reported participating. However, Eufy remained unresponsive to TechCrunch's inquiries regarding the total number of participants, the overall sum disbursed, the volume of videos collected, or whether the company subsequently deleted the footage after AI training. Since then, Eufy has launched similar campaigns, continually incentivizing its customers to submit videos for the enhancement of its AI systems. Among these is an ongoing in-app "Video Donation Program," aimed at further improving Eufy's AI. This program offers a range of rewards, from an "Apprentice Medal" – essentially a badge displayed next to the user's name – to more tangible gifts like cameras or gift cards. For this specific campaign, Eufy focuses on videos involving human subjects. The Eufy app further promotes participation through an "Honor Wall," publicly ranking users based on the number of video events they have donated. The current leading contributor, according to the app, has provided a staggering 201,531 videos. Within the donation program's app page, Eufy reiterates its commitment to privacy, clarifying that "donated videos are only used for Al training and improvement. Eufy will not provide the video to third parties." Despite these clarifications, Eufy once again did not respond when asked about this particular initiative. Ultimately, after a back-and-forth with the tech news site, Anker admitted to having misled users and committed to rectifying the problem.

The pressing issue of water scarcity resonates deeply with Mario Bustamante, based in Chile, who shared with TechCrunch that it constitutes a major challenge in the region. This critical need for solutions has driven companies like Instacrops to innovate. The firm's strategic pivot from a hardware-centric approach to one powered by artificial intelligence profoundly transformed its operations, enabling it to manage vast amounts of data with a leaner team. Bustamante elaborated on this shift, highlighting the dramatic increase in data processing capabilities: "We are processing — more or less — 15 million data points per hour. Almost 10 years ago, that was the amount for a year." This exponential growth translates directly into tangible benefits, as he noted, "We’re reducing cost, team members, and generating more impact with less." At its core, Instacrops' technology empowers farmers by leveraging data from IoT sensors, whether newly installed or integrated into existing farm networks. This data is then fed into the startup's advanced LLM models, which process over 80 distinct parameters. These inputs include crucial metrics such as soil moisture, humidity, temperature, pressure, crop yield, and NDVI – a key plant productivity indicator derived from satellite imagery. The insights generated from this comprehensive analysis enable precise recommendations on when and where to irrigate specific areas. These critical advisories reach farmers directly on their mobile phones. While Instacrops provides its own chatbot application, the company also prioritizes integration with WhatsApp. Bustamante anticipates a full transition to the messaging platform, stating, "I think in the next year, we will be 100% WhatsApp because it’s a universal tool for any farmer." For farms with a higher degree of technological sophistication, Instacrops extends its capabilities to directly control irrigation systems, providing an even more integrated solution. Currently, Instacrops strategically targets high-value crops across Latin America, focusing on produce such as apples, avocados, blueberries, almonds, and cherries. Farmers access these vital irrigation insights by paying an annual fee calculated per hectare of their farmland.

OpenAI semble s'orienter vers une révision de sa gestion des droits d'auteur et de la propriété intellectuelle au sein de sa nouvelle application vidéo, Sora. Au cœur de cette approche potentielle se trouve le concept de "opt-in", suggérant qu'OpenAI empêcherait la création de vidéos avec des personnages protégés par le droit d'auteur à moins que les studios et autres ayants droit n'aient expressément autorisé Sora à le faire. Sam Altman, PDG d'OpenAI, a rapporté que de nombreux ayants droit se montrent très enthousiastes face à cette nouvelle forme de "fan fiction interactive", y voyant une source potentielle de grande valeur grâce à l'engagement généré. Il a cependant souligné leur volonté de spécifier les modalités d'utilisation de leurs personnages, y compris la possibilité de l'interdire totalement. Même avec cette nouvelle stratégie, Altman a reconnu la probabilité de rencontrer des "cas limites de générations qui passent alors qu'elles ne devraient pas". L'espoir de l'entreprise est que ce nouveau type d'engagement soit "encore plus précieux que le partage des revenus", tout en souhaitant que les deux aspects soient significatifs.

