PANews has published a long first-person essay by Digital Life Kha'Zix on what happened after his company spent the past two and a half years pushing AI across nearly every function. The firm, he wrote, now has close to 100% AI penetration, with agents used in finance, HR, legal, business development, talent management, and operations. Yet instead of shrinking, the company is hiring more people and preparing to move into a larger office because the current space is no longer enough. The core argument is that broad AI adoption did not turn the company into a lean, automated machine run by a small number of people. In his account, it had the opposite effect: the more routine, describable, and repeatable work agents absorbed, the more the organization was pushed back toward old-fashioned human tasks such as trust-building, judgment, responsibility, training newcomers, and direct communication with clients and creators. He argues for a "thin middle platform, thick front line" structure, saying companies with fewer than 100 native employees should be cautious about building a heavy AI platform team. He also says data accumulation and data governance matter more than agents themselves. The essay closes on a broader point: AI may change tools, workflows, and efficiency, but the real test for any organization still comes down to how it treats people.

TechFlowPost’s elsewhere project has published an online exhibit tracing the funding history of Unitree Robotics through archived WeChat messages, investment memos, photos and videos from early backers. The materials show that founder Wang Xingxing received a RMB 2 million investment from Variables Capital in 2018, described as Unitree’s first institutional financing and the second investment after Yin Fangming. The records also revisit a period when fundraising was difficult, cash was tight, and Wang at one point used his own money to cover payroll. The exhibit lays out a multi-stage financing path. After early backing from Yin Fangming, Variables Capital, Anchuang Technology and Binhe Investment, firms including Sequoia China, Vertex Ventures, Shunwei Capital, Chuxin Capital and Matrix Partners China entered from 2019 onward. A turning point came before the 2024 Lunar New Year, when investors such as Meituan, Jinshi Investment and Source Code Capital joined Unitree’s B2 round. The story also includes internal assessments of Wang, 2020 operating metrics showing an 18-person company with RMB 11.8282 million in annual revenue, and investor notes that documented Unitree’s product focus, customer base and pricing against Boston Dynamics.

A Foresight feature, based on a past AMA organized by ULTILAND, argues that South Korea’s growing influence in contemporary art and creative design now extends well beyond the usual focus on Seoul. In the discussion, ULTILAND art project lead Hannah and co-founder Tangtang describe Korea’s ascent as the product of a broader cultural ecosystem built over time, one that includes museums, galleries, art schools, collectors, biennials, and a strong youth culture. They also point to the country’s global cultural export machine, from K-pop and film to fashion and beauty, as a factor that has drawn wider attention to Korean art, architecture, and design. The conversation maps Seoul through several distinct art districts, including Samcheong-dong and Bukchon, Hannam-dong, Gangnam and Cheongdam, and Seongsu-dong, while naming the National Museum of Modern and Contemporary Art and Leeum Museum as key institutions. It also highlights Frieze Seoul and Kiaf Seoul as a two-way link between international art systems and Korea’s domestic market. Beyond the capital, the speakers single out Gwangju, Busan, and Jeju as cities with different cultural logics, ranging from biennial-driven public discourse to coastal urban culture and nature-linked art spaces. The feature says Korea’s appeal lies not only in institutions or fairs, but in the way art is woven into everyday urban life.

Bloomberg reported that internal Slack messages show Phoebe Gates, daughter of Bill Gates, and Phia co-founder Sophia Kianni knew as early as December last year that their shopping comparison startup had been using cookie stuffing to claim affiliate marketing commissions. The practice involves placing tracking cookies in a user’s browser without a legitimate referral click, allowing a company to take credit for sales it did not drive. Corporate lawyer Ariel Givner wrote on X that this conduct is typically treated as federal wire fraud in U.S. courts and could carry a maximum penalty of up to 20 years, along with fines and restitution. Bloomberg had previously reported on July 9 that Phia’s browser extension opened background tabs during checkout, inserted its own cookies, and in some cases overwrote valid referral records. After the feature was shut down, charts reviewed by Bloomberg showed Phia’s average daily revenue fell from about $80,000 to between $10,000 and $28,000. The report also said cookie stuffing accounted for 51% of all sales attributed to Phia in June. No prosecutors have formally charged Gates or Kianni at this stage.

