Today's briefs
Harvey Releases Tenet: A Kimi K3-Based Model Post-Trained for Long-Horizon Legal Agent Tasks
Harvey has released Harvey Tenet, a specialized legal AI model built by post-training the Kimi K3 base model using Fireworks AI infrastructure, targeting long-horizon agentic workflows in legal work. The model is designed for complex, multi-step legal tasks such as contract analysis, due diligence, and case research that require sustained reasoning across large document contexts. This release is notable as a concrete example of domain-specific post-training on a capable open base model to create a vertically specialized agent — a pattern increasingly viable for enterprise AI teams. Developers building legal tech or vertical AI agents can study this architecture: Kimi K3 as a strong reasoning base, combined with domain-specific RLHF or supervised fine-tuning via a scalable serving layer like Fireworks. It also signals that legal AI is moving beyond simple retrieval-augmented generation toward genuine long-horizon autonomous task execution.
MarkTechPost
Vercel Launches 'Is Agentic': A Free Tool That Scores Any Website for AI Agent Readiness
Vercel has released 'Is Agentic,' a free public tool that audits any website's readiness to be consumed and navigated by AI agents, running over 100 checks powered by Ora. The tool evaluates factors like structured data availability, navigation clarity, API surface exposure, and other signals that determine how effectively an autonomous agent can interact with a site. For developers building agentic applications or web infrastructure intended for agent consumption, this provides an actionable checklist and scoring system to identify gaps before deploying agent-facing products. It also reflects a broader industry shift toward designing for agent-first consumption rather than human-first interfaces. Teams shipping public-facing products should audit their sites now as agentic traffic becomes a first-class concern.
MarkTechPost
Alibaba Raises $10.2 Billion in New Share Offer to Fund AI Expansion
Alibaba has priced a $10.2 billion new share offering, with proceeds earmarked to accelerate its AI and cloud computing expansion. This is one of the largest capital raises in the tech sector this year and positions Alibaba to compete aggressively in the global AI infrastructure and model race, particularly against domestic rivals Baidu, ByteDance, and Tencent, as well as Western hyperscalers. For developers working in or evaluating cloud AI platforms, Alibaba's Qwen model family and associated cloud services are likely to see significant investment and capability growth as a result. The raise also signals continued institutional confidence in Chinese AI as a commercial force despite geopolitical headwinds. Developers building multi-cloud or Asia-Pacific-facing AI systems should monitor Alibaba's roadmap closely over the next 12–18 months.
Tech - South China Morning Post
Building End-to-End Document Intelligence Pipelines with deepDoctection
deepDoctection is an open-source Python framework designed to construct full document intelligence pipelines, handling tasks such as layout detection, table extraction, OCR, and semantic structuring from PDFs and scanned documents. The framework integrates with popular deep learning backends and provides modular components that developers can chain together to process complex, unstructured document formats at scale. For teams building RAG pipelines, legal tech, financial document processing, or compliance tooling, deepDoctection offers a more structured alternative to ad hoc PDF parsing libraries. Its end-to-end design means developers can move from raw document ingestion to structured, queryable output with significantly less custom glue code. This is particularly valuable as document-heavy enterprise AI use cases expand and the quality of document parsing becomes a key bottleneck in pipeline accuracy.
MarkTechPost
Galbot Robots Complete 100 Consecutive Autonomous Tennis Rallies at World Humanoid Robot Games
Galbot's humanoid robots achieved 100 consecutive autonomous tennis rallies at the World Humanoid Robot Games, demonstrating a significant milestone in real-time dynamic motion planning and physical AI coordination. The feat required continuous closed-loop perception, trajectory prediction, and motor control without human intervention across an extended sequence — conditions that stress-test the robustness of embodied AI systems far beyond lab demonstrations. For developers working in robotics, reinforcement learning, or physical AI, this benchmark signals that humanoid dexterity in dynamic, unstructured tasks is advancing rapidly from proof-of-concept to repeatable performance. The event also highlighted Chinese robotics firms accelerating practical humanoid capabilities alongside internationally recognized competitors. This is worth watching for teams building simulation environments, robot learning frameworks, or deploying physical AI in real-world settings.
Unite.AI
China's Rural Cities Are Becoming the Engine of Its National AI Compute Infrastructure
A feature from the South China Morning Post details how rural Chinese cities, previously known for agriculture and livestock, are being transformed into major AI data center hubs as part of China's national strategy to build distributed compute infrastructure away from coastal urban centers. The buildout is driven by lower land costs, access to renewable energy, and government incentives designed to reduce concentration risk in AI infrastructure and support regional economic development. For developers and infrastructure architects, this signals that China's AI compute capacity is scaling geographically in ways that will meaningfully increase total national throughput over the next several years. It also reflects a deliberate industrial policy approach to AI infrastructure that differs from the market-driven clustering patterns seen in the US. Teams assessing competitive AI capacity globally should factor in this distributed buildout as a structural contributor to Chinese AI capabilities.
Tech - South China Morning Post
China's Mid-Tier Internet Companies Are Quietly Embedding AI Across Everyday Business Operations
A South China Morning Post analysis reports that China's second-tier internet companies — firms below the visibility of Alibaba, Tencent, and ByteDance — are aggressively integrating AI into core business operations including customer service, logistics, content moderation, and supply chain management. Unlike the headline-grabbing model races, this wave of deployment is characterized by practical, ROI-driven adoption targeting operational efficiency rather than frontier model capabilities. For enterprise AI developers, this pattern illustrates how mid-market AI integration is becoming a competitive baseline in China's digital economy, potentially pressuring global counterparts in similar sectors. The piece suggests that the real commercial impact of AI is currently happening in unsexy, back-office automation rather than consumer-facing AI products. Developers building enterprise AI tooling should take note of the deployment patterns emerging in this segment as a preview of what broader enterprise adoption looks like at scale.
Tech - South China Morning Post