Partiful, l'application de planification d'événements sociaux qui se décrit comme le "Facebook events pour les gens branchés," a fermement pris la place de Facebook comme plateforme de référence pour l'envoi d'invitations à des fêtes. Cependant, Partiful partage également avec Facebook une propension à collecter une quantité massive de données utilisateur, et l'entreprise aurait pu faire mieux pour assurer la sécurité de ces informations. À présent, Partiful s'est transformé en un puissant graphe social, similaire à celui de Facebook, capable de cartographier aisément l'identité de vos amis et des amis de vos amis, vos activités, vos déplacements et l'ensemble de vos numéros de téléphone. Suite à des spéculations autour de l'application, TechCrunch a créé un nouveau compte pour la tester. Nous avons rapidement découvert que l'application ne supprimait pas les données de localisation des images téléchargées par les utilisateurs, y compris les photos de profil publiques. TechCrunch a constaté qu'il était possible pour quiconque, en utilisant uniquement les outils de développement d'un navigateur web, d'accéder aux photos de profil brutes des utilisateurs stockées dans la base de données backend de Partiful hébergée sur Google Firebase. Si la photo d'un utilisateur contenait la localisation géographique précise de l'endroit où elle avait été prise, n'importe qui aurait également pu visualiser les coordonnées exactes de ce lieu. Cette faille de sécurité est problématique car tout utilisateur de Partiful aurait pu révéler la localisation de l'endroit où la photo de profil d'une personne a été prise. Certaines photos de profil d'utilisateurs de Partiful contenaient des données de localisation très granulaires qui auraient pu être utilisées pour identifier le domicile ou le lieu de travail de la personne, en particulier dans les zones rurales où les maisons individuelles sont plus facilement distinguables sur une carte. Il est courant que les entreprises qui hébergent des images et des vidéos d'utilisateurs suppriment automatiquement les métadonnées lors du téléchargement afin de prévenir des atteintes à la vie privée de ce type. TechCrunch a vérifié le bug en téléchargeant une nouvelle photo de profil Partiful que nous avions préalablement capturée devant le Moscone West Convention Center à San Francisco, et qui contenait la localisation précise de la photo. Lorsque nous avons vérifié les métadonnées de la photo stockée sur le serveur de Partiful, elles contenaient toujours les coordonnées exactes de l'endroit où l'image avait été prise, à quelques pieds près. Après avoir découvert cette faille de sécurité, TechCrunch a alerté les co-fondatrices de Partiful, Shreya Murthy et Joy Tao, par email, Partiful ne disposant pas de moyen public pour signaler les failles de sécurité. TechCrunch a partagé un lien vers une photo de profil brute d'un utilisateur de Partiful contenant la localisation réelle de cet utilisateur au moment où la photo avait été prise, une adresse résidentielle à Manhattan. Tao a déclaré à TechCrunch vendredi que la vulnérabilité était "déjà sur le radar de notre équipe, et a récemment été priorisée comme une correction à venir." Partiful a initialement fourni un calendrier pour corriger la faille d'ici "la semaine prochaine," mais étant donné la sensibilité des données impliquées, TechCrunch a demandé une correction pour vendredi. Partiful a confirmé avoir corrigé le bug samedi. TechCrunch a constaté samedi que les métadonnées avaient été supprimées des photos existantes téléchargées par les utilisateurs. La photo de profil que nous avions téléchargée avec notre localisation réelle avait également vu ses métadonnées supprimées. Lorsqu'on lui a demandé si Partiful disposait des moyens techniques, tels que des journaux, pour déterminer s'il y avait eu un accès direct ou en masse aux photos de profil des utilisateurs stockées dans sa base de données, la porte-parole de Partiful, Jess Eames, a répondu que c'était "toujours sous enquête mais nous n'avons trouvé aucune preuve de cela pour l'instant." Eames a déclaré que l'entreprise "effectue régulièrement des revues de sécurité avec des experts dans le domaine, non pas comme une action ponctuelle mais comme faisant partie de nos processus continus." Partiful n'a pas fourni à TechCrunch les noms des experts lorsqu'on les a interrogés. Partiful a levé plus de 27 millions de dollars auprès d'investisseurs depuis sa fondation en 2022, y compris un tour de financement de série A de 20 millions de dollars mené par Andreessen Horowitz. TechCrunch a demandé aux co-fondatrices de Partiful si elles avaient commandité une revue de sécurité de leur produit avant le lancement, mais elles n'ont pas souhaité répondre.

The California Department of Insurance (CDI) has accused Tesla’s insurance arm, along with its partner State National Insurance Company, of engaging in “willful unfair claims settlement practices.” These include “egregious delays in responding to policyholder claims in all steps” of the process and “unreasonable denials,” actions which the CDI states have allegedly caused “financial harm” and “distress to policyholders.” The regulator asserts that despite initial interventions in 2022, the situation has only deteriorated. Indeed, the CDI noted that, according to filings, “In 2025, the Tesla Companies have already had more complaints, more justified complaints, and committed more violations than in the three previous years combined.” CDI first engaged with Tesla in December 2022, following a “marked uptick in claims-related consumer complaints” against the company. During this period, the regulator discovered that Tesla’s “Head of Claims” position had been vacant for months and accused the companies of failing to report these critical claims-handling problems. As a consequence, CDI subjected both Tesla and State National to a six-month probationary period to monitor their efforts to reduce violations. Tesla and State National “conceded” they had underestimated the volume of claims and the staffing required to handle them, promising to increase hiring. While Tesla eventually hired a new Head of Claims in April 2023, and both companies “reported improvements in the quality” of their claims handling and the “resolution of consumer complaints” through the remainder of that year, this respite was short-lived. By 2024, the CDI observed a “significant increase” in both consumer complaints against Tesla and “violations of the law.” The number of consumer complaints surged from a mere 83 in 2022 to 829 in 2024, with the CDI finding Tesla in violation of the state’s insurance code in 775 of those cases. The problems have intensified further this year. Through September 22, the regulator has received 1,481 complaints against Tesla and identified 1,969 insurance code violations. Cumulatively, since 2022, Tesla has accumulated nearly 3,000 violations of state insurance law. The vast majority of these involve Tesla’s failure to respond to customers within the mandatory 15-day period, alongside 166 identified violations where Tesla failed to conduct a “thorough, fair, and objective investigation” into a claim. “CDI repeatedly notified [Tesla] of its claims-mishandling issues and violations of law,” the regulator stated, adding, “While [Tesla] repeatedly committed to improvements, the number of justified complaints and violations continued to mount, demonstrating [Telsa’s] failure to correct its practices.” Consequently, Tesla and State National now face potential penalties of up to $5,000 for each “unlawful, unfair, or deceptive act” and up to $10,000 for each “willful” act. The companies have 15 days to respond to the allegations.

