Loading…
Latest reads
Every post, summarised
Grounded summaries of engineering blogs, newest first. Each links to the original.
- 4,671 posts
- 20 companies
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
Latest posts
GLM-5.2: Built for Long-Horizon Tasks
GLM-5.2 is presented as a flagship model for long-horizon tasks, extending GLM-5.1’s capability to a solid 1M-token context for sustained engineering work. It was trained extensively on 1M-context coding-agent scenarios involving large-scale implementation, automated research, performance optimization, and complex debugging, and adds effort-level controls to balance capability, latency, and computational cost. Its architecture uses IndexShare, sharing one lightweight indexer across every four transformer layers and reducing per-token FLOPs by 2.9× at 1M context; MTP changes for speculative decoding increase acceptance length by up to 20%. On reported long-horizon and standard coding benchmarks, GLM-5.2 is described as the highest-ranked open-source model: it trails Opus 4.8 by 1% on FrontierSWE, ranks second only to Opus 4.8 on PostTrainBench, and scores 81.0 on Terminal-Bench 2.1 versus GLM-5.1’s 63.5.
What’s New in the AI Platform: Agents for ML Engineering, Our Deep Learning Platform, and New Capabilities for Real-Time ML
The announcement presents three additions to the Databricks AI Platform: Genie Code support for ML engineering, AI Runtime’s serverless GPU environment, and expanded real-time ML capabilities. Genie Code integrates with Unity Catalog, Feature Store, training, serving, monitoring, and MLflow, assisting with feature engineering, model training, deployment, evaluation, and production operations. AI Runtime provides on-demand serverless NVIDIA A10 and H100 GPUs, supports high-performance multinode training with RDMA and high-performance data loading, and adds Lakeflow Jobs, DABs, MLflow, and Unity Catalog integration. For real-time ML, the platform adds declarative feature engineering, streaming features, online feature serving on Lakebase, and enhanced Model Serving targeting 300K+ QPS with under 10ms p99 latency overhead. Reported customer examples include faster workflows, lower infrastructure costs, and production scaling beyond 100K QPS.
Introducing the Agentic CDP: A New Species of CDP for a New Era of Agents
Traditional customer data platforms (CDPs) were built for human-managed, batch-based campaigns, but the post argues that agentic buying requires millisecond speed, hyper-personalization, and richer context. It contrasts the familiar Golden Record with Golden Context, which combines customer data with current business goals and the history and outcomes of prior decisions. The proposed Agentic CDP uses “Infinity Campaigns,” always-on engagement loops that use LLMs and agents to adapt messaging, timing, and channels for individuals. It is also embedded in the data foundation, bringing customer, business, and decision context together under existing governance, and is designed for agents and humans from the outset. Databricks presents CustomerLake as an implementation of these principles for its platform.
Cloud Environment Setup and Cloud Subagents in Agents Window
This release updates cloud agents in the Agents Window of the Cursor desktop app, adding cloud environment setup, cloud subagents, and more reliable movement between local and cloud sessions. Cursor can help configure a development environment in under 10 minutes while showing progress in a shared terminal, then capture the result as a reusable snapshot for faster startup and software testing by future agents. The /in-cloud command launches a subagent in its own virtual machine and branch, isolating long-running or parallel tasks while the local workspace remains responsive; /babysit lets a cloud agent iterate on a pull request remotely. Cloud subagents can run in the background, and sessions can be pulled back to a local computer for testing, with multiple cloud agents running in parallel.
Kimi K2.7 Code vs Claude Fable 5: Landing pages that cost 94% less
The experiment compares Kimi K2.7 Code and Claude Fable 5 across 12 landing-page prompts covering B2B SaaS, a rooftop speakeasy, developer tools, and other categories. Kimi cost about 94% less overall, averaging roughly 16 times cheaper than Fable, while producing similar-quality output, especially when given visual references. The team first established a prompt-only baseline, then used a custom design MCP server to provide screenshots, UI elements, and other references directly to multimodal Kimi; this improved hierarchy, typography, composition, readability, loading behavior, and image handling. GPT-5.5 scored screenshots and source code on a 0–100 rubric, and although Fable scored higher, the relatively small quality gap led the authors to consider Kimi’s lower cost a reasonable trade-off for iterative landing-page generation.
