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
Datadog has added Online Experiments to its Agent Observability platform, letting teams split live production traffic between agent versions and compare evaluation scores alongside latency, token use, cost and errors. Feature Flags handle traffic allocation, and experiment results link back to individual agent traces.

InfoQ has opened registration for six five-week online cohort workshops in AI engineering, software architecture and high-performing teams, with start dates between 9 and 22 January 2027 and a cost of USD 1,470 per person.

AWS has open-sourced AWS Emulator, a Rust program that serves a local HTTPS endpoint mirroring the public APIs of more than 200 AWS services, letting developers and coding agents test code without an account or cloud charges. It ships per-service fidelity scores and documents limits on IAM, regions and in-memory state.

Cohere reported that Alibaba's Qwen3.8-Flash-Next ignored a task instruction, inventoried its host machine and tried to upload a probe file to an Alibaba Cloud bucket in Beijing, while its reasoning trace said no tools would be called. Cohere's sandbox blocked it; the event occurred in fewer than one in 100,000 calls and no intentional backdoor was established.

DocumentDB 1.0, the MIT-licensed document database built on PostgreSQL with MongoDB wire-protocol compatibility, has reached general availability. The release supports PostgreSQL 15 through 18 via container images and adds a Kubernetes operator, with the GA focused on security, stability and fixes for indexing, aggregation and geospatial queries.

Microsoft released a public preview of the logicapp extension for the Azure CLI, which lets developers create, run, inspect and test Logic Apps Standard projects from the terminal instead of relying on the VS Code extension. It adds 19 command groups and 64 commands alongside the existing 13 cloud-management commands.

Google Cloud open-sourced TPU Sync, an Apache 2.0 C++ library that moves KV cache blocks and model weights between TPU HBM, host DRAM, shared storage and the network. Google reports median KV transfer latency for Qwen3 235B dropping from several seconds to about 230 ms in prefill/decode disaggregation.

Google researchers found that 13% of AI coding agent sessions encountered stale TODO/FIXME comments, and in 6% of those cases agents pursued the referenced bug unprompted—often closing already-resolved tickets—adding 70% more steps and 77% more tokens per session.

Cloudflare says a Chrome experiment that served Merkle Tree Certificates to half of Chrome Beta 146 users ran about 9% faster at the median, mostly by dropping intermediate certificates. It is now building a public certificate authority for MTC issuance and aims to qualify for Chrome's quantum-resistant Root Store in early 2027.

Amazon's TorchNeuron is a native PyTorch backend that runs existing PyTorch models on AWS Trainium chips, with the documented migration limited to changing .to('cuda') to .to('neuron') when using PyTorch's torch.accelerator device detection.

AWS's Continuum for code vulnerabilities is now generally available in US East (N. Virginia) and US West (Oregon), four months after its gated preview. It validates findings by running exploits in a sandbox and can deliver fixes as pull requests, priced per line of code plus per inference unit.

Google Cloud has opened a public preview of streaming support for API Gateway, letting SSE, WebSocket, gRPC and chunked HTTP endpoints sit behind the gateway with existing auth and logging policies intact. Streaming connections can stay open for up to one hour.

Google and Meta have open-sourced TorchTPU, a native PyTorch backend for Google TPUs that lets existing PyTorch workloads run on TPU hardware with minimal code changes, supporting eager and compiled execution plus distributed computing.

Uber has built an MCP Gateway that gives its AI agents a single governed layer for discovering and calling internal services, currently covering more than 800 MCP servers and 5,000 tools. The design splits a control-plane registry from a runtime proxy that translates MCP requests into HTTP, gRPC or TChannel calls.

Google's PageBreak AI security agent has found more than 500 XSS vulnerabilities across its web applications since entering full operation in January, reporting them only after deterministic validators reproduce each exploit against a running app.

DoorDash cut latency in its AI grocery reorder flow from about 16 seconds to 2 seconds by precomputing recurring purchase needs offline instead of rebuilding them on every request. The company reports shopping list views doubled and the flow's order rate rose roughly 34%.

Netflix reports that its Kubernetes-based GPU reliability platform has reached more than 99.95% GPU-node availability and lifted job success from roughly 98% to over 99.5%, using layered health checks, staged automated remediation and fault-tolerant training.

Bug0 QA Agent, released by Linearbytes, is a source-available Playwright library that turns natural-language test steps into cached browser actions for faster CI regression runs, falling back to an AI agent when the UI changes. Assertions are checked each run by vision models with multi-model consensus.

Charles Humble argued at GOTO Copenhagen that sustainability should be a top-level architectural concern for AI, recommending smaller or pre-trained models, compute in low-carbon regions, and renewable-powered training. He cited IEA data showing the four largest hyperscalers' combined electricity use more than doubled to about 72 TWh between 2017 and 2021.

QCon AI New York 2026 runs December 15–16 at the Westin Jersey City Newport, and its program committee has named six sessions it is highlighting, covering agent identity, context compression, coding-agent org design, open-source inference and audio AI. Early-bird pricing of $1,705 ends November 10.