AI news for builders and product teamsUpdated Nov 17, 2026, 13:01 UTC
InfoQ · AI, ML & Data Engineering
Reporting and perspectives from InfoQ · AI, ML & Data Engineering. Headlines and excerpts link to the original articles.
Latest stories
Newest first
Airbnb extended its open-source Chronon feature platform with Push Mode and Near Real-Time Model Transform, cutting recommendation embedding staleness from nearly two days to under a minute, typically 10 to 30 seconds. Offline NDCG improved 1.67% and online A/B testing showed roughly a 0.33% increase in uncancelled bookings.

Uber has introduced Normalized Compute Units in its Odin stateful orchestration platform, assigning each server type a performance factor so workloads get comparable compute capacity rather than an equal number of cores. On roughly 1.4 million CPU cores of automatically scaled Odin workloads, Kafka CPU allocation fell about 8%.

Atlassian's State of Developer Experience 2026 reports that the share of developers saving ten or more hours a week with AI fell from 68% to 26% in a year, even as 95% of developers say their organizations actively use AI.

The Agentic AI Foundation grew from three founding projects to ten in its first year, adding OpenSandbox from Alibaba, Open Responses from OpenAI, Agent Skills from Anthropic and OpenEnv from Hugging Face at AGNTCon + MCPCon North America 2026 in San Jose. Much of the October 22-23 program shifted from new protocol features to security and agent identity.

DigitalOcean has put its Agent Learning Services — Insights, Signals and Simulations — into public preview, while Inference Router optimization and Agent Data Services remain in private preview. The services aim to let agents learn from real production sessions and be tested before changes ship.

InfoQ is running three online certification cohorts in November 2026: two AI engineering programs (led by Adi Polak on November 12 PST and Lan Chu on November 13 CET) and a High-Performing Teams program with Olimpiu Pop on November 17.

Cloudflare released Auto Router in public beta through AI Gateway, where a classifier scores each request on complexity, ambiguity, stakes and context dependence and picks a model. Its own benchmark shows ten points lower success than Claude Opus 5.5 at roughly a third of the cost.

Google and Meta open-sourced TorchTPU, a native PyTorch backend for Google TPUs, announced at PyTorch Conference North America 2026 in San Jose; Meta says it plans to move larger LLM workloads onto TPUs with it.

Google has open-sourced ML Drift, an Apache 2.0 cross-platform GPU compute engine that backs LiteRT and abstracts OpenGL ES, OpenCL, Metal and WebGPU. Google reports up to 40% lower frame latency in YouTube Shorts and 12% lower memory overhead than llama.cpp and MLX LM in Gemma benchmarks.

An unattended 11-hour run of Gemini 4 Argon produced two accepted performance optimizations to Google's Agent Development Kit, a roughly 200,000-line Python codebase, cutting in-memory session-preparation overhead by about 40% and cold package import time by 21%. Four other proposed changes were rejected at developer review.

Google has released Data Agent Kit as generally available: a free, Apache 2.0-licensed set of MCP tools and 37 skills that connect third-party coding agents to more than fifteen Google Data Cloud services. Its guardrails come in two forms — advisory skills like accidental-data-loss-prevention, and IAM deny policies that actually block tool calls.

Anthropic launched OSS Scanner, a free opt-in service that uses its strongest models, including Claude Mythos, to generate vulnerability reports and candidate patches for open-source projects, delivered to maintainers without human review.

A Harvard study of 718 firms using Jellyfish data found AI coding tools raise code volume but not completed work proportionally, with code review emerging as a key bottleneck. Agents lifted lines of code 30% but resolved Jira issues only 12%, while review time per pull request rose 49%.

Vercel Labs has open-sourced fx, an experimental coding agent written in Zig that ships as a 6.3 MiB native binary and can be used as a CLI, invoked from scripts, or embedded via a libfx library with a WebAssembly build.

Brandon Thomas used Claude Opus 5.5 to build open-source Rust replacements for seven Adobe apps, then told Hacker News the project is not close to ready, contradicting Reddit posts that projected near-total parity within a month.

TornadoVM 7.3 replaces JNI with the Foreign Function & Memory API and drops the experimental JVMCI dependency, giving one compiler-metadata path across JDK 21, 25, 26 and 27. The release adds an optimized CUDA backend with cuTile and cuBLAS/CUTLASS calls from TaskGraphs, plus multi-GPU execution.

Google released EmbeddingGemma 2 under the Apache 2.0 license, a multimodal embedding model based on Gemma 4 that maps text, code, images, video and audio into one shared 768-dimensional space for on-device search and RAG.

Reverse Engineer Anything (REA), an MIT-licensed open-source project with over 200K GitHub stars, links coding agents like Claude Code and Codex to disassemblers such as Ghidra, Hopper and IDA through MCP servers so agents can inspect binaries and test hypotheses.

Google researchers have open sourced EnvHarness, a framework that adapts agent-training environments to an agent's weaknesses without changing the agent, the environment or the evaluation system. In tests across WebArena, SWE-bench Verified and ALFWorld, several models gained 2.9 to 3.7 percentage points.

An InfoQ webinar on running agentic systems in production featured practitioners describing silent failures such as semantic and embedding drift, and recommending automated review, continuous evaluation, and per-agent access controls. The panel framed deployment approval as a strategic business decision rather than a purely technical one.