MLOps news and analysis on Web Pulse: 3339 curated articles tracking the latest MLOps developments, tools, and research, updated continuously from vetted sources.
Slack's engineering team published findings from over 200 automated agentic test runs comparing three architectures: an agent driving Playwright via Model Context Protocol, an agent issuing Playwright CLI commands, and L…
Honeycomb launched Canvas Connectors in general availability, offering 12 connectors including GitHub, PagerDuty, Jira and Amplitude that let the Canvas AI agent read from and act in systems outside Honeycomb's telemetry…
Honeycomb announced early access to AI Ecosystem, an AI agent fleet-level analysis layer built on its context-rich data model that covers AI Agent Fleet Performance, LLM Cost Tracking, and Agent Conversations with one-cl…
A developer built AcquireIQ, an M&A intelligence terminal that ingests unstructured data rooms, code repositories and diligence documents into a persistent vector memory layer for autonomous deal-flow analysis. The syste…
A developer's guide outlines LLMOps practices for moving generative AI applications from prototype to production, covering prompt versioning, evaluation sets, monitoring, cost control, and security. It recommends storing…
TensorFold 0.3.6.2 delivered decode speeds over 62 tokens per second on a single stream and 119 tokens per second across five concurrent streams running Qwen3.8-Flash-Next on a single Nvidia DGX Spark, according to bench…
A developer built MemoryOps, an AI incident response agent that pairs an LLM with the Hindsight memory layer so it can recall similar past incidents and retain new resolutions for future investigations. The system separa…
A technical guide released this week by developer and researcher NO1ennn argues that production AI agents waste most of their spend routing frontier models like Claude and GPT-4o at $15 to $18 per million input tokens on…
A developer built WorkMemory AI, an incident-response and engineering-memory platform that records software incidents and connects new problems to relevant past investigations so teams can reuse prior fixes. The system u…
A developer has released MCP-USE, a full-stack framework that provides both client and server SDKs for building deployable agent applications on top of the Model Context Protocol. The framework adds TypeScript client orc…
A developer outlines a RAG evaluation harness that separates retrieval failures from generation failures, using a dataset of question/answer/context triplets and four core metrics. The approach uses an LLM-as-judge to sc…
Microsoft reported that starting at 10:00 UTC on 29 September 2026, a subset of customers using Azure OpenAI Service, Foundry Agent Service, Foundry Models, and Cognitive Services in Sweden Central experienced intermitte…
Cloudflare introduced Worker Previews, a feature that gives each Git branch its own isolated, production-like environment for a Worker with a stable URL, per-branch configuration and state, and scoped observability, crea…
Pydantic AI 2.50.0 adds support for running agents on Jev, the typed-decision model from TypeSafe, via a new TypeSafeModel adapter and a DecisionModel base class, with a decide span added under the chat span in Logfire o…
A developer outlined a two-stage on-device retrieval-augmented generation pipeline for iPhone that pairs an INT8-quantized MiniLM-L6 bi-encoder for approximate retrieval with a CoreML cross-encoder (MiniLM-L12) that resc…
Rsync.ai launched as a self-hosted, source-available data platform under the Elastic License 2.0, offering batch and change-data-capture pipelines, scheduled SQL models, and table-level lineage with 21 connectors. The to…
A new arXiv paper (2609.31857) finds that at 5% pairwise human-label overlap, LLM-judge validation can make wrong deployment decisions 25% of the time and select the wrong best judge among ten candidates 65% of the time.…
A developer built PipelineSage, a Python and Streamlit agent that diagnoses CI/CD pipeline failures by recalling similar past incidents from the Hindsight agent memory system, re-ranking them, and grounding an LLM diagno…
SimplyBlock has shipped business continuity and disaster recovery functions for Red Hat OpenShift deployments, targeting VMware-to-OpenShift migrations with storage-layer replication built directly into its platform. The…
UniServe, a serving engine for FastH3 8-Step text-to-video-with-audio generation from the hao-ai-lab, delivers lower median end-to-end latency and 20–46% higher throughput than FastVideo, vLLM-Omni and SGLang across ever…