Jeff Bezos, le visionnaire derrière Blue Origin, a récemment partagé son ambitieux aperçu de l'avenir extraterrestre de l'humanité lors d'une discussion avec John Elkann, figure éminente de la lignée Agnelli d'Italie. Bezos a postulé que les gens habiteraient l'espace principalement par choix, déléguant le travail exigeant aux robots tandis que de vastes centres de données d'IA flotteraient en orbite. Cette déclaration audacieuse trace inévitablement des parallèles avec son homologue de l'industrie spatiale, Elon Musk, qui a constamment défendu la colonisation de Mars, projetant un million de résidents d'ici 2050. La question se pose alors : ces deux milliardaires sont-ils déconnectés des réalités actuelles, ou sont-ils au courant de connaissances qui échappent à l'individu lambda ? L'optimisme de Bezos s'est étendu au-delà de l'espace, car il a également défendu avec force le boom actuel des investissements dans l'IA. Il l'a décrit comme une bulle "industrielle" bénéfique, la distinguant des vagues "financières" plus précaires. Sa conclusion rapportée, "Il n'y a jamais eu de meilleur moment pour être enthousiasmé par l'avenir", a probablement suscité, on l'imagine, un certain degré de scepticisme tacite parmi les participants réunis dans l'auditorium de Turin.

OpenAI has officially acquired Roi, an AI-powered personal finance application, with its CEO being the sole individual transitioning to the acquiring company, a trend increasingly observed across the AI industry. While specific details regarding the integration of Roi's technology into OpenAI's ecosystem or the precise unit Vishwajith will join remain undisclosed, the acquisition strongly indicates OpenAI's strategic focus on personalization and comprehensive life management as key frontiers for future AI products. Roi notably brings a team experienced in tackling the complexities of financial personalization at scale, offering insights that could be broadly applied to various AI challenges. Founded in New York in 2022, Roi successfully raised $3.6 million in early-stage funding, attracting investors such as Balaji Srinivasan, Spark Capital, Gradient Ventures, and Spacecadet Ventures, according to PitchBook data. The platform's core mission was to consolidate users' entire financial footprint – including stocks, cryptocurrency, DeFi assets, real estate holdings, and NFTs – into a single application capable of tracking funds, generating insights, and facilitating trades. Beyond its comprehensive tracking and trading capabilities, Roi distinguished itself by offering users access to a sophisticated AI companion, engineered to provide financially astute responses tailored to individual preferences. Upon initial setup, users could actively personalize Roi's behavior, supplying information ranging from their professional background to their desired interaction style. This design philosophy highlights a core belief held by Roi and its co-founder: that software should transcend generic outputs, instead adapting, learning, and communicating in ways that feel uniquely personal, human, and crucially, maintain user engagement. As the Roi team previously articulated in a blog post, they envision everyday products evolving from static, predetermined experiences into "adaptive, deeply personal companions that understand us, learn from us, and evolve with us."

Naveen Rao, formerly head of artificial intelligence at Databricks, is reportedly in discussions to secure $1 billion in funding for his nascent startup, Unconventional, Inc. The company, focused on developing an innovative computing architecture, is aiming for a $5 billion valuation, according to four sources familiar with the ongoing talks. Rao has already amassed hundreds of millions of dollars and intends to initiate the startup's operations without waiting for the full $1 billion round to conclude. Instead, he plans to raise the remaining capital in successive installments, a common financing strategy known as a "tranched" round. Our sources indicate that Rao's ambitious vision for Unconventional, Inc. is to directly compete with Nvidia by engineering a novel AI machine that integrates both custom-designed silicon chips and dedicated server infrastructure.

OpenAI s'apprête à accueillir sa troisième conférence annuelle des développeurs, DevDay 2025, ce lundi. Plus de 1 500 personnes sont attendues à Fort Mason, San Francisco, pour ce qui est annoncé comme le "plus grand événement jamais organisé" par l'entreprise, promettant une série d'annonces, des discours d'executifs d'OpenAI et une conversation informelle entre le PDG Sam Altman et le designer de longue date d'Apple, Jony Ive. Cet événement se déroule alors qu'OpenAI fait face à une concurrence plus intense que jamais dans sa quête pour séduire les développeurs. Au cours de la dernière année, les modèles d'IA d'Anthropic et de Google sont devenus de plus en plus performants pour les tâches de codage et la conception web. Cette évolution a contraint OpenAI à lancer des modèles d'IA améliorés à des prix plus compétitifs pour rester dans la course. En arrière-plan, Meta a constitué une équipe impressionnante de talents en IA au sein de son nouveau groupe, Meta Superintelligence Labs, qui pourrait représenter une menace supplémentaire pour OpenAI dans un avenir proche. Bien qu'aucune annonce de lancement ne soit confirmée pour DevDay 2025, les rumeurs vont bon train. OpenAI pourrait potentiellement dévoiler le navigateur alimenté par IA sur lequel il travaille, ou fournir une mise à jour sur l'appareil d'IA qu'il développe avec Ive et d'anciens cadres d'Apple. Des mises à jour concernant le GPT Store, dont OpenAI a peu parlé depuis son lancement l'année dernière, sont également possibles. TechCrunch sera sur place à San Francisco pour couvrir l'événement en direct, permettant de suivre toutes les actualités. Seul l'un des événements sera diffusé en direct pour les participants à distance. Les participants sur place pourront assister à des présentations et des conférences de personnalités telles que Michael Truell, PDG de Cursor, Daniel Lurie, maire de San Francisco, et Kimberly Tan, associée chez Andreessen Horowitz. Plusieurs employés d'OpenAI prononceront également des discours sur leurs travaux, notamment Laurentia Romaniuk, chercheuse sur le comportement des modèles, et Alexander Embiricos, responsable de Codex. DevDay 2025 proposera également une série de "sideshows" alimentés par l'IA. Parmi eux, le "Sora Cinema", décrit comme un "mini-théâtre confortable avec du pop-corn" présentant des courts métrages générés par le modèle vidéo d'OpenAI, Sora. Il y aura aussi une cabine téléphonique dotée d'un "portrait vivant" du célèbre informaticien Alan Turing "qui répond". En fin d'après-midi, deux événements majeurs clôtureront le DevDay. Ces deux dernières sessions ne seront pas diffusées en direct, mais elles seront mises en ligne sur YouTube plus tard dans la journée. À 15h15 PT, une "Developer State of the Union" se tiendra avec Greg Brockman, président d'OpenAI, et Olivier Godement, responsable des produits de la plateforme OpenAI. Les deux dirigeants d'OpenAI sont prévus pour "présenter de nouvelles capacités" et partager les perspectives pour les développeurs. Enfin, à 16h15 PT, Altman et Ive tiendront un "Closing Fireside Chat" pour discuter de "l'art de construire à l'ère de l'IA". Cette conversation durera environ 45 minutes.

