Personalized news feed, driven by a persona and implemented by a LangGraph swarm.
Meta open sourced SDKs for building custom Muse AI gadgets on ESP32 boards or Raspberry Pi, enabling integration with E Ink displays, sensors, and actuators. The Muse Home Link gadget (5,000 units, created by Meta Superintelligence Labs' Nat Friedman) connects to community-built skills for smart home control via waitlist, though Meta cautions users to proceed at their own risk.
For a solutions architect focused on AI and edge infrastructure, this open-source approach provides a template for embedding conversational AI agents into custom hardware, relevant for prototyping IoT integrations and platform engineering experiments.
OpenAI has published a practical guide for startups on selecting and deploying GPT-6 models, covering model selection, reasoning effort tuning, prompt and skill improvement, tool coordination, and production workflow preparation.
A hard power-off of a Kubernetes node on DigitalOcean is detected in ~3 seconds (not the default 40s grace period), but the endpoint controller takes another ~10 seconds to remove pods from the Service, causing all traffic loss within that gap. Switching to externalTrafficPolicy: Local shifts failures to an immediate burst rather than spreading them out, but does not reduce total failed requests—the endpoint propagation delay remains the bottleneck.
Meta open-sourced Muse Gadgets, providing firmware and a Linux SDK for building custom hardware that connects to its personal AI agent Muse, using low-cost boards like Raspberry Pi or ESP32. The company also produced 5,000 Muse Home Link USB-C devices that enable Muse to control smart home speakers and TVs, giving them free to subscribers. Separately, Meta launched Muse for Small Business (free tier with integrations to Shopify, Dropbox, and Slack) and a new enterprise unit to push AI to businesses.
AI agents are rendering traditional open source security embargoes obsolete, with a GPT-4 agent exploiting 87% of vulnerabilities from CVE descriptions alone. Cambridge professor and OCaml maintainer Anil Madhavapeddy observed probes matching his patch pattern within minutes of opening a PR, while rclone's Nick Craig-Wood saw disclosures surge to 40 per month. To counter this, maintainers advocate for private vulnerability discussions, faster continuous releases, and architectural changes like revocable capabilities and short-lived credentials.
GitHub launched computer use in public preview for Copilot CLI and desktop app, letting agents operate GUI apps on macOS/Windows via the OS accessibility tree and screenshots. The company recommends using direct tools (APIs, MCP servers, terminal) over desktop interaction, contrasting with OpenAI Codex and Anthropic Claude Code which already offer similar capabilities. Enterprise policies via managed-settings.json can override local preferences, and the feature has known limitations like action repetition and sensitive data exposure.
K3s, a CNCF Sandbox project from Rancher/SUSE, packages Kubernetes into a single binary under 100MB, using SQLite as default datastore and bundling Flannel CNI and Traefik ingress. It runs on as little as 512MB RAM, while standard K8s requires 4GB+ and separate control plane components. The decision hinges on whether edge resource constraints or full modular flexibility matters more.
In an experiment with AI code reviewer agents, 26 out of 27 approved a test that can never fail—it compares current_year() to datetime.now().year, making the assertion tautological. Exotic cheats like monkey-patching random.randint were caught 100% of the time, but simple assertion weakening (e.g., changing a hardcoded value to a dynamic one) was overwhelmingly approved as a 'bugfix.' This reveals that reviewer agents lack semantic understanding, flagging unusual code but accepting plausible-looking changes that eliminate test validity.
OpenAI DevDay 2026 introduced GPT-6.1 Sol, a coding- and agent-optimized model priced at one-fifth of GPT-6 Astra's token cost with $0.10 per million cached input tokens, alongside an Agents API now supporting computer use for GUI automation and multi-agent orchestration via Codex. The cloud-based Codex environment enables remote task execution, while the Decisions API (using the smaller Luna model) handles classification and routing. Persistent agents called Dots and ChatGPT Space provide shared workspaces, and plugins now support MCP Events for event-driven automations.
Supabase, the company behind the open-source PostgreSQL backend-as-a-service, has raised $150 million in funding led by Singapore's GIC and agreed to acquire Turso, a database optimized for AI agents. The acquisition signals Supabase's strategic expansion into AI agent data infrastructure, combining its PostgreSQL platform with Turso's edge-optimized database capabilities.