Announcing the Agentic Resource Discovery specification
Agentic Resource Discovery (ARD) is an open specification for publishing, discovering, and verifying AI capabilities across the web, addressing the lack of a standard that works across organizations and ecosystems. It uses organization-hosted catalogs, including an ai-catalog.json file at a well-known domain path, plus federated registries that crawl, index, and return capabilities with trust metadata. Clients can query a registry with plain-language intent or fetch a known catalog directly, verify the publisher’s cryptographic identity, then connect at runtime through the capability’s native protocol or API. Google Cloud is bringing ARD to Gemini Enterprise Agent Platform through Agent Registry, which supports hosted discovery and governance features including namespaced URNs, egress policies, specification pinning, and Agent Identity trust-manifest verification. The specification is available under Apache 2.0, with native Agent Platform support planned in the coming months.
A2UI + MCP Apps: Combining the best of declarative and custom agentic UIs
Agentic UIs must balance deep customization with seamless integration: MCP Apps provide standard-web flexibility in an iframe, while A2UI renders declarative JSON through the host application's native components. The post presents three architectural patterns for combining them, including A2UI over MCP, where MCP servers return application/a2ui+json resources for native rendering and can separate static MCP Resources from dynamic Tool calls. This arrangement avoids iframe-related visual fragmentation, keeps backend tool access separate from frontend rendering, and supports rendering across React, Flutter, or Angular through a predefined component catalog. The broader combination also reserves MCP Apps for highly tailored experiences and describes A2UI in MCP Apps, where an embedded app can process A2UI responses and route user actions through an AppBridge; the conclusion recommends choosing among patterns according to project constraints.
Clustering billions of products for agentic commerce with Catalog API
Shopify’s Catalog must reconcile billions of listings whose merchants use different product and variant structures, so AI shopping agents can identify the same real-world product. Product clustering groups related listings under a Universal Product Identifier, with precision prioritized over recall because wrong products can lead to incorrect purchases while missed variants remain findable. The intra-store system starts with rules-based singleton detection, then uses ANN and sparse average linkage for pre-chunking, and applies a two-stage LLM process that proposes clusters and critiques them using the core value proposition framework. Dynamic per-chunk strict JSON schemas enforce complete product coverage, support arbitrary chunking, and improved recall by 8% on the Toloka evaluation dataset in one schema change. The resulting pipeline is presented as a scalable intra-store foundation, while cross-store clustering into global UPIs is identified as the next step.
Agentic Resource Discovery: Let agents search
Agentic Resource Discovery (ARD) is a draft, open specification for discovering agent capabilities at runtime, addressing a model in which users and developers must pre-install, integrate, and maintain each MCP tool, skill, or A2A agent. It defines a static ai-catalog.json manifest and a dynamic POST /search registry API, using indexed signals such as publisher identity, representative queries, compliance attestations, and tags to move capability selection outside the LLM. Hugging Face's Discover Tool adapts Hub semantic search into ARD catalog entries for running Spaces, exposing skills, MCP servers, or raw Space metadata according to the requested media type and wrapping agents.md files into installable SKILL.md artifacts. The tool is available through the hf CLI, REST, and MCP, while planned work includes federation modes and Hub support for profile-level manifests.
Building AI Agents for AR Glasses and XR Devices with NVIDIA XR AI
Building AI experiences for augmented reality and wearable devices requires bridging hardware with live media streams, multimodal models, enterprise tools, and runtime infrastructure. NVIDIA XR AI provides an open-source, modular framework connecting extended reality headsets and smart glasses to GPU-accelerated services across cloud, edge, and workstations. In this architecture, camera frames and microphone audio ingest into an XR Media Hub that routes data while keeping raw video pixels in shared memory to minimize overhead. The ecosystem uses NVIDIA Cosmos models for vision-language grounding, NVIDIA Nemotron models for reasoning and tool invocation, and the Model Context Protocol for enterprise integrations. Optional agent orchestration via NVIDIA NeMo Agent Toolkit and spatial streaming through NVIDIA CloudXR support complex workflows across healthcare and manufacturing.
What is data pipeline architecture?
