英文科技媒体分类 feed · InfoQ · AI, ML & Data Engineering

Google has open-sourced AQuA, a reference implementation that analyzes production AI agent sessions, verifies recurring failure patterns, and investigates root causes against deployed source snapshots. In a 32-session replay, developers reduced seat selection bypasses from 15 sessions to two and dietary preference omissions from seven to zero after prompt changes. By Leela Kumili

Netflix is developing an ontology-driven knowledge graph to accelerate production investigations and provide AI agents with a shared understanding of its operational environment. By Craig Risi

Hannes Hapke shares how Dataiku’s open-source Kiji Inspector applies Sparse Autoencoders (SAEs) directly to decision tokens to expose why agents pick specific tools. Moving past superficial chain-of-thought narratives, he discusses causal ablation, layer activation extraction via patched vLLM engines, and production monitoring workflows for software architects and senior engineering leaders. By Hannes Hapke

Microsoft has released a public preview of the logicapp extension for the Azure CLI, adding local project creation, workflow authoring, a runtime lifecycle, connector management, run inspection, and test scaffolding for Logic Apps Standard, work that previously required the VS Code extension. The product team pitches the terminal surface as determinism for AI-assisted development. By Steef-Jan Wiggers

Google Cloud has open-sourced TPU Sync, a native C++ library that moves KV cache blocks and model weights between TPU HBM, host DRAM, shared storage, and the data center network. By Daniel Dominguez

JioHotstar evaluated 6 approaches to code retrieval for AI coding agents using SWE, CodeScale Bench, and an internal question-answering harness. Structural search performed well on concrete tasks, while semantic retrieval helped with ambiguous requirements. The findings highlight trade-offs in accuracy, token use, and tool design, underscoring the need to match retrieval tools to task types. By Leela Kumili

QCon AI New York has published its full program for December 15–16, 2026. Engineers from LinkedIn, Netflix, DoorDash, Figma, Pinterest, Honeycomb, Harvey, Google DeepMind and OpenAI describe AI systems they run in production. Talks cover agent permissions and guardrails, eval loops, inference cost, AI code review, and team adoption. By Artenisa Chatziou

Cloudflare has announced the general availability of Basin, a serverless analytics platform built on Apache Iceberg and R2 Object Storage. The managed service supports open data formats and integrates with tools such as DuckDB, Snowflake, and Apache Spark, with no data egress fees. By Renato Losio

In a recent article, Google senior staff software engineer Sam Saccone and senior staff data scientist Madhuri Garikapati highlighted the cost of agent distraction, where AI coding agents get derailed by stale technical debt, such as old TODO comments, obsolete bug references, and outdated specs, leading them to embark on unrequested "side-quests". By Sergio De Simone

Aditya Mulik explains how Walmart Global Tech deployed multi-agent LLM workflows across 4,700 stores for nightly batch inventory forecasting. He discusses tackling production hurdles - schema violations, cross-department data mismatches, and network drops - via MCP grounding and loop agents, while slashing cloud inference costs from $100k+ to $10k per month shifting from PTUs to pay-as-you-go. By Aditya Mulik

Verification layers sort outputs by trust rather than fixing errors. In a benchmark of six reliability layers, voting and self-correction failed to recover accuracy - they only withheld doubtful records. Production systems gain more reliability by replacing generation with deterministic lookups, tracking yield over raw accuracy, and resolving extraction bottlenecks before stacking checks. By Pushpdeep Mishra, Wejdan Bagais

Google has released Gemini 4 Argon framing it as a general-purpose physical-reasoning model for robot control, Real2Sim scene construction, multimodal labeling of robot video, and iterative policy improvement, rather than as a single end-effector policy. By Daniel Dominguez

Cloudflare is developing a public certificate authority for Merkle Tree Certificates (MTCs) to reduce the bandwidth overhead of post-quantum cryptography. A Chrome experiment served billions of MTCs to 50% of Chrome Beta 146 for selected domains, reporting 9% faster median performance. Wider deployment depends on browser support, transparency, and production-scale validation. By Leela Kumili

At the recent PyTorch Conference North America 2026, Amazon introduced TorchNeuron, a native PyTorch backend that enables developers to run existing PyTorch code on AWS Trainium AI chips with virtually no code modifications. By Sergio De Simone

In this episode, Edoardo Vacchi discusses the evolving landscape of WebAssembly on the backend, running Java workloads on GPUs using TornadoVM, and optimizing distributed AI inference using llm-d and vLLM on Kubernetes. Edoardo explains how llm-d manages cache locality (KV and prefix caches), request routing, and batch inference across Kubernetes clusters to maximize expensive GPU utilization. By Edoardo Vacchi