{"date":"2026-11-03","title":"What happened in AI — Nov 3, 2026","generated_at":"2026-11-03T22:17:43Z","intro":["Observability went OpenTelemetry-native: Claude Platform traces now flow into Databricks MLflow, and LangSmith BYOC reached GA on AWS with Azure and Google Cloud in Beta.","Alongside it, a study of eval-versus-deployment behavior gaps and nearly $1M in stolen-API-key charges show where agent operations can fail quietly."],"highlights":["Claude Platform ships native OpenTelemetry ingestion into Databricks MLflow.","LangSmith BYOC is GA on AWS, Beta on Azure and Google Cloud.","Agent safety evals may not predict deployment behavior because of eval-awareness.","Mandiant: stolen AI API keys produced nearly $1M in charges for one company.","'Weight Space Programming' paper: LLMs can read and write network weights.","MoE inference for Apple Silicon fits 125B parameters in under 10 GB of peak memory."],"article_count":24,"categories":[{"name":"Agent harnesses, memory and tool use","slug":"agent-harnesses-tool-use","summary":"Teams are consolidating agents onto shared harnesses and adding runtime guardrails around MCP tool calls.","articles":[{"title":"We built one harness for every agent, inspired by Ramp's Inspect","summary":"Metaview runs sourcing, screening and reporting agents on one shared harness, modeled on Ramp's Inspect, instead of per-agent scaffolding.","source":"hackernews_ai","url":"https://www.metaview.ai/resources/blog/how-we-built-one-harness-for-every-metaview-agent","published":"Tue, 03 Nov 2026 14:05:10 +0000"},{"title":"How We Optimized Browser Use for Deep Agents","summary":"Deep Agents' browser-use rework targets multi-step web tasks; the write-up covers what changed in the harness to make them reliable.","source":"langchain_blog","url":"https://www.langchain.com/blog/how-we-optimized-browser-use-for-deep-agents","published":"2026-11-03T16:00:00+00:00"},{"title":"A complete (?) library of agent optimization methods","summary":"A catalog of agent optimization methods with evidence for each, plus a prompt that audits your own agent against every entry.","source":"hackernews_ai","url":"https://mahalsystems.ai/aot/optimizations/","published":"Tue, 03 Nov 2026 14:01:23 +0000"},{"title":"Show HN: MCPgraphy – a proxy that learns when your agent's MCP tools work","summary":"A proxy that learns which MCP tools succeed in which state, then warns or blocks the next likely-bad call.","source":"hackernews_ai","url":"https://github.com/hemanth1999k/mcpgraphy","published":"Tue, 03 Nov 2026 05:07:22 +0000"},{"title":"Vercel: Computer use CLI for AI agents","summary":"Vercel's agent-desktop CLI exposes a desktop app's accessibility tree as JSON with refs the agent acts on.","source":"hackernews_ai","url":"https://agent-browser.dev/desktop","published":"Tue, 03 Nov 2026 05:26:34 +0000"},{"title":"Show HN: Replacing Turn-Based LLM Agents with SQLite and System One Models","summary":"Claims a SQLite-backed System One coding agent matches Claude Fable 5.1 on a 74-point xz benchmark (69/74) at 1/20 the cost; self-reported, unreplicated.","source":"hackernews_ai","url":"https://github.com/DevLoops-ai/unified-brain","published":"Tue, 03 Nov 2026 17:58:18 +0000"}]},{"name":"Observability, evals and agent reliability","slug":"observability-evals-reliability","summary":"Tracing is going OpenTelemetry-native across vendors, while eval research questions how well tests predict deployment behavior.","articles":[{"title":"Claude Platform now supports native OpenTelemetry observability in Databricks MLflow","summary":"Claude Platform traces now ingest natively via OpenTelemetry into Databricks MLflow, so teams can mine recurring failures from production