{"date":"2026-11-14","title":"What happened in AI — Nov 14, 2026","generated_at":"2026-11-14T22:15:57Z","intro":["Coordinating many coding agents is turning into a product feature: Claude Code Projects now scopes, delegates and reviews parallel work itself, and one founder's all-agent org merged 1,119 PRs in a week.","The day's best engineering writing put agent failures outside the model, in stale repo context, drifting job state, and per-tool permissions that leak data once agents are chained."],"highlights":["Claude Code Projects now runs a coordinator that splits work into parallel threads and reviews the results.","An all-agent org on Amp merged 1,119 PRs in a week and grew its own bureaucracy.","Google: stale specs and resolved bugs in a repo read as instructions to coding agents.","Three agents that each respect their permissions can still leak data together.","DeepSeek opened its app's search, file, OCR and voice tools as APIs and ported DeepEP to Huawei Ascend."],"article_count":39,"categories":[{"name":"Agent fleets: coordination moves into the product","slug":"agent-fleets-coordination","summary":"The manager role in multi-agent coding is being automated: a coordinator scopes, delegates and reviews, and the human steers from one thread.","articles":[{"title":"Anthropic upgrades Claude Code Projects to run parallel AI workers","summary":"Claude now scopes a request, splits it into parallel threads, reviews the outputs and assembles the result, and keeps working after you step away.","source":"hackernews_ai","url":"https://claude.com/resources/articles/projects-redesigned","published":"Sat, 14 Nov 2026 00:28:11 +0000"},{"title":"My agent org speedran a company","summary":"An all-agent engineering org on Amp merged 1,119 PRs in one week, and the agents rebuilt big-company bureaucracy when told to organize like one.","source":"hackernews_ai","url":"https://perfloop.ai/blog/my-agent-org-speedran-a-company","published":"Sat, 14 Nov 2026 19:27:11 +0000"},{"title":"Show HN: CodeRelay – Let your coding agent delegate work to a teammate's agen","summary":"Hands work between teammates' coding agents with API contracts pinned to Git commits, so two agents stop editing the same files.","source":"hackernews_ai","url":"https://github.com/AymanAhsan/CodeRelay","published":"Sat, 14 Nov 2026 16:56:39 +0000"},{"title":"Goal Engineering","summary":"LangChain argues open-ended agents need explicit metrics to optimize, and that a badly scoped goal mostly buys wasted tokens.","source":"langchain_blog","url":"https://www.langchain.com/blog/goal-engineering","published":"2026-11-14T16:00:00+00:00"}]},{"name":"Context and state: where agents actually fail","slug":"context-and-state-reliability","summary":"Two of the day's best writeups locate agent failures outside the model: stale repo context and drifting job state.","articles":[{"title":"Agent Distraction","summary":"Google warns that obsolete specs, resolved bugs and stale docs in a repo read as instructions to a coding agent and send it on side-quests.","source":"hackernews_ai","url":"https://developers.googleblog.com/agent-distraction/","published":"Sat, 14 Nov 2026 03:47:16 +0000"},{"title":"Why a 98%-accurate model leaves an agent's state right only 21% of the time","summary":"Per-step errors compound across a long job; the AXI paper holds job state on the host and gates every reply deterministically before applying it.","source":"hackernews_ai","url":"https://zenodo.org/records/23958737","published":"Sat, 14 Nov 2026 10:48:00 +0000"},{"title":"Write the eval before the fix: building a WhatsApp transcriber with LLM-TDD","summary":"A worked example of LLM-TDD: each bug in a voice-transcription pipeline gets a failing eval case before any prompt change.","source":"hackernews_ai","url":"https://waldo.vanderlore.de/blog/whatyousay-privacy-voice-pipeline/","published":"Sat, 14 Nov 2026 12:43:31 +0000"},{"title":"Show HN: Portal2Bench – Can AI agents solve Portal 2?","summary":"An agent benchmark on Portal 2 test chambers capped at $30 per run: GPT-6.1 Sol solves 20 of 26, Opus 5.5 solves 11.","source":"hackernews_ai","url":"https://kingstonyip.com/portal2bench","published":"Sat, 14 Nov 2026 00:40:55 +0000"}]},{"name":"Agent permissions don't compose","slug":"agent-security-permissions","summary":"Per-tool permissions are not a security boundary; the fixes on offer are reference monitors, staged effects and hard spend caps.","articles":[{"title":"Transitive Policies for Agents: Agent Permissions Don't Automatically Compose","summary":"Three agents that each obey their own tool permissions can still leak customer data once an orchestrator chains them