Google's family AI, China's AI cash gap, a 100k-site scam, and 'AI building AI'
Google's new 'CC' agent wants to run your household, China's top AI labs earn a fraction of their US rivals, and a ClickFix scam is hitting 100,000+ sites — plus a Gmail storage trick and an honest look at Anthropic's 'Claude builds Claude' claim.
- Long Press · Daily Five19 Sept 202601 / 05
commentaryGoogle's new AI agent 'CC' is built to run your family's life
Google is testing 'CC', a new Gemini-powered AI agent built to run a family's household. It gets its own Google account and can read emails you forward it, plan meals, track tasks, fill school permission slips, and send a 'Your Day Ahead' brief every morning. It's genuinely useful — but it also means one more Google login that can see your entire household, and for now it's US-only, adults, and behind a waitlist.
The problem
Running a household means juggling schedules, forms, meals and reminders across email, calendars and half a dozen apps — and no single assistant has been trusted to actually do any of it for you.
What changed
Google is testing 'CC', a Gemini-powered agent that gets its own Google account and can coordinate up to six family members — reading emails you forward it, planning meals, tracking tasks, filling school permission slips and generating a 'Your Day Ahead' brief each morning. It runs in an isolated cloud environment on Google's Antigravity infrastructure and builds shared memory about household preferences.
The catch
It's an experiment: US-only, adults with personal Google accounts, waitlist for now. And it means one more login with standing access to your family's emails and documents — Google says CC asks for confirmation before sending anything outside the family group.
What it means for you
This is where consumer AI is heading — from one-off prompts to persistent agents with memory and permissions. Worth watching, but think hard about how much of your household you want a single agent to see before you join the waitlist.
- Long Press · Daily Five19 Sept 202602 / 05
newsChina's best AI earns about a tenth of OpenAI + Anthropic
China is building world-class AI models but making very little money from them. Seven of its top labs — ByteDance, Alibaba, DeepSeek, Moonshot, Z.ai, MiniMax and Kuaishou — pull in roughly $10.7B in combined annual recurring revenue, versus more than $100B for OpenAI and Anthropic alone. The models compete on benchmarks; the business doesn't — yet. Their strategy leans on open-weight releases and low prices, chasing reach over revenue, which for users means more powerful free AI.
The problem
Benchmark scores make headlines, but they don't reveal whether an AI lab can build a business — and China's models increasingly match the West on capability.
What changed
New Rhodium Group estimates (via SCMP) put seven leading Chinese developers — ByteDance, Alibaba, DeepSeek, Moonshot, Z.ai, MiniMax and Kuaishou — at roughly $10.7B in combined annual recurring revenue, with ByteDance around $4B and Alibaba about $2.4B.
The catch
ARR is a run-rate metric, not audited annual revenue, so treat it as a scale indicator. The gap is real though: OpenAI and Anthropic together report $100B+, helped by enterprise deals and premium subscriptions.
What it means for you
Two AI business models are emerging — one built for monetization, one for distribution. China's open-weight, low-price push means more capable free and open models you can actually use, even if the labs behind them earn far less.
- Long Press · Daily Five19 Sept 202603 / 05
tipFree up Google storage in seconds instead of paying for more
Before paying Google for more storage, clean out the biggest offenders. Open Gmail, click the search bar and type larger:10M older_than:2y (or has:attachment larger:25M), hit enter, then choose 'Select all conversations that match this search' and delete. Years of forgotten giant attachments clear in seconds, reclaiming free space instead of forcing a paid plan.
The problem
Google's free 15GB fills up fast — mostly with years of huge email attachments you'll never open again — and the default nudge is simply to pay for more storage.
What changed
Gmail's search operators do the cleanup for you: type larger:10M older_than:2y (or has:attachment larger:25M) in the search bar, then use 'Select all conversations that match this search' to bulk-delete the space hogs in one go.
The catch
Deleted mail sits in Trash for 30 days before it frees space, and this runs on the web. Skim the results first so you don't bin an important old attachment.
What it means for you
You can reclaim gigabytes in under a minute and skip a storage upgrade entirely — a genuine money-saver if you're bumping against the 15GB ceiling.
- Long Press · Daily Five19 Sept 202604 / 05
securityIf a website tells you to run a command, it's a scam
A scam is spreading across more than 100,000 websites this week. You land on a page that says 'verify you're human', then it tells you to copy a code and paste it into Windows Run or your Terminal — and doing so installs malware yourself. It's called ClickFix, and the rule that keeps you safe is simple: no legitimate website ever needs you to run a command. Close the tab.
The problem
Scammers have found a trick that sails past antivirus: instead of hacking your device, they convince you to run the malicious command yourself.
What changed
A supply-chain attack on marketing platform Brevo injected code into more than 100,000 websites this week, showing some visitors a fake 'verify you're human' page that instructs them to copy a command and paste it into Windows Run or Terminal — a technique called ClickFix.
The catch
It works because the page looks like a normal anti-bot check, and the malware runs the moment you paste and hit enter. Antivirus can't help if you launched it yourself.
What it means for you
Memorise one rule: no real website ever needs you to copy-paste a command into Run, Terminal or PowerShell. If a page asks, it's an attack — close the tab.
- Long Press · Daily Five19 Sept 202605 / 05
hot take'AI is building AI'? Anthropic's own numbers say: not yet
Anthropic says Claude now leads about 26% of the work of building its next models, up from roughly zero in February, with around 30,000 AI agents running research tasks. It's a striking trajectory — but the fine print matters: Anthropic says humans still supervise everything and the models are not autonomously designing or training their successors. Fast is not the same as autonomous, so watch the trend and ignore the 'AI is building itself' headlines.
The problem
'AI is building itself' makes for a terrifying headline — and it's easy for the claim to outrun what the evidence actually shows.
What changed
Anthropic disclosed that Claude led about 26% of its model R&D work by August, up from effectively zero in February, with roughly 30,000 agents doing research and engineering and around 90% of R&D now involving Claude in some form.
The catch
Anthropic is explicit that humans still supervise everything and the models are NOT autonomously designing or training their successors — '26% led' means tasks completed from a high-level prompt under human oversight, not independent research.
What it means for you
The honest read: the trajectory is genuinely fast and worth tracking, but 'fast' isn't 'autonomous.' Treat breathless 'AI builds AI' headlines with skepticism and watch the actual numbers instead.