runs.","source":"databricks_blog","url":"https://www.databricks.com/blog/claude-platform-now-supports-native-opentelemetry-observability-databricks-mlflow","published":"Tue, 03 Nov 2026 21:00:00 GMT"},{"title":"LangSmith BYOC is now available on AWS, Azure, and Google Cloud","summary":"LangSmith BYOC is GA on AWS and Beta on Azure and Google Cloud, keeping traces, evals, datasets and prompts in your own cloud.","source":"langchain_blog","url":"https://www.langchain.com/blog/langsmith-byoc-aws-gc-azure","published":"Tue, 03 Nov 2026 22:06:58 GMT"},{"title":"Show HN: Traccia – Control what AI agents can do","summary":"An OpenTelemetry-native, framework-agnostic SDK adding tracing, evaluation and runtime policy enforcement for agents.","source":"hackernews_ai","url":"https://github.com/traccia-ai/traccia-py","published":"Tue, 03 Nov 2026 20:06:53 +0000"},{"title":"Eval-production behavior gaps in agent safety and the role of eval-awareness","summary":"Compares agent behavior in safety evals versus deployment and starts tracking the gap, flagging eval-awareness as a confound.","source":"hackernews_ai","url":"https://alignment.openai.com/eval-production-behavior-gap/","published":"Tue, 03 Nov 2026 18:01:38 +0000"},{"title":"Why human insight is a strategic advantage for building better agents","summary":"LangChain argues long-horizon agent evals need human review of the trajectory, not just the final answer.","source":"langchain_blog","url":"https://www.langchain.com/blog/human-insight-strategic-advantage-agents","published":"2026-11-03T21:07:00+00:00"},{"title":"Show HN: A tool to optimize AI model cost without benchmark score trade-offs","summary":"A tool that finds the cheapest model blend that still matches target benchmark scores per setting.","source":"hackernews_ai","url":"https://jeff-kline.github.io/posts/same-ai-results-for-less/index.html","published":"Tue, 03 Nov 2026 12:38:49 +0000"}]},{"name":"Security and identity for agents","slug":"security-identity-agents","summary":"Agent identity delegation and stolen API keys are the concrete operational risks today.","articles":[{"title":"ServiceNow on-behalf-of token exchange with Bedrock AgentCore Gateway","summary":"Walks through ServiceNow on-behalf-of token exchange via Bedrock AgentCore Gateway, so agents act with the user's identity, not a shared one.","source":"aws_ml_blog","url":"https://aws.amazon.com/blogs/machine-learning/servicenow-on-behalf-of-token-exchange-with-bedrock-agentcore-gateway/","published":"Tue, 03 Nov 2026 16:47:40 +0000"},{"title":"Stolen AI API keys generated nearly $1M in charges for a single company","summary":"Mandiant traced nearly $1M in charges at one company to stolen AI API keys.","source":"hackernews_ai","url":"https://cloud.google.com/blog/topics/threat-intelligence/llmjacking-monetizing-compromised-cloud-models/","published":"Tue, 03 Nov 2026 00:32:32 +0000"},{"title":"Fuzzers vs. AI: Comparing Fuzzilli and BigSleep on JavaScript Engines","summary":"Project Zero compares Fuzzilli with Google's BigSleep AI on finding V8 vulnerabilities.","source":"hackernews_ai","url":"https://projectzero.google/2026/10/fuzzing-v-AI.html","published":"Tue, 03 Nov 2026 09:04:15 +0000"},{"title":"Progress on provable inference for large language models","summary":"Anthropic-affiliated researchers report progress on cryptographically verifying that an LLM's outputs came from the claimed model.","source":"hackernews_ai","url":"https://www.anthropic.com/research/provable-inference","published":"Tue, 03 Nov 2026 06:44:10 +0000"}]},{"name":"Inference and on-device deployment","slug":"inference-on-device","summary":"Memory-efficient MoE serving and on-device toolchains keep lowering the hardware bar.","articles":[{"title":"Show