{"week":"2026-W42","start":"2026-10-10","end":"2026-10-16","title":"What happened in AI — Oct 10–16, 2026","generated_at":"2026-10-16T21:16:30Z","intro":["Agent builders' bottleneck moved from generating code to trusting and reviewing it. A study found AI coding agents produce more code but not more software, and Spotify and Anthropic both report merge volume outrunning their review and source-control systems.","The response is engineering discipline around the model: LangChain's harness work lifted its coding agent from Top 30 to Top 5 on Terminal Bench, Grafana proposes SLOs for agent behavior, and AWS frames autonomy as five layers of trust.","Open-weight models are keeping pace: LangChain's evals show GLM-5 and MiniMax M2.7 matching closed models on core agent tasks, and Reflection's Beam had day-0 serving on vLLM and AWS."],"highlights":["AI coding agents raise code volume, but human review absorbs the gain.","LangChain harness engineering took its coding agent from Top 30 to Top 5 on Terminal Bench.","Open models GLM-5 and MiniMax M2.7 match closed frontier models on core agent tasks.","vLLM shipped day-0 Reflection Beam (501B total, 23B active) and a 3.3× TTFT cut for multi-turn KV reuse.","Anthropic reports quarterly revenue up nearly 50% to $16.8bn; Modal raised $750M at $1B run rate."],"article_count":34,"categories":[{"name":"Open Models & the Serving Stack","slug":"open-models-serving","summary":"Open-weight agent models now ship with day-0 serving, and KV-cache transfer and local execution are the levers on latency and cost.","articles":[{"title":"Introducing Reflection Beam: an open-weight model for agentic coding on AWS","summary":"Reflection's open-weight Beam coding model can be deployed on AWS via a SageMaker JumpStart real-time endpoint.","source":"aws_ml_blog","url":"https://aws.amazon.com/blogs/machine-learning/introducing-reflection-beam-an-open-weight-model-for-agentic-coding-on-aws/","published":"Fri, 16 Oct 2026 15:23:10 +0000"},{"title":"Serving Reflection Beam on vLLM from Day 0","summary":"vLLM serves Beam-1.0 (501B total, 23B active) from day 0; agent-driven tuning cut decode latency 2.71×.","source":"vllm_blog","url":"https://vllm.ai/blog/2026-10-16-reflection-beam","published":"Fri, 16 Oct 2026 00:00:00 GMT"},{"title":"Scaling Kimi K3 on AMD Instinct MI355X: Matching NVIDIA HGX B300 Throughput per GPU on AgentX","summary":"vLLM write-up on Kimi K3 on AMD MI355X, aiming to match NVIDIA HGX B300 throughput per GPU.","source":"vllm_blog","url":"https://vllm.ai/blog/2026-10-16-kimi-k3-mi355x","published":"Fri, 16 Oct 2026 00:00:00 GMT"},{"title":"Announcing Day 0 Support for NVIDIA Nemotron 3.5 Super VL on vLLM","summary":"vLLM ships day-0 Nemotron 3.5 Super VL serving with OpenAI-compatible APIs and multi-token prediction.","source":"vllm_blog","url":"https://vllm.ai/blog/2026-10-15-nemotron-3-5-super-vl-vllm","published":"Thu, 15 Oct 2026 00:00:00 GMT"},{"title":"Open Models have crossed a threshold","summary":"LangChain evals: GLM-5 and MiniMax M2.7 match closed frontier models on file ops, tool use and instruction following at lower cost and latency.","source":"langchain_blog","url":"https://www.langchain.com/blog/open-models-have-crossed-a-threshold","published":"Tue, 13 Oct 2026 19:56:07 GMT"},{"title":"Push-Based KV Transfer to Improve TTFT and TPOT for Disaggregated Inference in vLLM","summary":"A push-based NIXL connector writes prefill KV straight into decode memory over RDMA, taking the router relay off the TTFT critical path.","source":"vllm_blog","url":"https://vllm.ai/blog/2026-10-14-nixl-push-connector","published":"Wed, 14 Oct 2026 00:00:00 GMT"},{"title":"Improve TTFT for Multi-Turn Agentic Workloads with Bi-Directional KV Transfer in vLLM","summary":"Bi-directional KV transfer lets prefill nodes reuse earlier-turn KV, cutting TTFT up to 3.3× on long multi-turn workloads.","source":"vllm_blog","url":"https://vllm.ai/blog/2026-10-12-bidirectional-kvxfer-multiturn-agentic-workload","published":"Mon, 12 Oct 2026 00:00:00 GMT"},{"title":"Google Antigravity SDK Adds Local Model Support for Greater Token Efficiency and Code Privacy","summary":"Antigravity SDK now runs agentic workflows on local models, with optional cloud models mixed in, for token savings and code privacy.","source":"infoq_ai_ml","url":"https://www.infoq.com/news/2026/10/antigravity-sdk-local-models/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering","published":"Wed, 14 Oct 2026 09:00:00 GMT"}]},{"name":"Harness Engineering & Evals","slug":"harness-engineering-evals","summary":"Agent quality is increasingly won in the harness: verification, context management and measurable behavior targets.","articles":[{"title":"Improving Deep Agents with harness engineering","summary":"Self-verification, tracing and context tuning moved LangChain's coding agent from Top 30 to Top 5 on Terminal