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
Grafana Labs proposes using evaluations, service-level objectives (SLOs) and error budgets to measure whether AI agents behave acceptably, turning hallucination, task fulfilment, toxicity, cost and prompt injection into monitorable targets. An example SLO requires 95% of conversations judged fulfilled over 30 days, with the remaining 5% as error budget.

AWS has open-sourced Strands Decider 2B, a two-billion-parameter model that picks between predefined options with confidence scores, alongside Strands Box, a developer-preview kernel sandbox that enforces execution policies outside the model.

Microsoft will scale its Maia accelerator family to multi-gigawatt deployments from 2027, making Maia the primary inference system for Microsoft products. Maia 300 puts scale-up and scale-out on one Ethernet fabric and moves KV state directly between prefill and decode workers, but practitioners questioned the tokens-per-watt efficiency claim.

Meta is extending its Private Processing architecture, previously used for WhatsApp and Meta AI, to AI glasses so cloud AI workloads and persistent personal context can be processed inside confidential virtual machines. The system is not yet deployed on the glasses and no timeline has been given.

Google added local model support to its Antigravity SDK, letting developers run agentic workflows on their own machines while keeping cloud models available for planning. A demo of auditing three vulnerable Python modules used about 95 of 3,400 tokens in the cloud.

QCon AI New York 2026, running December 15-16 at The Westin Jersey City Newport, has confirmed sessions on production AI from engineers at AWS, LinkedIn, DoorDash, Netflix and Google Cloud. Topics include agent identity and authorization, guardrails, eval loops, LLM cost control and AI-assisted code review.

Spotify reports that AI-assisted development more than doubled its monthly merged code changes year over year, from roughly 8,100 to 17,000 in August, forcing changes to testing, observability and rollback capacity. The company says incident reviews found no material direct contribution from AI-authored code.

Researchers from Meta, MIT, and the University of Washington propose Context Language Models (CLMs), which let an LLM treat its own context as an editable file instead of relying on fixed summarization or retrieval pipelines, reporting accuracy gains alongside lower compute on several benchmarks.

Cloudflare Traces is now in open beta, automatically turning security rules, transforms, cache, routing and origin handling into OpenTelemetry spans without instrumentation. From December 1, tracing pricing moves to ingestion and retention volume, with 50GB ingestion and 10GB-month storage included on paid plans.

Google released Android Bench 2.0, adding long-horizon tasks, agentic evaluation, and continuous scoring to its benchmark for AI models and agents on Android development work. The top reported score is a 32% long-horizon task pass rate for Claude Opus 5.5.

GitHub migrated over 800,000 lines of Copilot runtime code from TypeScript and Node.js to Rust in about 14.5 weeks using AI-assisted development, delivering it through 128 pull requests. A measured startup and single-turn scenario dropped from 5.25 seconds to 292 milliseconds.

Cloudflare released Clef, open-weight 9B and 27B multimodal decision models that return probabilities for predefined options instead of generating text, plus a fine-tuning service. The models are available via Workers AI and Hugging Face.

A Coleman Parkes survey for Undo of 300 senior engineering leaders found AI coding agents have shifted the main bottleneck to debugging and code comprehension: teams spend 16.9 hours weekly debugging versus 9.8 hours writing code, and 79% say the overall release cycle is no faster.

InfoQ is hosting a free 60-minute live panel on October 14, 2026, with five practitioners covering agent autonomy, verification of AI-generated changes, sensitive-data exposure, and production RAG. Registration includes access to the recording and the chance to submit questions in advance.

Cloudflare ran frontier AI models inside a controlled test harness to probe its Web Application Firewall, generating 1,107 mutation attempts across 45 scenarios. Human triage left 49 findings, 48 involving command injection or SSRF, and the work led to three Managed Ruleset changes.

Google has made Spanner Omni generally available, letting its distributed SQL database run in customer data centers, on other clouds or on a laptop by replacing Colossus and TrueTime with software. No availability SLA is offered, and Google's documentation lists BigQuery, Knowledge Catalog and Gemini Enterprise as excluded integrations.

QCon London 2027 has announced a 15-track program for its April 13–16 conference, covering production AI, architecture, distributed systems, data platforms, performance, and Staff+ leadership. More than 75 practitioner speakers will present production systems and trade-offs.

Akka used 65 open source projects to test a spec-driven workflow for AI-assisted software porting, measuring specification structure, context, model and effort selection, automated validation, token consumption and runtime performance. The first tranche took 99.3 hours and 9.41 billion tokens, with reported lines-of-code or performance improvement in 57 of 65 ports.

A team of developers working at AWS has open-sourced Pizza Bot, a self-hosted Apache 2.0 application that runs AI agents in the background and returns results through an inbox-style interface. It supports scheduled and webhook-triggered tasks, delegation to specialist workers, and human approval steps.

Archestra released OpenAPPA, an open-source security engine that runs outside an agent's prompt and execution loop. The team reports a 0% attack success rate on the Bench-Corp and AgentThreatBench benchmarks, versus 10% for Claude Code's auto mode and 31% for Microsoft FIDES.