Gemini API adds background tasks to Managed Agents

#1 𝕏

AI at Meta introduced Muse Image, its most advanced image-generation model offering precise edits, multi-reference composition, Instagram-powered context and agentic tool use via Muse Spark in the Meta AI app, Instagram Stories and WhatsApp.

Also covered by: @AI at Meta, @Alexandr Wang

#2 𝕏

Logan Kilpatrick rolled out major updates to Managed Agents in the Gemini API—adding background task support, remote MCP & function calling, and network credential refresh—and now you can try them on the free tier.

Also covered by: @Logan Kilpatrick

#3 📝 Claude Code Blog

Claude Cowork is coming to mobile and web - Anthropic announced that Claude Cowork will be available on mobile and web, expanding access to collaborative Claude features. The post highlights product rollout and productivity-focused use cases.

#4 📝 Simon Willison

github-code Web Component - An experimental Web Component was built (using GPT-5.5) to embed code from GitHub URLs by converting them to raw.githubusercontent links and fetching/displaying specified ranges of lines with line numbers.

#5 𝕏

Harrison Chase built a local-first personal AI agent that leverages orchestration, persistent memory, and LangGraph-powered tool integrations. It runs background workflows and spins up child agents for modular, private on-device automation.

#6 𝕏

Philipp Schmid rolled out four new Managed Agent features in the Google DeepMind Gemini API—Background Execution (`background: true`), Remote MCP servers, Custom Function Calling, and credentials refresh across turns.

Also covered by: @Logan Kilpatrick

#7 𝕏

Alexandr Wang released Muse Image today—the first agentic image generation model from MSL that pairs with Muse Spark to reason through prompts, search the web, and plan outputs for first-try accuracy, now live in the Meta AI app.

Also covered by: @AI at Meta, @Alexandr Wang

#8 𝕏

Santiago introduced a first-of-its-kind benchmark testing five frontier AI models across 12 non-explicit child-safety risks—grooming, impersonation, minor profiling and emotional dependency—finding failure rates from 2% to 34%.

#9 𝕏

Hugging Face live-streamed “Training Agents 2,” a hands-on tutorial demonstrating how to use model distillation to train and optimize custom AI agents with their new framework.

#10 𝕏

Guillermo Rauch demonstrates how Eve.dev’s beautiful filesystem enables an open ecosystem of pluggable models, skills, channels, and tools—just define tools/github.ts and export createGithubTools() to give your agent full GitHub powers.

#11 📝 Claude Code Blog

Choosing a Claude model and effort level in Claude Code - This post explains how to choose a Claude model and appropriate effort level when using Claude Code, helping developers balance capability and cost. It provides guidance for coding workflows and selecting the right model for tasks.

#12 𝕏

NVIDIA AI introduces MOTIVE, a new NVIDIA Research method that re-weights video-model training signals toward moving regions, achieving 74.1% human preference over baselines.

#13 𝕏

Julien Chaumond is surprised by Anthropic’s unexpected release of open-source demos built on the Qwen model in collaboration with Neuronpedia.

#14 𝕏

Aravind Srinivas says Vera CPUs—custom-built for agentic runtimes—are now running Perplexity Computer’s NVIDIA-backed sandbox infrastructure with significant performance gains, with more details coming soon.

#15 𝕏

LlamaIndex 🦙 improved LlamaParse Cost Optimizer with intelligent tier routing that defaults simple pages to a cost-effective tier and sends complex pages to higher-accuracy agentic or agentic plus tiers.

#16 📝 Mario Zechner

Field Guide to Fable — Thariq Shihipar, Anthropic - Anthropic's Thariq Shihipar announces Fable is rolling out and demonstrates "capability overhang" by showing Claude Code fetch a Pokémon list and filter for names ending in aw—Croconaw and Drednaw—when ordinary chat models fail. He says Claude Code cut 80% of its system prompt, the ask‑user‑question tool went from barely working under Opus 4 to generating embedded HTML questionnaires under Fable, he built a full keynote deck in four hours, and urges teams to demand good, fast, and cheap.

#17 𝕏

Lenny Rachitsky (Lenny’s Newsletter) The 2026 survey reveals the tech workforce is bifurcating: 50% feel amplified by AI—more capable, confident, and excited—while the other 50% feel shaken about their value and future, and this split now predicts career sentiment more than a...

Also covered by: @Lenny Rachitsky (Lenny’s Newsletter)

#18 𝕏

bolt.new launched Fable 5 in Bolt’s Max agent, so when you select Max mode it auto-chooses the best model for your task (including Fable).

#19 𝕏

Google Research launched three FireSat satellites to scale the Earth Fire Alliance’s AI-powered, continuous high-resolution wildfire detection network. Built with @EarthFireAll and partners, this milestone leverages AI for enhanced climate resilience.

#20 𝕏

Google Research demonstrates that coordinating routing for just a small fraction of trips can disperse traffic and measurably boost average driving speeds.

#21 𝕏

Google DeepMind launched the Predicting the Past Skill in Google @antigravity, integrating Gemini with expert Aeneas and Ithaca models. It lets historians study and translate Greek and Latin texts using plain English.

#22 𝕏

Google DeepMind launched the Antigravity skill to tackle AI-driven history analysis’s three core challenges by translating complex workflows into plain English.

#23 𝕏

Harrison Chase asks if ATIF is emerging as the standard format for agent traces or if everyone’s still rolling their own.

#24 𝕏

Claude launched mobile and web support for Claude Cowork, enabling seamless task handoff from desktop to phone. Beta access rolls out over the next several weeks for Max plan subscribers, with additional plans to follow.

#25 𝕏

Claude merged Chat and Cowork into a single web and desktop app, unifying all your projects and artifacts in one workspace. Tasks now kick off with Claude just as easily as starting any conversation.