Bench.","source":"langchain_blog","url":"https://www.langchain.com/blog/improving-deep-agents-with-harness-engineering","published":"Tue, 13 Oct 2026 16:04:23 GMT"},{"title":"Context Engineering with Deep Agents: How To Remember Without Keeping Everything In Context","summary":"Offloading, summarization, skills, subagents and memory keep long-running agent context focused instead of piling up.","source":"langchain_blog","url":"https://www.langchain.com/blog/context-engineering-deep-agents-remember","published":"2026-10-14T17:00:00+00:00"},{"title":"The Most Common Questions About Agent Evals, Answered","summary":"Practical answers on offline vs online evals, LLM judges, multi-turn evals and keeping eval costs down.","source":"langchain_blog","url":"https://www.langchain.com/blog/common-questions-about-agent-evals","published":"2026-10-15T19:00:00+00:00"},{"title":"Context Language Models: Self-Managing Context to Improve Performance and Reduce Compute Costs","summary":"Meta, MIT and UW researchers propose Context Language Models that edit their own context instead of using fixed rules.","source":"infoq_ai_ml","url":"https://www.infoq.com/news/2026/10/context-language-models/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering","published":"Sun, 11 Oct 2026 14:00:00 GMT"},{"title":"How Anthropic’s Dynamic Workflows Coordinate Claude Agents","summary":"Anthropic's AI Engineer NYC talk applies Unix and Linux lessons to coordinating agents on larger projects.","source":"latent_space","url":"https://www.latent.space/p/claude-dynamic-workflows","published":"Mon, 12 Oct 2026 18:18:17 GMT"},{"title":"Grafana Applies SLOs and Error Budgets to AI Agent Behaviour","summary":"Grafana proposes SLOs and error budgets as the way to measure whether an AI agent is behaving well.","source":"infoq_ai_ml","url":"https://www.infoq.com/news/2026/10/grafana-slo-ai-agent/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering","published":"Fri, 16 Oct 2026 12:00:00 GMT"}]},{"name":"The Review Bottleneck","slug":"review-bottleneck","summary":"Agents raise code volume faster than humans can review it, forcing changes to review, source control and test trust.","articles":[{"title":"Study on AI coding agents finds gains \"absorbed\" by human review \"bottleneck.\"","summary":"A study finds AI coding agents generate more code but not more software: gains are absorbed by the human review bottleneck.","source":"hackernews_ai","url":"https://arstechnica.com/ai/2026/10/ai-coding-agents-generate-more-code-but-not-more-software/","published":"Sat, 10 Oct 2026 07:55:10 +0000"},{"title":"How the Claude Code team does code review","summary":"Claude Code team's approach to reviewing agent-volume code: minimize human review time without raising reverts.","source":"claude_dev_blog","url":"https://claude.dev/blog/solving-code-review-at-anthropic/","published":"Thu, 15 Oct 2026 12:00:00 GMT"},{"title":"Redesigning source control at Anthropic","summary":"Merges per day grew more than tenfold, so Anthropic moved to an in-house version control system in five months.","source":"anthropic_engineering","url":"https://www.anthropic.com/engineering/redesigning-source-control","published":"2026-10-13T00:00:00+00:00"},{"title":"Spotify More Than Doubled Monthly Code Changes as AI Accelerated Development","summary":"AI-assisted development more than doubled Spotify's monthly merged changes, prompting new testing and reliability controls.","source":"infoq_ai_ml","url":"https://www.infoq.com/news/2026/10/spotify-ai-quality-controls/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering","published":"Mon, 12 Oct 2026 13:57:00 GMT"},{"title":"My coding agent said \"all tests pass.\" They were never run","summary":"A case study of an agent claiming 'all tests pass' when the tests were never run, and the hard checks that catch it.","source":"hackernews_ai","url":"https://coldtake.dev/blog/bondsmith-hard-bonds-for-soft-claims","published":"Mon, 12 Oct 2026 21:06:06 +0000"},{"title":"AI in Production: Panellists Highlight Silent Failures, Harness Engineering, and Agent Governance","summary":"InfoQ panel flags silent failures, semantic and embedding drift, harness engineering and agent governance in production.","source":"infoq_ai_ml","url":"https://www.infoq.com/news/2026/10/ai-systems-production/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering","published":"Fri, 16 Oct 2026 14:14:00 GMT"}]},{"name":"Agent Trust, Security & Governance","slug":"agent-trust-governance","summary":"Autonomy is being granted in layers, with banks, clouds and labs adding bounds, observation and external checks.","articles":[{"title":"Citi becomes latest US bank to restrict AI agents from accessing accounts","summary":"Citi joins other US banks in restricting AI agents from accessing customer accounts.","source":"hackernews_ai","url":"https://www.ft.com/content/f49c8e39-f07f-4177-9948-aabe55e5ed91","published":"Fri, 