Users facing an overflowing Snapchat Memories archive now have a clear choice: either subscribe to one of the platform's recently introduced storage plans or take steps to export their cherished Snaps to avoid permanent loss. Snapchat confirmed to TechCrunch that an introductory storage plan is available, providing up to 100GB of space for $1.99 monthly. For those with existing subscriptions, Snapchat+ users will automatically receive an increased capacity of 250GB as part of their $3.99 monthly fee, while Snapchat Platinum subscribers benefit from a substantial 5TB of storage for their $15.99 monthly premium. Opting not to subscribe means the platform will prioritize retaining older Snaps, with the most recent ones being purged to maintain the storage cap. Despite Snapchat offering a temporary 12-month grace period for Memories exceeding the standard 5GB limit, a segment of users is already preemptively choosing to export their content rather than committing to a recurring payment. This article will guide you through the available alternatives for safeguarding your Snapchat Memories, particularly focusing on the free export methods. One primary method for preserving Snapchat Memories involves downloading them directly to your device's camera roll. While a straightforward process from within the app, this approach is limited to batches of 100 Snaps at a time, potentially making it a laborious endeavor for users with extensive archives. Nonetheless, this method offers the advantage of selective saving, allowing users to curate precisely which Snaps they wish to keep. The procedure involves a series of steps that must be repeated for each batch of 100 Memories. For those seeking a less manual alternative to batch exports, Snapchat offers its "Download My Data" tool. This feature enables users to request their entire Memories archive, which is subsequently compiled and sent to them as a .zip file via email. Accessing this tool through the app's settings will guide users through the necessary steps to initiate the request. It is important to note that the time required to process and deliver this data can vary considerably depending on the total volume of content stored.

Asked for comment, a Google spokesperson indicated that no announcement was ready to be made “just yet.” These examples prominently displayed a variety of engaging prompts designed to demonstrate Gemini's potential. Users were shown suggestions for creative photo manipulation, such as “Teleport me to deep space,” “Give me a vintage or grunge look,” or “Turn my drawing into a storybook.” Beyond visual transformations, other prompts, often set against a vibrant background, highlighted diverse use cases for Gemini's AI, including “Brainstorm out loud with Live” or receiving a “daily news roundup.” The core objective behind this redesign is to proactively guide users, offering concrete suggestions on how to leverage Gemini’s AI, rather than compelling them to independently explore the chatbot's full range of capabilities. This revamp also aims to significantly enhance the Gemini app's visual appeal and overall user engagement.

The adoption of electronically latched door handles has become widespread among automakers in recent years, allowing for a sleeker design and enhanced aerodynamics. Yet, this trend often leads to manual emergency door releases being hidden in hard-to-find locations within vehicles. Rivian, however, is set to modify this approach with its R2 model, launching next year. According to Bloomberg, the automaker will now place the interior manual releases in a "more clearly visible" position, closer to the main electrically powered door handles.

The world of venture capital, often perceived as a realm of unwavering success, is paradoxically introduced within YC Arena with a potent reminder: "Many rejected founders went on to build incredibly successful companies afterwards," a note at the game's start declares. It continues, "Rejection means nothing — even the most successful founders got rejected multiple times,” setting a reflective tone on the journey of entrepreneurship. While YC Arena offers diverse challenges, from matching company names to their logos to guessing the year a startup completed YC based on its description (revealing a marked increase in AI-focused ventures in recent years), the "YC Partner Simulator" stands out. This particular game uniquely compels players to confront and refine their own decision-making processes, mirroring the high-stakes choices faced by actual investors. A core tenet emphasized by YC co-founder Paul Graham, pertinent both in startup pitches and journalistic correspondence, is the paramount importance of conciseness. "You have to be exceptionally clear and concise," he wrote. "Whatever you have to say, give it to us right in the first sentence, in the simplest possible terms." This advice proved insightful during a subsequent playthrough of the simulator. Rather than scrutinizing the specifics of a company's pitch, attention was redirected to the efficiency with which their core offering was articulated. While this strategy is admittedly not ideal for real-world startup evaluation, where the substance of the company's work is crucial, it significantly enhanced the accuracy of in-game selections. Ultimately, the efficacy of swift, focused evaluation underpins Y Combinator’s model. As Sam Altman observed in 2016, "It turns out that in 10 minutes, if the only question you’re trying to answer is, ‘Does this person have the potential to be the next Mark Zuckerberg?’ … You can answer that question in 10 minutes.” He acknowledged, "Not with 100 percent accuracy, obviously, but good enough that our business model works,” underscoring the surprising utility of rapid assessment in identifying entrepreneurial promise.

