MLOps news and analysis on Web Pulse: 3397 curated articles tracking the latest MLOps developments, tools, and research, updated continuously from vetted sources.
A developer found that seven of eight custom agents defined in Claude Code had zero invocations over a 30-day period, despite being injected into the system prompt on every request. The unused agent definitions consumed …
Modelplane released version 0.4, adding an NVIDIA Dynamo serving stack that brings gang scheduling and peer-to-peer weight transfer to its inference clusters. The release, built with NVIDIA's Dynamo team, also pulls serv…
Google's Antigravity CLI, the successor to Gemini CLI, lacks a built-in OpenTelemetry exporter, prompting a guide to monitor its agent loop using SigNoz and a custom hook script that converts tool completion, loop passes…
Inference engineering techniques for large language models (LLMs) can be categorized into those that manage tradeoffs along an efficient frontier and those that push the entire frontier outward, according to an article o…
OpenRouter's tracking dashboard shows the deepseek/deepseek-v4-flash endpoint at atlas-cloud/fp4 recorded a B3IT total-variation level shift of 0.55, while its LT logprob tracking remains impossible because the endpoint …
The z-ai/glm-5.3 endpoint on Modal was retired from B3IT monitoring on 2026-09-08 after the tracker could not find enough border inputs, while its LT logprob probe also failed to return usable logprobs. OpenRouter billed…
A developer detailed how their batch runner for an AI content pipeline, despite having resumability built in from day one, still failed silently for three months due to four distinct bugs. The fixes, which landed in a si…
A new approach treats AI system prompts as deployable artifacts, applying release engineering practices such as versioning, canary deploys, and rollback to prompt changes. The proposal, outlined in a technical essay, inc…
A developer released merge-proof, an open-source Git/GitHub Action tool that detects when a pull request's green CI does not prove the final merged state was validated against the base it merged into. The tool, which run…
Databricks engineers used Unity Gateway tracing and Genie One to identify and fix seven MCP-server bugs in one hour, eliminating an estimated $1.2 million a year in wasted AI agent spend, including $499K in wasted tokens…
A developer benchmarked three Gemma 4 deployments on a single AWS T4G GPU for under $3, comparing vLLM, JAX, and PyTorch runtimes. The project built a common harness to measure decode throughput, revealing that previous …
A new report from nlocoding.com details how development teams can integrate AI coding tools into Agile workflows, citing 2026 data showing 97% of developers using AI assistants report faster delivery but only 41% see mor…
AI engineering is shifting from prompt optimization to building deterministic systems with observability and local-first architectures, according to a technical article. The piece advocates for treating LLM calls like mi…
Shaide, a Kubernetes-native AI platform for distributed multi-model inference, has been released under the Apache 2.0 license. It uses vLLM for inference, llm-d for orchestration, KV-cache-aware scheduling, and an intern…
Cisco ThousandEyes built OdysseyCD, an internal continuous delivery platform on Kargo, to replace manual release cycles with a centralized, policy-driven promotion model, now managing 137,000+ release gate evaluations ev…
Selfship.ai has launched as a SaaS platform that autonomously monitors AI agent traces, clusters failures by user intent, and ships fixes as pull requests, re-evaluating traces post-merge to confirm regressions are resol…
A developer's unattended automation stack of Windows Scheduled Tasks running Claude Code CLI sessions went offline for five days after an AI agent edited a shared PowerShell script and introduced a Unicode em dash in a c…
Arize AX now includes a managed cost-control agent that runs inside its platform to autonomously diagnose and fix expensive LLM calls in agentic applications, addressing the non-linear cost scaling that occurs as context…
Axem co-founders open sourced Shaide, a Kubernetes-native AI platform for distributed multi-model inference, under the Apache 2.0 license. The platform uses vLLM for inference and llm-d for orchestration, supports indepe…
A developer from MonkeyCode argues that the real cost of AI agent runs is debugging time, not tokens, and advocates for a trace-loop approach to log and diff tool calls. They provide a minimal trace collector and diff sc…