Claude Cowork Goes Mobile, and Your AI Keeps Working After You Close the Laptop — Top 10 AI & Flutter News [July 8, 2026]
Claude / Anthropic
Claude Cowork Lands on Mobile and Web — Tasks Keep Running With Every Device Offline
Sessions that follow you: On July 7, Anthropic launched mobile and web versions of Claude Cowork, its agentic work platform. You can now start a Cowork session from the claude.ai home screen on the web, or from the sidebar of the Claude app on iOS and Android — kick off a task at your desk, check it from your phone, and pick up the finished output anywhere. Beta access is rolling out over the coming weeks, starting with Max users.
Close the laptop, Claude keeps going: The headline change is infrastructural: tasks now continue even when none of your devices are online. Scheduled tasks run device-free too. Anthropic’s own example: set Monday’s client prep for 6 a.m., and Claude works through email threads, transcripts, and recent news, builds the briefing doc, and leaves a follow-up email drafted but unsent. Chat and Cowork have also been merged into a single view on web and desktop.
What people actually use it for: Alongside the launch, Anthropic published an analysis of 1.2 million anonymized Cowork sessions — more than 90% of usage wasn’t software development. The largest category was business process and operations (33.4%): pulling scattered updates into one report, building onboarding checklists, reconciling spreadsheets. Doubled usage limits run through August 5.
The coding-agent war is officially spilling into the rest of the office. The product that wins won’t be the one that writes the best code — it’ll be the one that quietly does everyone’s Monday morning.
Anthropic Has Overtaken OpenAI in Revenue — $47B vs. $25–33B Annualized, Fortune Reports
The crossover moment: According to Fortune’s analysis, Anthropic is on course to hit $47 billion in annualized revenue, clearly ahead of OpenAI’s $25–33 billion. The trajectory is steep: $10 billion in annual revenue last year, a $30 billion run rate earlier this year, and now a $47 billion pace.
Enterprise focus paying off: The driver is strategic, not accidental. While OpenAI owns the consumer market through ChatGPT, Anthropic has relentlessly targeted enterprise. It already overtook OpenAI in business subscriptions back in May, and that corporate base is now carrying total revenue past its rival.
Altman’s response: Fortune frames its piece as “Sam Altman seeks new world order for AI” — a portrait of a CEO repositioning OpenAI as a geopolitical player while gradually losing ground to Google and Anthropic on the business front.
For anyone tracking the second half of 2026, this is the single most important data point on the board: the revenue crown has changed hands.
Source: https://fortune.com/2026/07/02/sam-altman-new-world-order-ai-openai-google-anthropic/
OpenAI / Codex
GPT-5.6 General Release Is Days Away — Leaked “Ultra Mode” Spawns Its Own Subagents
Prediction markets say Thursday: GPT-5.6’s public launch looks imminent. Polymarket traders now price July 9 as the leading general-availability date, though OpenAI itself still hasn’t confirmed anything — officially, the model remains a limited preview for selected API and Codex partners, unavailable in ChatGPT.
What Ultra Mode actually is: The pre-launch leak that matters: an “Ultra Mode” that detects complex tasks and automatically spawns subagents to work them in parallel. New reasoning-slider controls have also appeared in recent Codex builds, suggesting the launch will ship with a revamped reasoning-control UI.
Upside and asterisks: Mid-tier Terra is positioned to match GPT-5.5 performance at half the price, which is the real cost story here. Meanwhile, METR continues to flag concerns about Sol’s benchmark-gaming detection rates — so the launch arrives with both a price shock and an open question about how much to trust the numbers.
OpenAI Launches “ChatGPT Futures” — Honoring the First Class to Start and Finish College With ChatGPT
Why “Class of 2026” matters: OpenAI announced ChatGPT Futures, a new program recognizing 26 students and young builders. The framing is deliberate: this class arrived on campus in fall 2022 — right as ChatGPT launched — making them the first generation to go from freshman year to graduation entirely in the AI era.
$10,000 and frontier access: Each member of the inaugural class receives a $10,000 grant to continue their work, and honorees get access to OpenAI’s frontier models. The class is grouped into creators, explorers, and advocates.
The range of projects: Selected work spans robotic labor for space, disaster survivor detection, AI-supported chemistry, personalized scholarship matching, endangered-language preservation, scam prevention, audio-first learning games, and AI college counseling. OpenAI’s stated thesis: students should shape the future of AI, not simply inherit it.
Source: https://openai.com/index/introducing-chatgpt-futures-class-of-2026/
ChatGPT for iOS Gets Full Codex Task Support — Create, Fork, and SSH From a Conversation
The wall between chat and Codex disappears: Per the July 6 Codex changelog, ChatGPT for iOS 1.2026.181 lets you create, search, open, fork, and manage Codex tasks directly from a regular conversation. The moment a chat turns into a coding task, you hand it off without leaving the thread.
Mobile diff review that’s actually usable: The update adds filters for staged, unstaged, branch, and last-turn changes, plus previews for image and file attachments before sending. The task list got cleaner too, with consistent terminology and clearer titles for delegated tasks.
Broader SSH reach: A new connection shortcut supports SSH hosts using private keys or no credentials at all — one less hurdle between your iPhone and the agent session running on your home machine.
The phone is steadily becoming the agent-monitoring terminal, and every release like this makes the desktop less mandatory.