Supabase has successfully secured an impressive $380 million in funding over the past year, propelling its valuation upwards by more than 500%, according to PitchBook's estimates for its Series C valuation. The open-source database service now confirms a total of $500 million raised since its inception. Established in 2020 by CEO Paul Copplestone and CTO Ant Wilson, Supabase emerged as a Y Combinator startup several years before the LLM-powered 'vibe-coding' craze took hold. Its core offering provides developers with an open-source, Postgres-based alternative to Google's Firebase, a database also designed to power AI applications. Notably, leveraging its open-source nature and a robust community of 4 million developers, Supabase is extending an opportunity for its community members to acquire company stock as part of this latest Series E funding round.

Sora has made a notable entry into the mobile application market, with its iOS app garnering an estimated 164,000 installs within its first two days, September 30 and October 1. This initial performance positions Sora ahead of rivals such as Anthropic’s Claude and Microsoft’s Copilot in day-one downloads, while mirroring the debut success of xAI’s Grok. However, established players like OpenAI’s ChatGPT and Google’s Gemini iOS apps experienced even more robust launches, each securing at least 80,000 downloads on their respective first days. It is important to note that Sora is currently invite-only, which may skew direct comparisons. Had the new video app been accessible to all users, it is plausible that its install figures could have been considerably higher. To offer a more equitable comparison, Appfigures conducted an analysis focusing specifically on downloads within the U.S., supplemented by Canadian figures where apps were available at launch. This deeper dive revealed that ChatGPT and Gemini indeed outpaced Sora in day-one iOS downloads, recording 81,000 and 80,000 respectively. Sora tied with Grok, both achieving 56,000 day-one installs, and comfortably surpassed the launches of AI apps Claude, which saw 21,000 downloads, and Copilot, which registered just 7,000. Beyond download numbers, Sora also swiftly ascended the U.S. App Store’s top charts, clinching the No. 3 overall position by its second day. In comparison, ChatGPT reached the No. 1 spot on its second day, while Grok secured No. 4, Gemini No. 6, Copilot No. 19, and Claude No. 78.

The current landscape of artificial intelligence integration reveals a significant proliferation of diverse tools, with companies adopting a wide array of AI products for specific tasks. This rapid evolution means new applications emerge and recede with remarkable speed, highlighting a market that has yet to consolidate around a few dominant solutions in each category, as noted by a16z partners Olivia Moore and Seema Amble. A substantial portion of current AI expenditure is directed towards "human augmentors" or "copilots," designed to boost workforce productivity. This suggests that startups are not yet prepared for a complete transition to fully agentic workflows. However, this dynamic is expected to shift. As computer usage increasingly becomes a seamless mode of operation and the capability to build end-to-end agentic flows improves, a transition away from copilots towards comprehensive agent tools is anticipated, especially given the eagerness of users to experiment with such advancements. Unsurprisingly, the leadership board for AI tools remains largely dominated by major labs, with OpenAI securing the top position and Anthropic following closely at number two. The report also highlights a strong presence of "vibe-coding" tools, featuring Replit at third place, Lovable at 18th, Cursor at sixth, and Emergent at 48th. Cognition, specializing in enterprise-oriented coding tools like Devin and Windsurf, also made the list at 34th. The future trajectory of vibe coding remains an open question: will the sector consolidate into a single leading platform, or will multiple robust businesses emerge to serve various application types? The answer, for now, remains elusive. A noteworthy observation was the swift adoption of consumer-oriented tools such as CapCut and Midjourney by startups. Moore remarked on how these "delightful consumer tools" are rapidly pulled into enterprise environments as individuals adopt and integrate them into their teams and workplaces. In terms of application types, horizontal applications constituted over 60% of the listed tools, while vertical applications accounted for 40%. The most prominent vertical software companies were concentrated in sales, recruiting, and customer service. Crucially, the report also identified AI making inroads into numerous sectors that previous generations of startups found challenging to penetrate. Moore articulated this transformation, stating that what were once service firms or consultancies are now manifesting as software companies in the age of AI. Amble offered Crosby Legal as an example, a platform that can quickly review legal contracts, effectively replacing a traditional meeting with an in-house general counsel. Currently, most tools function as copilots, assisting employees in making faster decisions rather than supplanting entire workforces with automated agents. However, as AI technology advances, leading to the development of full agent co-workers, the balance is projected to shift significantly towards end-to-end agents, capable of executing tasks like outreach far more rapidly than humans. A final, significant finding in the report underscores the escalating intertwining of consumer and enterprise businesses. Individuals are increasingly bringing personal applications used at home into their professional lives, and founders are leveraging their preferred personal tools to build their companies. This marks a departure from a previous era where a clear delineation existed, with a distinct technology stack reserved for startup development. Canva serves as a prime example, a popular consumer app that also commands a substantial enterprise audience, despite taking years to introduce an enterprise plan. As the distinctions between individual and enterprise use cases blur, companies are demonstrating a greater willingness to integrate both spheres. This convergence expands the total addressable market (TAM) beyond one segment to encompass both, potentially prompting companies to "professionalize" more quickly by establishing enterprise teams – including go-to-market, sales, and support – to secure enterprise revenue sooner, rather than relying solely on individual consumers. Looking ahead, Moore and Amble anticipate rapid changes to this landscape in the coming years. Established companies are now actively launching AI features to maintain relevance and accessibility, while new entrants continually emerge with innovative ideas. The concept of a "legacy player" in this context can often refer to a company from just 12 months prior. The question remains whether the same note-taking applications, for instance, will still be prominent in another year, or if an entirely new set will have taken their place.

