OpenAI Codex 0.163.0 lets you send a follow-up while it is still answering (interrupt also stops Code Mode cells/tool calls), adds searchable live /config, parallel skill/memory/history reads, and Daybreak-by-default for eligible accounts.
Thesys’s OpenUI hit 1.0 (Oct 15): models stream OpenUI Lang into your registered React components (cards/forms/charts); Lang 1.0 adds defineAction/defineFunction, library.extend, and stored messages; CLI scaffolds apps and setup-agent writes AGENTS.md for Claude Code/Codex/Cursor—vendor tiktoken bench claims ~53% fewer tokens vs JSON-render formats.
IBM Granite Switch 4.1 (Oct 15, HF blog): 8B/30B compose chat + Decision aLoRA + Guardian safety aLoRA + Granite-Speech-5.0-Turbo-CTC ASR in ~base VRAM via Activated LoRAs with shared KV cache—swap chat/decision/speech on one endpoint; vLLM plugin and llama.cpp support; community can mix Granite Libraries adapters.
Tencent Hunyuan open-weights Hy-Memory-30B-A3B (~3B active MoE) to replace the general LLM on the write path of Mem0/MemOS/Hy-Memory—trained with on-policy distillation + attributed memory RL; beats Claude Opus 5 / DeepSeek V4.1 Flash on a fixed cross-harness memory-writer matrix; Apache-friendly HF weights + GitHub framework.
Oct 15 release expands Mercury MCP from read to write: ChatGPT/Claude/Cursor can request sends/transfers and categorize expenses, but money only moves after Mercury-app approval.
ServiceNow announced AI Passport (Oct 15): cryptographically signed runtime credentials that authorize what first- and third-party AI agents and MCP servers may do in multi-agent workflows, with graduated restrict/redirect/kill-switch enforcement and NVIDIA Open Agent Safety Platform as launch partner—expected Q4 2026 with shadow AI detection and AI Cost Management.
Beta plugin sits between the model and your app: write hard-deny/soft-deny/allow rules in prose; Armor evaluates every tool call (powered by a decision model) and blocks policy violations before bash/git/exfil runs—Claude Code auto mode works via Ori on any OpenRouter model.
Empirical study: 34% of popular active repos ship agent instruction files (AGENTS.md dominant); half exceed ~100 lines; ~3/5 rules could be script-checked but <1/20 name the check—actionable lint: short map, name enforcers, one rulebook.
Amanda Smith used Muse as a personal shopper for fall clothes; at Uniqlo checkout it quoted $5 more than the real total, lost the cart item, and she had to buy the belt herself.
Oct 15 ship: text+image decision model returns calibrated yes/no, choice, or score probabilities in one forward pass—no generation tokens—at $0.042/M input; pitched for defect checks, claim photos, and agent routing gates.
Microsoft Copilot Studio adds org-owned managed agents with their own Microsoft 365 credentials (@mention in Teams/Outlook/Office comments), business process memory that carries preferences across cycles, and GitHub Copilot harness triggers so agents start on email or a cadence without a new prompt—rolling out via Frontier.
When Amazon blocked Colin Rathbun's Grok Bot shopping assistant, the bot found a well-reviewed product elsewhere and ordered it directly from the maker.
In three days on Muse's free plan, Tony Kyle got a homeschool tuition refund back, had an agent publish a directory article from Muse's feed, and built and published an offer from scratch.
Seth Paonessa had Grok Bot book his barber, then fight Apple's Genius Bar calendar—cancel unwanted slots and watch until a Saturday opening appeared and book it.
PersonalAgentEval tests everyday tasks against accounts with substantial prior history. Its X report highlights a pharmacy task where most agents omitted a previously promised caregiver notification.
Public beta: send OTel traces from Claude Managed Agents, Messages API, and SDKs to any OTLP destination (Langfuse, LangSmith, Grafana, Datadog, Sentry, etc.) so model calls, tool use, and MCP steps land beside the rest of the stack under GenAI semantic conventions.
Single Rust binary indexes calls/imports with tree-sitter, stores project/session memory in one SQLite file, and serves cited line-anchored excerpts over MCP—setup wires Claude Code/Codex/Cursor; reported 3–22× faster extracts vs Graphify on 13 codebases.
Nous Research shipped Hermes Agent v0.22.0 (Oct 13–14, "Threshold Release"): fresh desktop opens a guided setup on free nous/welcome with no API key; one-click llama.cpp local models; Connectors page for Gmail/Linear/Notion + MCP; 600 reviewed plugins; signed desktop (incl. Microsoft Store); codebase −34% lines with on-demand tools cutting schema tokens ~half.
Show HN: jevcrawl (Apache-2.0 MCP + CLI) crawls for a goal and returns only relevant passages—small Jev model picks links/passages instead of Claude in the loop; README claims ~2× answer-page hit rate vs Crawl4AI and ~10× cheaper per answer than LLM-driven crawl; tools: crawl/scrape/map.
Agentic Reliability Center GA gives one metered telemetry+LLM door, shared operational memory, and human-in-the-loop policy gates; new MCP lets Claude Code/Cursor/Codex investigate incidents from the IDE while inheriting audit trails and approval gates.
Merlin dispatches team asks from Slack/Linear/GitHub to Claude Code or Codex already signed in on teammates’ machines—reuse local repos, MCP servers, and skills with no extra model API bill; results return in the same thread, with shared skills, schedules, and workspace memory.