A nostalgia trend branded "2026 is the new 2016" is spreading across TikTok, but the objects of affection are not the polished highlights of a past era. They are dog-face filters, flower-crown overlays, blurry iPhone 6 photos, oversaturated selfies, rough Tumblr collages, and challenge videos that once looked cheap because they were everywhere. According to BlockTempo, TikTok searches for "2016" jumped 452% in a single week, Spotify playlists themed around 2016 rose 71% from a year earlier, and Google Trends hit a record high for the term in mid-January. GWI found that 42% of Gen Z said they missed the 2010s most, the highest share among all generations. The article argues that this is not nostalgia in the strict psychological sense. Drawing on the work of University of Southampton researchers Constantine Sedikides and Tim Wildschut, it says nostalgia usually helps individuals restore continuity in their own identity. What is happening here looks more collective. In BlockTempo’s reading, young users are not simply revisiting the past for comfort; they are marking the loss of an internet where human beings were assumed to be the primary authors. The piece backs that claim with several data points: a joint estimate from MIT CSAIL and the Oxford Internet Institute said 64% of newly published online content in 2026 was AI-generated; Reuters’ Digital Media Report 2026 said 79% of visual content on Instagram, TikTok and Pinterest came from AI; and Imperva found automated traffic had crossed half of all web traffic for the first time in a decade.

Xiaohongshu’s dots model lab has released dots3-note preview as the first open-source entry in its dots3 family, extending public access to the same model line that previously drew attention after dots-note-3.0 achieved an officially certified perfect 42 at the International Mathematical Olympiad. The newly released model has 280B total parameters, 16B active parameters, a 512K context window, and multimodal understanding across text, vision and speech. Rather than centering the release on math alone, Xiaohongshu positions dots3-note preview around long-horizon, open-ended tasks such as travel planning, wedding preparation, store operations and home design, where goals are fluid and answers are rarely unique. The company also outlined the training and evaluation stack behind the release. That includes TEMPO, short for Test-time-scaled Value Estimation with Macro-step Policy Optimization, a reinforcement learning method built for long trajectories; a Self-Critiquing mechanism used both in training and inference; and two newly open-sourced benchmarks, VibeSearchBench and VibeLifeBench. Xiaohongshu said the preview model can handle tasks ranging from playing Slay the Spire II and solving ARC-AGI 3 to reading floor plans, generating decoration proposals, and building a native Apple Vision Pro app end to end. According to the company, a full dots3-note release is planned in the near term, while the broader dots3 lineup will also include jazz and aria.

ChainFeeds published its Aug. 14 research roundup, pulling together five feature pieces selected from its Aug. 13 Web3 briefing. The package spans Bitcoin market structure, crypto venture and trading narratives, wallet design for AI agents, Ethereum’s long-term cryptography roadmap, and startup advice from Y Combinator CEO Garry Tan. The Bitcoin section, citing Glassnode, says BTC remains stuck between two key on-chain cost bases: the Median Realized Price near $63,000 and the Short-Term Holder Cost Basis near $68,700. Spot activity has fallen to one of the lowest levels in years, ETF demand has yet to show a convincing return, and leveraged longs in derivatives have built up ahead of a broader recovery. The report says a downside break would put focus on the June low around $58,500. Other pieces in the roundup argue that stablecoins and perpetuals remain crypto’s clearest native product-market fits, examine how products such as x402 and MetaMask Agent Wallet try to give AI agents controlled spending and execution power, and frame recursive STARK aggregation as a key tool for Ethereum’s post-quantum, privacy, and scaling agenda. The final feature highlights Garry Tan’s warning that chasing hot sectors can pull founders away from areas where they have genuine edge, while AI-driven coding is changing what counts as a durable moat.

Andreessen Horowitz’s latest conversation with Y Combinator president Garry Tan centered on a blunt lesson from his own career: founders lose more by chasing what looks hot than by making a single bad product call. Tan revisited two decisions he now sees as defining mistakes — moving away from web programming after the Web 1.0 crash in 2003 and turning down an offer to join Palantir, which he said may have cost him roughly $2 billion to $4 billion in hindsight. In both cases, he said he was reading the map instead of the terrain, paying too much attention to what looked impressive and too little to what the smartest people around him were actually building. The discussion then turned to AI’s effect on startups, software and management. Tan said coding has become far less scarce, which shifts the bottleneck toward judgment, initiative, data and network effects. He argued that a pure per-seat SaaS business without defensible data or user lock-in may not hold up over the next five to 10 years. He also described how AI agents, reusable workflow files and long-context systems could change how founders operate, and pointed to a Brex example to show how agents may alter the role of middle management. On the longer horizon, Tan said AI’s full organizational impact may take 20 years to play out, while the next consumer battleground could center on voice, memory and what he called the 2027 “harness wars.”