Enterprise software titan Salesforce is poised to capitalize on the burgeoning trend of 'vibe-coding' – where developers articulate their needs in natural language and AI agents then generate the corresponding code – with its newly unveiled AI-powered developer tool. The company officially announced its latest offering, Agentforce Vibes, on Wednesday. This innovative coding tool is designed to empower developers to work with greater autonomy on Salesforce applications and agents by automating much of the technical implementation. Agentforce Vibes promises to support developers through every stage, from the initial app idea and building phases to observability, all while embedding robust enterprise security and governance controls. Central to this new tool is Vibe Codey, an autonomous AI coding agent. Seamlessly connected to a company’s existing Salesforce account, Vibe Codey can leverage and reuse previously written code and adhere to established coding guidelines, ensuring new applications are consistent with existing products. Dan Fernandez, Salesforce’s Vice President of Product for Developer Services, conveyed to TechCrunch that Agentforce Vibes’ deep integration with a company's current Salesforce account offers enterprises an ideal scenario. It enables them to explore vibe-coding without the typical security concerns or the burden of starting each new project from scratch. "We’re trying to give you everything," Fernandez asserted, underscoring the tool's comprehensive nature. He elaborated that users are spared the time-consuming process of setting up model context protocols – systems that facilitate secure communication between AI models and external tools – along with development environments and various tools. Instead, everything, including AI requests to get started, comes prebuilt and ready, a key differentiator that significantly lowers the barrier to entry. Fernandez further clarified that Agentforce Vibes represents not Salesforce's first foray into vibe-coding, but rather the latest sophisticated addition to its expanding suite of AI developer tools. This release marks a significant convergence, bringing together powerful capabilities. "We are taking the power of the client tools and Agentforce for developers and making it tailored to Salesforce development," Fernandez explained, emphasizing an end-to-end enterprise vibe coding experience tailored for the "agentic enterprise." The company's decision to utilize Cline within Agentforce Vibes followed an evaluation of numerous open-source coding tools. Fernandez noted that Cline's strong support for Model Context Protocols (MCP), crucial for AI models to securely interact with external tools and data, was a decisive factor. This launch arrives at an intriguing moment for the broader vibe-coding industry. However, potential cost pressures often associated with such innovations are mitigated when vibe coding is integrated into a larger product suite, as is the case with Salesforce’s Agentforce Vibes. Each Salesforce organization will receive 50 requests per day utilizing OpenAI’s GPT-5 model. Any additional requests beyond this limit will then be processed through a Salesforce-hosted Qwen 3.0 model. Salesforce is currently offering Agentforce Vibes free of charge to its existing users, with priced usage plans anticipated for the future.

Senator Ted Cruz (R-TX) has effectively halted a legislative initiative designed to extend data privacy protections, currently afforded to federal lawmakers and public officials, to every citizen across the United States. On the Senate floor, Senator Wyden, the bill’s sponsor, voiced his opposition to preferential treatment for members of Congress. He underscored the necessity for constituents to receive comparable protections against violence, stalking, and other criminal threats. Cruz emerged as the sole senator to object to Wyden’s proposal. He asserted, without presenting supporting evidence, that the bill possessed the potential to disrupt crucial law enforcement operations, specifically referencing the ability to track the residences of sexual predators. However, Daniel Schuman, executive director of the non-profit American Governance Institute, countered Cruz's claim, branding it "wrong." Schuman clarified that federal law already mandates a publicly accessible registry of sex offenders, which lists their names and addresses online, ensuring parents and communities are aware of their whereabouts. Furthermore, Schuman explained that Wyden's legislation explicitly preserves this existing system, explicitly permits law enforcement to share information with at-risk individuals, and includes exemptions for the press, allowing news outlets to continue reporting freely on offenders. He stressed that nothing within the bill’s provisions would hinder parental notifications or restrict public access to this critical information. Following the objection, Senator Cruz indicated a broader interest, stating he was "interested in expanding the protection to as wide a universe as is feasible, as is practicable," though he noted that the precise method for achieving this had "not yet worked out."

Job seekers across various sectors should prepare for a notable increase in initial screening interviews. While this might initially sound promising, it does not signify a sudden surge in available job openings. Instead, recruiters, often overwhelmed by the extensive process of identifying qualified applicants for subsequent stages, are progressively delegating routine screening tasks—such as verifying backgrounds, assessing salary expectations, and checking availability—to artificial intelligence. At the forefront of this shift is Alex, a startup co-founded by Aaron Wang approximately 18 months ago. Wang, who previously held positions at Facebook and worked as a quant for a hedge fund, informed TechCrunch that Alex's voice AI tool is capable of conducting autonomous interviews with applicants soon after they submit their job applications. He emphasized the platform's reach, stating, "Our AI recruiter does thousands of interviews a day and helps people get hired at some of the biggest companies in the world." While Wang refrained from naming specific clients, he revealed that Alex's customer base includes Fortune 100 companies, prominent financial institutions, nationwide restaurant chains, and leading Big 4 accounting firms. Investors widely perceive the growth of AI interviewers as an inevitable trend that many businesses will embrace. This conviction was a key factor in Alex securing a new $17 million Series A funding round, led by Peak XV Partners. The round also saw participation from Y Combinator and Uncorrelated Ventures, alongside several chief human resources officers (CHROs) from undisclosed Fortune 500 companies, among others. This latest funding follows the company's $3 million seed round last year, which was led by 1984 Ventures. Alex, however, is not alone in offering AI recruiting services. The startup faces competition from other early-stage companies in the space, including HeyMilo, ConverzAI, and Ribbon. Alex's long-term ambition is to interview millions of job applicants, thereby creating professional profile data that is significantly richer and more comprehensive than what LinkedIn currently provides. Wang articulated this vision, asserting, "Our thesis is that a 10-minute conversation with you tells me a whole lot more about you than your LinkedIn profile does." For the immediate future, Alex is focused on empowering recruiters by freeing up their time to build relationships with pre-qualified candidates and to offer strategic advice to hiring managers.