Data pipeline architecture is the end-to-end blueprint for collecting, processing, storing and delivering data from source systems to people, applications and models. It distinguishes logical design, which defines stages and responsibilities, from physical design, which assigns tools and infrastructure to those stages. The common four layers are ingestion, processing and transformation, storage, and serving and consumption, with orchestration and observability spanning the pipeline. The brief compares batch, streaming, Lambda, Kappa and medallion patterns, explaining trade-offs involving freshness, cost, complexity and operational burden. It also contrasts ETL with ELT and presents governance, monitoring and right-sized processing as reliability practices, concluding that architecture should match the use case and balance freshness, cost and reliability.
Build Your Own Transaction Foundation Model for Financial Intelligence
Production financial intelligence systems frequently rely on hand-engineered tabular features and rule sets that are brittle, expensive to maintain, and blind to historical sequential patterns. NVIDIA demonstrates an accelerated reference pipeline that replaces standard BPE tokenization with a GPU-based domain tokenizer, converting raw transactions into semantic tokens with an 8,192-token context window. Using the NeMo AutoModel library, a compact 29-million parameter decoder-only transformer is pretrained from scratch on unlabeled transaction sequences using causal language modeling. Learned sequence embeddings are extracted, compressed using PCA, and concatenated with raw tabular features to train a downstream GPU-accelerated XGBoost fraud detection model. On the IBM TabFormer benchmark dataset, this combined approach achieves a 41.76% lift in Average Precision over the baseline model relying solely on raw tabular features.
NVIDIA Blackwell Tops MLPerf Training 6.0 with Industry-Leading Scale and Performance
NVIDIA submitted results across all MLPerf Training v6.0 benchmarks, setting performance records on workloads including the DeepSeek-V3 and GPT-OSS-20B Mixture of Experts models. To scale training up to 8,192 Blackwell GPUs, NVIDIA combined hardware cluster designs with scale-out networking via Spectrum-X Ethernet and Quantum InfiniBand. The engineering team eliminated CPU-GPU synchronization bottlenecks in token-dropless MoEs by implementing full-iteration CUDA graphs with synchronization-free operators and paged stashing. Further software optimizations included CuTe DSL kernel fusions, an MXFP8 attention block in the Transformer Engine, and transitioning router elementwise math to FP32. Across these benchmarks, the GB300 NVL72 platform achieved the fastest time to train at scale and the highest normalized per-accelerator performance.
Enabling Governed Vibe Coding for Enterprise Apps on Databricks
Databricks introduces three capabilities intended to bring vibe coding to enterprise applications, where speed alone does not provide business-data context, deployment safety, or cost control. App Spaces lets admins define resource and data access, on-behalf-of-user API scopes, and security policies for groups of apps, with each app inheriting those settings. Genie App Builder turns plain-language descriptions into working internal apps through generated plans, live previews, AppKit, and awareness of workspace data assets and Unity Catalog semantics. Serverless micro apps run in isolated lightweight virtual machines, start quickly when needed, scale to zero while idle, and use usage-based rather than reserved-capacity infrastructure. Together, the capabilities are presented as a way for business-proximate users to build on enterprise data while organizations apply consistent governance and support broader app portfolios; all three are coming to Databricks Apps, with private previews coming soon.
Introducing OpenSharing: the Next Evolution of Delta Sharing for the Agentic Era
OpenSharing is presented as the next evolution of Delta Sharing, extending an open zero-copy data-sharing protocol from tables and files to models, agents, semantic context, unstructured data, and reusable AI logic. The protocol is now an independent open-source project hosted by the Linux Foundation, while Databricks OpenSharing adds Unity Catalog governance and audit logging, Marketplace discoverability, and enterprise features. Genie Agent Sharing supports governed AI experiences across organizational boundaries, with controls for proprietary instructions, data access, daily prompt quotas, and row exports. SecureConnect removes per-recipient firewall changes through a Databricks-managed proxy, while Global Distribution uses local replicas to reduce egress fees and latency. The launch also supports Apache Iceberg REST Catalog API, external catalogs, and on-premises storage partners; providers define shares in Unity Catalog, recipients query live data through existing tools, and governance enforces access controls.