HN: Run faster MoE inference on Mac with less RAM","summary":"MoE inference for Apple Silicon runs a 125B-parameter model at up to 11 tokens/s in under 10 GB of peak memory.","source":"hackernews_ai","url":"https://github.com/davide-troiani/moeshift","published":"Tue, 03 Nov 2026 21:17:06 +0000"},{"title":"The Inference Stack Behind OCRia","summary":"OCRia's team details the serving optimizations behind their OCRia-1 document-parsing model after inference became the bottleneck.","source":"hackernews_ai","url":"https://www.llamaindex.ai/blog/the-inference-stack-behind-ocria","published":"Tue, 03 Nov 2026 18:14:26 +0000"},{"title":"Google Launches LiteRT Tensor API for Building, Customizing, and Optimizing ML Models for On-Device","summary":"Google's LiteRT Tensor API targets building, customizing and optimizing ML models for on-device deployment.","source":"infoq_ai_ml","url":"https://www.infoq.com/news/2026/11/litert-tensor-api/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering","published":"Tue, 03 Nov 2026 19:00:00 GMT"}]},{"name":"Models and research","slug":"models-research","summary":"Weight-space programming dominates research discussion; DeepSeek and Huawei shift the Chinese open-weight landscape.","articles":[{"title":"[AINews] LLMs can read & write neural network weights: \"Weight Space Programming\" cuts Autoresearch out of the RSI loop","summary":"Reports that LLMs can natively read and write neural network weights, a 'Weight Space Programming' result aimed at removing autoresearch from the self-improvement loop.","source":"latent_space","url":"https://latent.space/p/ainews-llms-can-read-write-neural-network-weights-weight-space-programming-cuts-autoresearch-out-of/","published":"Tue, 03 Nov 2026 05:35:00 GMT"},{"title":"Qwen 3.8 27B on my laptop can interpret model weights, too","summary":"Simon Willison runs Qwen 3.8 27B locally to interpret model weights, following the Weight-Space Programming paper.","source":"simon_willison","url":"https://simonwillison.net/2026/Nov/3/qwen-can-interpret-model-weights/","published":"2026-11-03T16:29:48+00:00"},{"title":"Verify, Then Vote: Overturning Wrong Majorities in Multi-Agent LLMs","summary":"Research on a verify-then-vote scheme that overturns wrong majorities in multi-agent LLM voting.","source":"hackernews_ai","url":"https://github.com/grainulation/papers","published":"Tue, 03 Nov 2026 06:16:17 +0000"},{"title":"DeepSeek Harness Adds Creator Mode For Text-to-plugin Generation","summary":"DeepSeek Harness adds Creator Mode, letting users generate agent plugins by chat.","source":"search_cn_open_weight_labs","publisher_name":"Dataconomy","publisher_domain":"dataconomy.com","url":"https://news.google.com/rss/articles/CBMikgFBVV95cUxPckFDR25sUkRoZC1yNGwtVjlVWm40dFRZTDZXbG9CaGFZUGFYWm1LMGEybTNIZzR2SHJRcXlZdWhybHJJeFpjQ2dNeTIxd1dDNFlRVkZ5aUxEX0RDbjZPcllSUV9paE1Ld1FyVjY4ZEVXQXNzZUM5UVFZVmJqdnV4WkJSNFRXcVRuSklneERVQncwUQ?oc=5","published":"Tue, 03 Nov 2026 08:50:23 GMT"},{"title":"Huawei Cloud to Retire openPangu-2.0 Models From MaaS, Recommends DeepSeek-V4.1-Flash and GLM-5.3","summary":"Huawei Cloud will retire openPangu-2.0 from its MaaS and point users to DeepSeek-V4.1-Flash and GLM-5.3.","source":"search_cn_open_weight_labs","publisher_name":"Pandaily","publisher_domain":"pandaily.com","url":"https://news.google.com/rss/articles/CBMiqAFBVV95cUxORXJiejlQNkFkN2xsVlU4OFRKWU80amRyV3M1OWpqMmNCRVhDbXExdzM2SjZzQ3ZwYlA1S1lERDJaNUZXTjVoY2Q5eWhKZEVUQVFDY2hnaFpZdkVPdTRvTDB0YkpEVElJYTJxaUFXTzZuRjdIVFpvY2hFWEJuUUx0SlFNZVFGQy1pREtpNE43Tk8wcldvX1ZSQkVKRkpraW14aTA2OVR5d2o?oc=5","published":"Tue, 03 Nov 2026 20:55:27 GMT"}]}],"video":{"youtube_id":"diX2bYhOZWA","title":"LLM Digest Daily Explainer: Nov 3, 2026","duration_s":109}}