The burgeoning trend of "vibe coding," which leverages AI-powered tools to create applications and websites through natural language prompts, is experiencing an undeniable surge in popularity. This rapid ascent has propelled a multitude of companies in the sector to remarkable growth, with many competitors quickly gaining significant traction. Nikhil Basu Trivedi, co-founder and general partner at VC firm Footwork, aptly notes, "This is one of those spaces where every company is growing like a weed." Despite this impressive expansion, Basu Trivedi highlights a crucial limitation among many existing vibe-coding startups like Lovable and Replit: while they excel at prototype development, they often fall short in enabling users to launch production-ready software. The core issue, he explains, lies in their failure to provide the comprehensive infrastructure nontechnical users require to bring a functional product to market. It is against this backdrop of rapid growth and identified need that Anything emerges, securing an $11 million financing round at a $100 million valuation. This investment was led by Footwork, with additional backing from Uncork, Bessemer, and M13, a testament to the company's compelling growth rate that convinced Basu Trivedi of its potential even in a crowded market. Co-founded by former Google colleagues Dhruv Amin and Marcus Lowe, Anything is specifically engineered to empower nontechnical individuals to generate complete web and mobile applications. Amin candidly remarks on the shortcomings of other vibe-coding tools, stating, "You haven’t really seen real businesses built on top of any of these tools." Anything's ambition is clear: "We want to be the Shopify of the space, where people build apps that make money on top of us." This vision is already manifesting, with Amin reporting that users have successfully utilized Anything to develop fully functional applications now available on the AppStore, including a habit tracker, a CPR training course, and a hairstyle "try-on" app. Crucially, some of these applications are beginning to generate revenue. Amin attributes this success to Anything's integrated approach, which eliminates the need for users to configure and connect various other essential tools to their AI-generated prototypes, providing an end-to-end solution. The concept of an AI-assisted, soup-to-nuts app builder crystallized for Amin and Lowe less than a year ago. Their partnership dates back to 2021, initially focusing on a bootstrapped development marketplace that combined AI coding tools with human developers. However, the pre-LLM era business, which was generating an annualized run rate of approximately $2 million, revealed a clear path forward: generative AI offered the promise of delivering applications faster and more cost-effectively than their existing marketplace model. Consequently, in 2023, they pivoted, winding down their previous venture to concentrate on developing their AI-powered app-building tool, securing pre-seed and seed funding from Uncork and Bessemer Venture Partners along the way. Despite the intense competition within the sector, Basu Trivedi remains unfazed. He believes there's ample demand, asserting, "It seems there’s enough demand out there for different types of app-building products," suggesting a broad market capable of supporting diverse solutions.