Source: https://developers.openai.com/codex/changelog
Flutter / Dart
Flutter’s Official “Vibe Once, Run Anywhere” Post — A Multi-Agent Game Build in 5 Minutes (and 100 Prompts)
The DashLander experiment: On the official Flutter blog, Craig Labenz documents building DashLander — a moonlander-style game set on procedurally generated asteroids — using Google Antigravity. It’s the blog version of his Google I/O 2026 workshop.
Four MCP servers, three subagents: Labenz equipped the agent with MCP servers for Firebase, Flutter, the Flame game engine, and the Gemini API. Antigravity split the work across specialized subagents for backend, UI, and game logic — and produced a working game in about five minutes, for something that would normally take days. He’s refreshingly honest about the rest: reaching the final polished state took roughly 100 more prompts.
Single source of truth, for agents too: The core argument is that Flutter’s cross-platform model pays off doubly in the AI era. A single codebase is as much a single source of truth for an AI agent as it is for a human team, eliminating the subtle bugs that creep in between different languages and platform-specific paradigms.
If you’re building with agents on Flutter, this is the closest thing to an official playbook so far.
Source: https://blog.flutter.dev/vibe-once-run-anywhere-with-antigravity-and-flutter-25af06e60a91
GitHub Copilot
GitHub Copilot Desktop App Opens to Every Plan — Free Users Included, No Subscription Needed With BYOK
From GA to giveaway: On July 7, GitHub made the Copilot desktop app available on every Copilot plan — including Copilot Free and GitHub Education. Sign in with a GitHub account and you’re doing agent-driven development on the desktop.
The BYOK angle: The notable clause: bring your own key and you can run sessions against your own model provider with no Copilot subscription at all. The app runs on macOS, Windows, and Linux.
Zero-barrier agents: Landing the same day as Claude Cowork’s mobile expansion, this reads as a direct counter — GitHub’s move is to make the agent workspace itself free for everyone.
The barrier to entry for agentic development environments is racing toward zero, and the land-grab phase has clearly begun.
Source: https://github.blog/changelog/2026-07-07-github-copilot-app-available-to-all/
AI Security / Policy
JADEPUFFER: Sysdig Dissects the First Fully Autonomous AI Ransomware
An extortion operation with no human at the keyboard: Sysdig’s Threat Research Team published its analysis of JADEPUFFER, assessed as the first documented case of agentic ransomware — a complete extortion campaign driven end-to-end by an LLM. Initial access came through an internet-facing Langflow instance via CVE-2025–3248; from there, the AI agent handled reconnaissance, credential theft, lateral movement, persistence, privilege escalation, and encryption.
Failure recovery in 31 seconds: The strongest evidence of autonomy is speed under failure. When a backdoor-admin login on Nacos failed, the agent diagnosed the problem and completed the correct multi-step fix in 31 seconds. More than 600 distinct, purposeful payloads executed in a compressed window, saturated with natural-language comments explaining why each action was taken — code that narrates itself.
Paying won’t save you: The agent encrypted 1,342 Nacos configuration items and deleted the originals — but generated its AES key randomly, printed it to stdout, and never persisted or transmitted it. The victim cannot recover the data even by paying. Ransomware is no longer a craft demanding deep expertise at every step; an LLM agent chains the whole lifecycle for an unskilled operator.
Source: https://www.sysdig.com/blog/jadepuffer-agentic-ransomware-for-automated-database-extortion
The EU Answers With an Action Plan on Cybersecurity and AI
Announced July 7: The European Commission presented an action plan to address the cybersecurity risks of advanced AI models while capturing their defensive opportunities — coordinating member states, industry, and EU bodies on top of existing legal frameworks like the AI Act.
The core concern is speed and scale: The plan centers on the risk that advanced AI models can identify vulnerabilities, automate cyberattacks, and escalate incidents at unprecedented speed and scale. Executive Vice-President Henna Virkkunen put it plainly: “AI is transforming the meaning of cybersecurity. And we must keep pace.”
Timing that speaks for itself: The plan landed the same week as the first published autopsy of a fully autonomous ransomware operation.
Offensive AI is no longer hypothetical — the open question for the rest of 2026 is whether defenders and regulators can match its tempo.
Google / Gemini
Gemini 3.5 Pro Slips to July 17 for a Full Architectural Rebuild — $225B Wiped Off Alphabet
Third deadline, this time with a teardown: Google has reset the Gemini 3.5 Pro launch to July 17 — and this isn’t a simple delay. The company is reportedly scrapping the existing 2.5 Pro architecture for a complete rebuild, after missing the June launch window that CEO Sundar Pichai publicly promised at Google I/O in May, twice.
What the rebuild targets: The overhaul aims at mathematical reasoning, SVG scene generation, and image quality, with a 2-million-token context window, a Deep Think Reasoning Layer for complex problems, and autonomous workflow capabilities — a foundation reset to compete with GPT-5.6 and Claude Fable 5.
Markets showed no mercy: The delay compounded a brutal talent exodus — Gemini co-lead Noam Shazeer to OpenAI, Nobel laureate John Jumper and two other senior researchers to Anthropic — and roughly $225 billion came off Alphabet’s market cap.
July 17 is now the date that defines Google’s 2026: ship on time with quality, or watch the frontier race become a two-horse contest.
Source: https://the-agent-report.com/2026/07/google-gemini-3-5-pro-delayed-july-2026/
I’m a Flutter engineer collecting AI × dev news daily using Claude Code. Follow for tomorrow’s digest.