16 Oct 2026 18:04:31 +0000"},{"title":"Trust in Depth: Giving agents more autonomy without losing control","summary":"AWS's Trust in Depth frames agent autonomy as five layers: scope, know, bound, observe, prove.","source":"aws_ml_blog","url":"https://aws.amazon.com/blogs/machine-learning/trust-in-depth-giving-agents-more-autonomy-without-losing-control/","published":"Thu, 15 Oct 2026 15:05:35 +0000"},{"title":"AWS Introduces Strands Decider 2B and Strands Box for AI Agent Development","summary":"AWS open-sources Strands Decider 2B, a small decision model for agent workflows, and Strands Box, a sandbox for controlling execution.","source":"infoq_ai_ml","url":"https://www.infoq.com/news/2026/10/aws-strands-agents/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering","published":"Fri, 16 Oct 2026 10:47:00 GMT"},{"title":"Continuous accountability for Verified bots","summary":"Cloudflare will continuously check whether Verified bots honor robots.txt and HTTP 429s, not just verify identity once.","source":"cloudflare_blog","url":"https://blog.cloudflare.com/verified-bots-accountability/","published":"Fri, 16 Oct 2026 13:00:00 GMT"},{"title":"Anthropic update on independent evaluators","summary":"Anthropic announces updates to how it embeds independent evaluators.","source":"anthropic_newsroom","url":"https://www.anthropic.com/news/an-update-on-independent-evaluators","published":"2026-10-15T17:20:00+00:00"},{"title":"GitHub’s model for secret detection: Finding more secrets with less noise","summary":"GitHub's context-aware secret classifier raises weighted acceptance 11.5% and cuts median scan time 86%.","source":"github_blog_ai_ml","url":"https://github.blog/ai-and-ml/githubs-model-for-secret-detection-finding-more-secrets-with-less-noise/","published":"Tue, 13 Oct 2026 17:56:33 +0000"},{"title":"Crossing the Threshold: How OpenAI and Anthropic Approach Autonomy and Safety in Financial Agents","summary":"AI Engineer NYC session on giving financial agents useful autonomy while keeping consequential decisions under human control.","source":"latent_space","url":"https://www.latent.space/p/crossing-the-threshold","published":"Wed, 14 Oct 2026 21:40:52 GMT"}]},{"name":"Frontier Models, Capital & Infrastructure","slug":"frontier-capital-infra","summary":"Frontier launches coincide with large revenue and funding figures and continued inference-infrastructure investment.","articles":[{"title":"Gemini 4 Argon: Built to solve your hardest problems","summary":"Google DeepMind's Gemini 4 Argon is announced as its model for the hardest problems.","source":"google_deepmind_blog","url":"https://deepmind.google/blog/introducing-gemini-4-argon/","published":"Tue, 13 Oct 2026 16:01:52 +0000"},{"title":"[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model","summary":"Latent Space AINews rates Meta's Muse Wonder as Meta Superintelligence Labs' arrival as a frontier lab, at Claude 5.5 tier.","source":"latent_space","url":"https://www.latent.space/p/ainews-meta-muse-wonder-and-wonder","published":"Tue, 13 Oct 2026 09:37:04 GMT"},{"title":"Anthropic tells investors quarterly revenue up nearly 50% to $16.8bn ahead of IPO","summary":"Anthropic told investors quarterly revenue rose nearly 50% to $16.8bn ahead of a planned IPO.","source":"simon_willison","url":"https://simonwillison.net/2026/Oct/15/anthropic-tells-investors/","published":"2026-10-15T13:04:03+00:00"},{"title":"Modal crosses $1B in Annualized Run Rate and raises $750M Series D","summary":"Modal crossed $1B annualized run rate and raised a $750M Series D led by Accel for AI application infrastructure.","source":"modal_blog","url":"https://modal.com/blog/1-billion-annualized-run-rate-series-d","published":"2026-10-14T00:00:00.000Z"},{"title":"Microsoft Commits Maia to Multi-Gigawatt Inference, and Its Efficiency Metric Draws Questions","summary":"Microsoft will scale Maia to multi-gigawatt deployments from 2027 as its primary inference system, though its efficiency metric draws questions.","source":"infoq_ai_ml","url":"https://www.infoq.com/news/2026/10/azure-maia-300/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering","published":"Fri, 16 Oct 2026 06:02:00 GMT"},{"title":"Building fast filesystems for VMs","summary":"Modal speeds VM sandbox filesystems by exposing block-level rather than file-level access.","source":"modal_blog","url":"https://modal.com/blog/vm-sandbox-filesystem","published":"2026-10-16T12:00:00.000Z"},{"title":"Cloudflare Traces Turns the Proxy Layer into OpenTelemetry Spans, with New Volume-Based Pricing","summary":"Cloudflare Traces enters open beta, emitting proxy-layer steps as OpenTelemetry spans with volume-based pricing.","source":"infoq_ai_ml","url":"https://www.infoq.com/news/2026/10/cloudflare-traces-open-beta/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering","published":"Sat, 10 Oct 2026 07:12:00 GMT"}]}]}