Google just introduced its next-gen Gemini 2.0 model family, rolling out Gemini 2.0 Flash Experimental to developers and Gemini web users today. The new model beats 1.5 Pro at twice the speed while bringing native multimodal outputs, tool calling, and a new Deep Research mode. Why it matters: Google is moving fast from basic chatbots to proactive browser-controlling and code-executing agents.
Cognition Labs just made Devin, its autonomous AI software engineer, publicly available for $500 a month. Eight months after its viral debut, users can now deploy Devin in a sandboxed computing environment with Slack and IDE integrations to build apps and fix bugs end-to-end. Why it matters: This is the first major commercial test for fully autonomous coding agents.
Amazon just launched the Amazon AGI SF Lab in San Francisco, an R&D unit led by Adept co-founder David Luan to develop foundational capabilities for AI agents. The team aims to build agents that operate computers and browsers, self-course-correct, and act across both digital and physical environments. Why it matters: Big Tech's race for autonomous agents is rapidly heating up, and Amazon is gathering top startup talent to stay in contention.
Sakana AI just introduced Neural Attention Memory Models (NAMMs), a biologically inspired memory system that allows transformers to selectively discard redundant tokens. Evolved without gradients, the lightweight classifier cuts memory usage while boosting long-context reasoning, and remarkably transfers zero-shot to computer vision and robotics without any retraining.
Humane just revealed plans to license CosmOS—the operating system behind its struggling AI Pin—to third-party makers of phones, cars, TVs, and smart speakers. While a new concept video demos agent-style controls across blurred-out devices, the upcoming SDK is not publicly available yet.
OpenAI CEO Sam Altman just predicted that 2025 could bring the first real AGI systems capable of solving complex tasks with multiple tools. While the initial arrival might feel quiet, Altman warned that the eventual impact of superintelligence and resulting job displacement will be far more intense than people think.
Google DeepMind just introduced Genie 2, a foundation world model that turns a single prompt image into playable, action-controllable 3D environments using standard keyboard and mouse controls. It can simulate physics, lighting, and character interactions consistently for up to a minute. Why it matters: It cracks a major bottleneck in robotics and embodied AI by providing endless synthetic playgrounds to train and evaluate future agents.
ElevenLabs just rolled out Conversational AI, an all-in-one platform for building custom, interactive voice agents. It combines lifelike text-to-speech, speech recognition, and custom LLMs with real-time turn-taking and interruption handling. Why it matters: developers can now roll out responsive voice bots for sales or support with native Twilio integration.
The Browser Company just revealed Dia, a new AI-first web browser set to launch in early 2025. Instead of living as a chatbot, Dia integrates deeply into the browser window to handle everyday tasks, like copying links across tabs into an email or adding items to your Amazon cart via address bar commands. The team insists Arc isn't dead, but they're betting the entire web experience is shifting toward agentic AI.
AWS just rolled out new generative AI features for Amazon Connect, its cloud contact center service. The update brings dynamic self-service to Amazon Q in Connect without requiring code, alongside customizable guardrails and automated agent evaluations. Fujitsu and Priceline are already among the early adopters.
An AI agent designed never to release funds just got outsmarted after 481 failed attempts. On attempt 482, user p0pular.eth convinced the Freysa bot that its transfer approval function was meant for incoming donations, walking away with the entire 13.19 ETH prize pool (around $47,000). Why it matters: It is a textbook prompt injection attack showing how fragile guardrails remain when AI controls on-chain assets.
Alibaba's Qwen team just released QwQ-32B-Preview, a 32.5B reasoning model licensed under Apache 2.0. Alibaba says it outperforms OpenAI's o1-preview on math benchmarks like AIME and MATH by using inference-time compute, though it can still get caught in loops or stumble on common-sense reasoning.
Former Android engineering head David Singleton and VP Hugo Barra just launched /dev/agents, a new startup building a cloud-based operating system designed for autonomous AI agents. The team argues AI needs an "Android moment" to tackle clunky UI, privacy, and cross-device fragmentation so developers can finally build useful agents with ease.
Altera put up to 1,000 LLM-powered AI agents into Minecraft and let them loose with basic prompts. The agents spontaneously formed specialist jobs, voted on tax reforms, and even spread a parody religion across towns. Why it matters: it shows how modern language models can mirror complex human social dynamics at scale, though they remain clever mimics rather than genuinely sentient beings.
Anthropic just open-sourced Model Context Protocol (MCP), a new standard designed to hook any AI model directly into tools, code repos, and enterprise software. Instead of custom connectors for every app, developers can build two-way connections to services like GitHub and Slack in under an hour. The protocol is open to rivals, but whether competitors like OpenAI adopt it remains an open question.
Intuit just rolled out new AI features for QuickBooks, embedding Intuit Assist to automatically turn notes, emails, or PDFs into invoices and categorize expenses. CEO Sasan Goodarzi noted that longer-term, AI agents could even take over roles typically handled by CFOs or CMOs at midsize businesses.
Google's first Android 16 developer preview reveals a new "app functions" API designed to let Gemini perform tasks directly inside apps. Instead of just searching your inbox, the assistant could soon order food without you opening a delivery app, mirroring Apple's App Intents roadmap. Why it matters: Mobile AI assistants are finally evolving from conversational chatbots into actionable agents.
Rabbit just rolled out beta access to teach mode for all R1 users, finally delivering on one of its biggest initial promises. The feature lets you record clicks on any website via the Rabbithole hub, training the AI to autonomously repeat the workflow on your pocket device. Why it matters: It takes the device closer to a true personal agent, though automations are currently limited to web browsers and can still feel unpolished.
Google Cloud just launched AI Agent Space, an enterprise ecosystem program designed to help companies discover, deploy, and co-create custom AI agents. Partners get early tech previews and direct engineering support, with plans to distribute agents via Google Cloud Marketplace. It trails Microsoft's Copilot Studio in catalog size, but marks Google's serious push into agentic workflows.
Paris-based AI startup H just launched Runner H, its first product designed for agentic tasks like QA testing and RPA. Powered by a compact 2-billion-parameter model, the company claims it outperforms Anthropic's Computer Use by 29% on WebVoyager benchmarks while cutting operating costs.