Announcing Apps on Databricks Marketplace
Databricks announces the Public Preview of Apps on Databricks Marketplace, allowing customers to discover, install, and run third-party data and AI applications inside secure Databricks workspaces. The offering addresses procurement challenges involving data movement, lengthy security reviews, custom integrations, and fragmented identity management by bringing applications to the customer’s data. Installed apps run in isolated sandboxes within the consumer’s Databricks account, inherit Unity Catalog governance, and use dedicated serverless compute with consumer-controlled external access through Serverless Egress Gateway policies. Providers can publish closed-source containerized apps once for no-egress distribution without maintaining per-customer infrastructure, while applications connect natively to services including SQL Warehouse, Lakebase, Model Serving, and Foundation Model APIs. The Public Preview launches with 20 partners, and planned additions include bundled assets, provider analytics, and commercial monetization.
Introducing OpenSharing SecureConnect
OpenSharing SecureConnect addresses the networking burden of sharing live data from provider storage behind private networks, where providers and recipients otherwise exchange firewall and egress details manually. It is a Databricks-managed proxy that routes recipient storage access through Databricks endpoints after a one-time provider setup, while the data remains in the provider’s bucket. Providers allowlist Databricks Serverless Data Plane endpoints and enable SecureConnect for a metastore; serverless recipients require no configuration, while classic and open recipients allowlist stable inbound IPs. Optional NCC provides private link connectivity, and mutual TLS is available for recipients. SecureConnect is in Public Preview and supports cross-region, cross-cloud sharing plus customer-managed and Databricks Default Storage.
Build On-Device AI Companions with the NVIDIA ACE Game Agent SDK and Unreal Engine 5 Plugins
Game developers integrating on-device artificial intelligence companions face challenges such as conversational latency, state synchronization, and loop prevention. NVIDIA announced the open-source ACE Game Agent SDK alongside a suite of Unreal Engine 5 plugins to streamline native, hardware-accelerated character pipelines. The SDK offers Agent, Chat, and RAG APIs, enabling characters to execute multi-step tool-assisted reasoning and query game databases. Unreal Engine 5 plugins introduce local runtime models including nemo-conformer-ctc-120m for speech recognition, Qwen 3.5 4B for language generation, and Chatterbox Turbo 350M for speech synthesis. Additional tooling expands to motion generation via Animotive Kimodo and rendering updates with the DLSS 4.5 Unreal Engine plugin.
How to Optimize Transformer-Based Models for Low-Precision Training
Accelerating transformer training requires understanding how low-precision formats such as FP8 and NVFP4 impact specific general matrix multiplication (GEMM) workloads. Transformer configurations do not explicitly reveal these shapes, making microbenchmarks necessary before committing to full training runs. NVIDIA Transformer Engine enables quantization and kernel dispatch across precisions, which can be evaluated in realistic autocast mode or kernel-only prequantized mode. Profiling the ESM2-15B model on NVIDIA B300 GPUs demonstrated that NVFP4 achieved a 1.79x blended forward propagation speedup over MXFP8 and up to 4.01x over BF16 in prequantized execution. Although large GEMM dimensions successfully overcome quantization overheads, dynamic scaling, Hadamard transforms, and kernel selection asymmetries moderate real-world gains.
Cart Assistant: Agentic Grocery Shopping on Uber Eats
Cart Assistant addresses grocery requests that begin as lists, recipe images, or vague plans and turns them into draft carts shoppers can review and edit. It uses a multi-prompt state graph that assigns planning, candidate retrieval, semantic relevance, constraint enforcement, quantity selection, guardrails, assembly, and content refinement to focused stages. LLMs handle ambiguity and structured outputs, while backend APIs, deterministic arithmetic, validation, optimization, and aggregation manage inventory, pricing, eligibility, sellable units, and cart construction. The design separates retrieval language from reasoning context, preserves quantity provenance, supports parallel item processing, and uses deterministic checks plus an LLM judge against synthetic and production-derived cases. According to the conclusion, this decomposition makes the system more controllable and inspectable while producing useful draft carts without removing shopper review, editing, and decision-making.
Top topics
- Architecture834
- Performance816
- LLMs660
- Machine Learning622
- Open Source563
- AI Agents553
- AI470
- Developer Experience448
Weekly digest
The best engineering reads, once a week.