Google's annual developer conference dominated the AI news cycle today, delivering a wave of product announcements that collectively signal the company's pivot from reactive AI assistant to always-on agentic infrastructure. Elsewhere, Andrej Karpathy made a landmark career move, SpaceX set its sights on a $60 billion coding startup, and Meta began cutting thousands of jobs in a sweeping AI-era restructuring.
---
Google I/O 2026: The Agentic Pivot
Google used I/O 2026 to reframe its entire product portfolio around one idea: AI that acts, not just answers. Under the banner of "always-on agents," Google announced that idle devices—phones, laptops, smart displays—would become perpetually running AI workers capable of handling multi-step tasks in the background.
The centerpiece model is Gemini 3.5 Flash, now generally available, which Google says runs four times faster than rival frontier models. It powers agentic workflows, long-horizon task execution, and coding tasks, and it is expanding across Search, Android Studio, and developer platforms. The model comes with a price increase over its predecessor, though Google has made clear it intends Gemini 3.5 Flash to underpin virtually everything it ships.
Built on top of it is Gemini Spark, a 24/7 cloud-based AI agent that runs on Google Cloud VMs and integrates natively into Gmail, Docs, Calendar, and third-party apps via the Model Context Protocol. Unlike existing AI assistants, Spark operates continuously within user-defined limits rather than waiting for a prompt—parsing credit card statements, monitoring school emails, or managing calendar conflicts without being asked. Gemini Spark launches next week for U.S. Google AI Ultra subscribers, whose plan Google has cut from $250 to $200 per month.
Gemini Omni rounds out the model family as a "world model" system designed to understand physics and object continuity. It can generate and edit videos from any combination of text, images, audio, and existing clips using conversational prompts, and it can turn a single photograph into a full video in seconds. Google's positioning is that Omni doesn't just stitch frames together—it understands how the world moves.
On the search side, Google made its most significant redesign since 2001. The new "intelligent search box" accepts text, images, video, files, and open Chrome tabs, and can generate dashboards, mini-apps, and trackers on demand rather than returning a list of links. AI Mode now has 1 billion monthly users. Google also announced Gmail Live, a native voice integration powered by Gemini that lets users verbally query their inbox and get explicitly sourced answers.
Project Genie received a significant expansion, integrating with Street View to create editable, interactive simulations of real-world environments—useful for training robots in rare conditions before physical deployment, and apparently also capable of reskinning a neighborhood as "Stone Age" or "Ocean World."
On the hardware front, Google announced Android XR, a line of AI-powered smart glasses developed with Samsung and Qualcomm, with design partnerships at Warby Parker and Gentle Monster. The initial fall rollout will be audio-only, with hands-free camera access and Gemini integration compatible with both Android and iOS. A tethered display variant called Project Aura enables 3D Google Maps and multi-window apps. Google is directly targeting Meta's dominant position in wearable AI hardware.
Other I/O highlights include Antigravity 2.0, a standalone desktop app and SDK for orchestrating parallel coding agents; AI Studio for Android, which generates native Kotlin apps from text prompts; Docs Live, which converts spoken ideas into formatted documents in real time; and a Universal Cart shopping agent embedded in Search, YouTube, and Gmail.
---
Andrej Karpathy Joins Anthropic
In a significant talent shift, Andrej Karpathy—who coined the term "vibe coding," co-founded OpenAI, and previously led Tesla's Autopilot team—has joined Anthropic's pretraining team to head a new group focused on using Claude to accelerate pretraining research. Karpathy described the next few years in large language models as "especially formative," a phrase widely read as a signal that Anthropic is making meaningful internal progress on frontier model development. The move gives Anthropic one of the most recognized names in applied AI research and has been interpreted in some quarters as a sign that Anthropic is pulling ahead in the lab-level model race.
---
SpaceX Eyes Cursor Acquisition at $60 Billion
SpaceX is reportedly planning to acquire Cursor, the AI coding startup, just 30 days after its IPO—with SpaceX's own IPO expected to file today and shares listing on June 12. The deal carries a reported $10 billion breakup fee, suggesting both sides are treating it as a serious commitment rather than exploratory talks. Cursor has grown rapidly as developer adoption of AI-assisted coding accelerates, making it one of the most valuable pure-play AI coding tools in the market.
---
OpenAI: IPO, Compute Lock-Ins, and a YC Blitz
OpenAI is preparing to file IPO paperwork targeting a valuation of up to $1 trillion, following its legal victory over Elon Musk—whose lawsuit was dismissed by a jury that found he filed too late. Separately, OpenAI launched a "Guaranteed Capacity" tier offering one- to three-year compute commitments with discounts, with Sam Altman warning the world will remain "capacity-constrained for some time." The company also announced it would offer $2 million in API credits to each of the 199 companies in Y Combinator's Winter 2026 batch in exchange for 2% equity at $100 million valuations.