Les sociétés de capital-risque sont convaincues d'avoir découvert la prochaine grande avancée en matière d'investissement : exploiter l'intelligence artificielle pour générer des marges dignes des logiciels au sein d'entreprises de services traditionnellement intensives en main-d'œuvre. Cette stratégie repose sur l'acquisition de firmes de services professionnels matures, l'implémentation de l'IA pour automatiser des tâches, puis l'utilisation des flux de trésorerie améliorés pour acquérir davantage de sociétés. General Catalyst (GC) est en première ligne de ce mouvement, ayant alloué 1,5 milliard de dollars de sa dernière levée de fonds à ce qu'elle nomme une stratégie de "création". Celle-ci vise à incuber des entreprises logicielles natives de l'IA dans des secteurs spécifiques, puis à utiliser ces entreprises comme véhicules d'acquisition pour racheter des firmes établies – et leurs clients – dans les mêmes domaines. GC a déjà investi dans sept industries, des services juridiques à la gestion informatique, avec l'intention d'étendre son champ d'action à une vingtaine de secteurs au total. Selon Bhargava, si l'on parvient à automatiser les activités de services, en abordant 30% à 50% des tâches de ces entreprises avec l'IA, et même en automatisant jusqu'à 70% des tâches essentielles dans le cas des centres d'appels, les calculs deviennent irrésistibles. GC cherche à doubler – au minimum – la marge d'EBITDA des entreprises qu'elle acquiert, a-t-il expliqué. Cette firme de premier plan n'est pas la seule à adopter cette vision. La société de capital-risque Mayfield a spécifiquement alloué 100 millions de dollars à des investissements dans les "coéquipiers IA", notamment Gruve. Cette startup de conseil en informatique a acquis une entreprise de conseil en sécurité de 5 millions de dollars, puis l'a fait passer à 15 millions de dollars de revenus en six mois, tout en atteignant une marge brute de 80%, selon ses fondateurs. Toutefois, cette tendance n'est pas sans impact sur les organisations. Des employés interrogés indiquent qu'ils consacrent en moyenne près de deux heures à chaque incident de "workslop" – des imperfections générées par l'IA – incluant le déchiffrage, la décision de le renvoyer ou non, et souvent la correction directe. Bhargava a contesté l'idée selon laquelle l'IA serait surévaluée, arguant au contraire que ces échecs de mise en œuvre valident en réalité l'approche de GC. "Je pense que cela montre l'opportunité, c'est-à-dire qu'il n'est pas facile d'appliquer la technologie de l'IA à ces entreprises", a-t-il déclaré. "Si toutes les entreprises du Fortune 100 et tous ces acteurs pouvaient simplement faire appel à un cabinet de conseil, intégrer de l'IA, signer un contrat avec OpenAI et transformer leur activité, alors notre thèse serait évidemment un peu moins robuste. Mais la réalité est qu'il est vraiment difficile de transformer une entreprise avec l'IA." Il a souligné la sophistication technique requise en IA comme la pièce manquante la plus critique du puzzle. "Il existe de nombreuses technologies différentes. Elles sont bonnes pour différentes choses", a-t-il précisé. "Il faut vraiment ces ingénieurs spécialisés en IA appliquée, issus d'entreprises comme Rippling, Ramp, Figma et Scale, qui ont travaillé avec les différents modèles, comprennent leurs nuances, savent lesquels sont adaptés à quoi, et comment les intégrer dans un logiciel." Cette complexité est précisément la raison pour laquelle la stratégie de GC, qui consiste à associer des spécialistes de l'IA à des experts du secteur pour construire des entreprises à partir de zéro, est pertinente, a-t-il soutenu. Néanmoins, il est indéniable que le "workslop" menace de compromettre – dans une certaine mesure – l'économie fondamentale de la stratégie. Même si une société holding est créée comme point de départ, si les entreprises acquises réduisent leurs effectifs comme le suggère la thèse de l'efficacité de l'IA, elles disposeront de moins de personnel pour détecter et corriger les erreurs générées par l'IA. Si les entreprises maintiennent leurs niveaux d'effectifs actuels pour gérer le travail supplémentaire créé par les sorties problématiques de l'IA, les énormes gains de marge sur lesquels les VCs comptent pourraient ne jamais se concrétiser. En fait, parce qu'elles acquièrent généralement des entreprises disposant déjà de flux de trésorerie, GC affirme que ses entreprises issues de la "stratégie de création" sont déjà rentables – un net écart par rapport au modèle traditionnel des VCs qui consiste à soutenir des startups à forte croissance et grandes consommatrices de liquidités. C'est également un changement probablement bienvenu pour les commanditaires (LPs) des sociétés de capital-risque, qui ont financé des années de pertes pour des entreprises qui n'ont jamais atteint la rentabilité. "Tant que la technologie de l'IA continuera de s'améliorer, et que nous verrons cet investissement massif et cette amélioration des modèles", a conclu Bhargava, "je pense qu'il y aura de plus en plus d'industries dans lesquelles nous pourrons aider à incuber des entreprises."

Spotify has unveiled a series of significant updates to its artificial intelligence policy, designed to enhance transparency regarding AI-generated music, vigorously combat spam, and unequivocally state that unauthorized voice clones are strictly prohibited on its platform. A cornerstone of these revisions is the adoption of the DDEX system, which mandates that labels, distributors, and music partners provide standardized AI disclosures within music credits. This mechanism will offer granular detail on AI integration, specifying, for instance, whether AI was utilized for AI-generated vocals, instrumentation, or post-production processes. As Sam Duboff, Spotify’s Global Head of Marketing and Policy, explained during a recent press briefing, "We know the use of AI is going to be a spectrum, with artists and producers incorporating AI in various parts of their creative workflow." He further noted that this industry standard would "allow for more accurate, nuanced disclosures," moving beyond a simplistic binary classification of AI use. Spotify simultaneously reinforced its stance on AI-enabled personalization, explicitly stating that unauthorized AI voice clones, deepfakes, and any other form of vocal replicas or impersonation are impermissible and will be promptly removed from the platform. While the DDEX standard is still in its developmental phase, Spotify has already secured commitments from 15 key labels and distributors who intend to integrate this technology, viewing its proactive step as a signal for broader industry adoption. Recognizing that AI tools significantly lower the barrier to music creation and distribution, potentially leading to an increase in spam, Spotify is also rolling out a new strategy to mitigate this issue. Starting this fall, the company will introduce an advanced music spam filter. This system is designed to identify and tag spam tactics, effectively preventing such tracks from being recommended to users. Duboff elaborated on the necessity of this measure, stating, "We know AI has made it easier than ever for bad actors to mass upload content, create duplicates, use SEO tricks to manipulate search or recommendation systems…we’ve been fighting these kinds of tactics for years." He acknowledged the escalating sophistication, adding, "But AI is accelerating these issues with more sophistication, and we know that requires new types of mitigations." The filter’s implementation will be gradual, allowing Spotify to fine-tune its targeting signals and adapt as the market evolves. In a related effort, Spotify will collaborate with distributors to tackle "profile mismatches," a fraudulent practice where music is illicitly uploaded to another artist’s profile across various streaming services. The company aims to preempt these incidents before the music goes live. Despite these stringent new measures, Spotify executives underscored their continued support for the legitimate and non-fraudulent use of AI. Charlie Hellman, Spotify’s VP and Global Head of Music, clarified the company’s position: "We’re not here to punish artists for using AI authentically and responsibly. We hope that artists’ use of AI production tools will enable them to be more creative than ever." He concluded by emphasizing the imperative to "stop the bad actors who are gaming the system, and we can only benefit from all that good side of AI if we aggressively protect against the downside." While Spotify refrained from disclosing specific metrics on AI-generated content, Duboff informed reporters that "the reality is, all streaming services have almost exactly the same catalog." He further explained that merely uploading tracks does not equate to listenership or monetization, reiterating that "AI usage is increasingly not a binary, but kind of a spectrum of how artists and producers are using it."