together.","source":"hackernews_ai","url":"https://glthr.com/transitive-policies-for-ai-agents","published":"Sat, 14 Nov 2026 12:33:40 +0000"},{"title":"The Missing Ring-0 Protections for AI Agents","summary":"Argues coding agents need a reference monitor they cannot edit, and that it only works if file and command effects are staged first.","source":"hackernews_ai","url":"https://hwtn.io/essays/t0013-missing-ring-0-protections-ai","published":"Sat, 14 Nov 2026 16:23:05 +0000"},{"title":"Before I Trust an Agent, I Try to Break Its Box","summary":"A hands-on attempt to escape a per-agent sandbox from the inside, listing what broke and what is still untested.","source":"hackernews_ai","url":"https://kunalkushwaha.github.io/2026/11/06/before-i-trust-an-agent/","published":"Sat, 14 Nov 2026 16:14:06 +0000"},{"title":"Show HN: Shotoku – Local-first action governance and budget caps for AI agents","summary":"Local, deterministic per-vendor and per-day spend caps for agents, enforced before a payment goes out, with a tamper-evident audit log.","source":"hackernews_ai","url":"https://www.shotoku.dev/","published":"Sat, 14 Nov 2026 13:02:53 +0000"}]},{"name":"Harness building blocks and reference stacks","slug":"harness-building-blocks","summary":"Builders shared the practical parts of a harness: tool loops, browsers, approval flows, and a catalog of what production teams run.","articles":[{"title":"Build Your Own AI Coding Agent Harness","summary":"Vercel's course builds a coding-agent harness from the tool loop up, covering context pruning, sandboxing and subagent delegation.","source":"hackernews_ai","url":"https://vercel.com/academy/build-ai-agent-harness","published":"Sat, 14 Nov 2026 12:09:46 +0000"},{"title":"Show HN: Browserr – An MCP-first browser for AI agents (one Rust binary)","summary":"An MCP-first browser engine in one Rust binary with no CDP or Chromium; it blocks when unsure and hands the page to a human.","source":"hackernews_ai","url":"https://browserr.dev","published":"Sat, 14 Nov 2026 03:22:22 +0000"},{"title":"Building a human-in-the-loop Slackbot agent","summary":"A reference Slackbot where code and Claude clear easy cases and people approve risky ones, built around outages, token overruns and bad output formats.","source":"hackernews_ai","url":"https://mays.co/human-in-the-loop-slackbot-agent","published":"Sat, 14 Nov 2026 14:34:08 +0000"},{"title":"WhatWeRun, a database of real production AI agent stacks","summary":"A catalog of production agent stacks drawn only from public writeups, with costs kept on the source's own basis.","source":"hackernews_ai","url":"https://whatwerun.dev","published":"Sat, 14 Nov 2026 19:42:04 +0000"},{"title":"Show HN: SunsetLint – find retiring AI model IDs in your GitHub repos","summary":"Finds retiring model IDs in your repos by file and line, next to the shutdown date and the provider's suggested replacement.","source":"hackernews_ai","url":"https://www.sunsetlint.com","published":"Sat, 14 Nov 2026 13:41:01 +0000"}]},{"name":"Models, APIs and non-NVIDIA infra","slug":"models-apis-infra","summary":"DeepSeek shipped on two fronts: its app's tools became APIs, and its expert-parallel library now targets Huawei Ascend.","articles":[{"title":"DeepSeek Opens Its App's Search, File, OCR and Voice Tools as APIs So Developers Can Rebuild the App","summary":"DeepSeek exposes its app's search, file, OCR and voice tools as APIs, so developers can rebuild the app's features on their own stack.","source":"search_cn_open_weight_labs","publisher_name":"Pandaily","publisher_domain":"pandaily.com","url":"https://news.google.com/rss/articles/CBMiowFBVV95cUxNREtNc2xqMVZHOGluRzU2dVJpclVpQmZqcEM3ZzN3WU5XYmdtdHNuWE80UzBmMHA0RWoxU2EtbnpDWXNRbEIwclVPMmZtN0pzV3NVd0R5bDZyMTg3b1RaOUpvX084Sm1wVGFmSVk4dUNmM19GOTluem10WDJBZ2N3ZERPUVRVLUFIbGl4VFdXRFh6Q1R6d1ZmYTNTM29yZ1lPdUY0?oc=5","published":"Sat, 14 Nov 2026 02:00:31 GMT"},{"title":"Training and Inference on Huawei Ascend NPUs","summary":"DeepEP-Ascend ports DeepSeek's MoE all-to-all communication library to Ascend NPUs with FP8 dispatch and a buffer API matching the NVIDIA version.","source":"hackernews_ai","url":"https://github.com/deepseek-ai/DeepEP-Ascend","published":"Sat, 14 Nov 2026 16:30:03 +0000"},{"title":"I Tried Every LLM Architecture Change Since 2020","summary":"An ablation of architecture changes since GPT-3 finds fine-grained MoE the biggest win, with linear attention and hyper-connections also paying off.","source":"hackernews_ai","url":"https://www.nikita-pavlichenko.com/blog/every-llm-architecture-change","published":"Sat, 14 Nov 2026 18:41:54 +0000"}]}],"video":{"youtube_id":"sQfxuUu8_mE","title":"LLM Digest Daily Explainer: Nov 14, 2026","duration_s":115.5}}