On the partnership front, AWS will begin reselling OpenAI models—including GPT and Codex—following Microsoft's loss of its exclusive license, opening new deployment options for enterprise cloud customers. OpenAI has also joined the C2PA content provenance standard and is embedding Google DeepMind's SynthID watermarking into all images generated via ChatGPT and its API.
---
Robotics Update
The robotics sector continued its busy week. Figure's F.03 humanoid robot completed seven consecutive days of fully autonomous package sorting without a failure. Boston Dynamics' Atlas demonstrated lifting a refrigerator. Tesla confirmed its Optimus robot is on track for limited production starting July or August 2026, with a major scale-up targeted for 2027.
A $15,000 humanoid robot kit aimed at researchers and enthusiasts made it to market, while a wall-climbing welding robot drew attention for demonstrating that physical autonomy is catching up to the software progress seen over the past few years. South Korea is exploring Hyundai's Spot and other wearable exoskeleton platforms to offset military manpower shortages.
---
Enterprise AI and Big Deals
Meta announced it is laying off 8,000 employees while simultaneously shifting 7,000 others into AI engineering roles, framing the restructuring as a long-term investment in AI infrastructure. Ford, GM, and Stellantis together cut more than 20,000 white-collar jobs driven by AI and software-driven automation, even as 400 AI-specific roles remain open across the three automakers.
Anthropic rolled out Claude to KPMG's 276,000 employees and their client-facing Digital Gateway platform, in one of the larger enterprise deployments announced this year. Cloudflare and Anthropic also integrated Claude Managed Agents, allowing developers to run AI agents at scale using lightweight V8 isolates for faster and cheaper deployment.
SandboxAQ integrated its physics-based drug discovery models directly into Claude, making molecular simulation accessible without specialized computing expertise. Google DeepMind published its Co-Scientist system in *Nature*, showing a multi-agent research platform that generated and lab-tested biomedical hypotheses autonomously.
On the M&A side: AWS acquired generative AI media startup fal to anchor its creative cloud pipelines; Google DeepMind paid $90 million to acquihire staff from Contextual AI; Mistral AI acquired Emmi AI to build physics models for heavy industry; and Analog Devices announced a $1.5 billion acquisition of Empower Semiconductor to meet surging data center chip demand. Google and Blackstone also formalized a $25 billion joint venture targeting 500 megawatts of TPU capacity by 2027.
Alibaba reported AI revenue of $1.32 billion for the quarter—its eleventh consecutive period of triple-digit growth—though heavy AI spending pushed profits into negative territory.
---
Security
GitHub disclosed an internal breach involving unauthorized access to its internal repositories, though the company said it found no evidence of impact to external customer data. A separate supply chain attack dubbed "Mini Shai-Hulud" poisoned over 600 npm and PyPI packages with a credential worm that deletes a developer's home directory if stolen GitHub tokens are revoked. Google, meanwhile, reported the first observed instance of an AI-built zero-day exploit in the wild—one that successfully bypassed multi-factor authentication—raising new concerns about the asymmetric advantage AI may be giving attackers.
---
Policy and Regulation
The U.S. government has demanded a 90-day advance preview of next-generation frontier AI models before public release. A coalition aligned with the MAGA political movement separately urged the Trump administration to mandate formal safety testing for advanced AI systems. Shield AI's Hivemind autonomy software was selected to power LUCAS, a new Pentagon program for low-cost coordinated attack drones; the system handles navigation and timing while humans retain authority over strike decisions.
---
Quick Takes
Cerebras launched enterprise trials for Kimi K2.6, achieving approximately 981 output tokens per second—roughly 6.7 times faster than the next-fastest GPU-based cloud provider, pushing past the 1,000 tokens-per-second threshold.
Coinbase reorganized its engineering teams into one-person pods managed by AI agents, a structural shift that drew wide attention for its implications on software team design.
Perplexity added Plaid and Polymarket integrations to power a new Personal CFO dashboard for financial tracking.
A user publicly reported recovering approximately $400,000 worth of Bitcoin locked for 11 years using Anthropic's Claude Code—a striking practical demonstration of AI-assisted cryptographic recovery.
Gemini now has 900 million regular users as of May 2026, roughly matching OpenAI's reported ChatGPT audience, according to Google.
xAI released Grok Build, a terminal-first coding agent that inspects repositories, drafts plans, and applies diffs; currently in early beta for SuperGrok Heavy subscribers.
A study found that four leading chatbots—ChatGPT, Claude, Gemini, and Grok—provide inaccurate answers when asked about elections and recent news topics.
AI will make its World Cup debut in 2026 for real-time tactical analysis, millisecond offside calls, and stadium digital twins.
Standard Chartered said it would cut more than 7,000 jobs over four years while using AI to replace what it called "lower-value human capital."
Mira Murati announced $100,000 grants aimed at addressing what she described as broken human-AI communication.