Cover: At Anthropic, junior tech hires fell from 27% to 9%.
Their own data: Entry-level tech hires fell from 27% to 9%. Photo: Anthropic’s chart. Blue line: junior share of hires.
Still growing: The company tripled in size in a year.
Agents vs people: 6 to 1 AI agent hours to employee hours. Photo: Anthropic’s chart of agent hours per human work hour.
More output per person: 5.1x code commits per engineer. Photo: Anthropic’s chart. Output per person, December 2025 = 100.
A new kind of work: Managing AI agents now fills 22% of staff’s AI sessions. Photo: Anthropic’s chart of time spent managing agents.
In their words: "Employment protections for incumbents will not benefit those not hired to begin with." Anthropic’s report, on the drop in junior hiring.
If you’re starting out: hiring now looks for a track record with AI tools, judgment and working on your own, and skill at managing agents, not just using them.
End slide: follow VVN, with sources and image credits.
Elias MwewaElias Mwewa
Cover: Jensen Huang: AI will not replace your capability.
Where: Nvidia’s CEO opened Germany’s Invest in Germany summit with Chancellor Merz. Photo: Merz and Huang on stage in Berlin, Oct. 19.
His message: "AI can only augment your capability, it will not replace your capability." Jensen Huang, Nvidia CEO.
On factories: "Electrons will never replace atoms." Jensen Huang, telling manufacturers that digital tech won’t replace making real things (AP).
The chancellor: "Please open your doors for AI to make use of your data." Friedrich Merz, German chancellor, to business leaders (Reuters).
What Nvidia announced: Plans for Germany. A declaration with the government: to work on AI models for robots and factories; Siemens: a partnership on AI-run factories.
The big idea: “The next wave of AI is physical.” Photo: Agile ONE, a humanoid robot from Agile Robots.
End slide: follow VVN, with sources and image credits.
Valley View NetworksValley View Networks
Cover: A meeting notes app just raised $320 million without selling a share.
What happened: Otter.ai raised $320 million from General Catalyst. Photo: Otter’s announcement, Oct. 20.
The business: $100M a year in recurring revenue. Photo: Otter’s AI agent joining a video call.
Users: Up 60% in a year. A year ago 25M, Now 40M.
The twist: Otter didn’t give up any shares to get the money.
Why it sells: "Meetings are terribly inefficient." Alex Gay, Otter’s chief marketing officer, to TechCrunch. Photo: A live Otter transcript.
The money lesson: pick a task everyone hates (Otter picked meeting notes), make the AI do the boring part, and give it time. Otter has been around for nearly a decade.
What’s next: The money goes to big companies and new countries. Photo: Otter’s app answering a question about a meeting.
End slide: follow VVN, with sources and image credits.
Elias MwewaElias Mwewa
Cover: Bill Gross is giving away 1.2 million AI-made startup ideas.
Who: Bill Gross founded Idealab in 1996. It has launched 175+ companies. Photo: From Idealab.ai’s launch video.
What launched: Idealab.ai uses AI to turn the news into business ideas. Photo: Idealab.ai’s homepage.
By the numbers: 1.2M business ideas, from more than 17 million sources. Photo: Idealab.ai’s live idea counter.
How it finds ideas: Each idea mixes three things. A problem: An unmet need or a point of friction; An enabler: New technology or falling costs; A shift: A change in rules, the economy or culture.
Every idea comes with a business brief, potential customers, competitors, first steps and reasons it might fail.
The offer: Cloudflare is giving $100,000 in credits to companies started through it. Photo: The offer on Idealab.ai.
The big goal: 1 million new businesses. 100 million new jobs. Photo: Bill Gross.
In his words: "The human is still making all the decisions." Bill Gross, to Fast Company. Photo: Idealab.ai’s idea search.
The catch: An idea is only a small part of building a business. Photo: An Idealab.ai idea brief.
End slide: read the full story on VVN, with sources and image credits.
Elias MwewaElias Mwewa
Cover: AI chipmaker Tenstorrent raises $1.5 billion.
The deal: $6.2B valuation after its $1.5 billion Series E. Photo: A Tenstorrent card, unboxed.
How fast it grew: Valuation after each raise. 2021 $1B, 2024 $2.6B+, 2026 $6.2B.
What it sells: AI chips and servers, from one card to whole racks. Photo: Tenstorrent’s Blackhole p150a card.
The other business: It licenses its chip designs to companies like Samsung and LG. Photo: LG CEO William Cho with Jim Keller.
Where the money goes: Mass production, more Galaxy servers and better chip designs. Photo: A Tenstorrent Galaxy server.
The pitch: "We built one architecture and sell it at every size." Jim Keller, Tenstorrent CEO.
Who runs it: Jim Keller designed chips for Apple, AMD and Tesla. Photo: AMD’s Zen chip, which Keller helped design.
Why it matters: Investors keep betting on alternatives to Nvidia. Photo: Nvidia headquarters, Santa Clara.
End slide: read the full story on VVN, with sources and image credits.
Valley View NetworksValley View Networks
Cover: DHS orders DeepSeek out of federal civilian agencies.
What happened: Civilian federal agencies can’t buy DeepSeek anymore, and must remove what they have. Photo: DHS Secretary Markwayne Mullin.
It goes further: Tools built using DeepSeek have to go too. Photo: DeepSeek’s own list of tools built on it.
Who it covers: Covers: Federal civilian agencies, What they buy, Their IT systems. Doesn’t cover: You, at home, Companies’ own use, National security systems.
Any exceptions?: Agencies can ask for one with a compelling justification. Photo: DHS headquarters, Washington.
The reasons: DHS points to three risks. DeepSeek is Chinese, and Chinese law allows broad government access; Staff prompts could send sensitive data to China.
The law: It uses a 2018 law that lets officials block and remove risky tech. Photo: The U.S. Capitol.
Not the first: NASA, the Navy and the House already blocked DeepSeek in 2025. Photo: The Pentagon.
If you sell to government: Check whether DeepSeek is inside your product. Photo: DeepSeek’s guide for developers.
End slide: read the full story on VVN, with sources and image credits.
Elias MwewaElias Mwewa
Cover: Meta’s AI agent gave a Marketplace buyer a seller’s address.
What happened: A Toronto tech YouTuber let Meta’s Muse agent run his Marketplace listings. Photo: Matt Robb, in his video about it.
The setting: Muse asked: Allow One Time or Allow Always? Photo: A Muse approval prompt (Meta’s example).
Sept. 26, from his account: 5:27 PM Sends his street address and agrees to C$10 for a C$15 keyboard; ABOUT 9:15 PM The buyer arrives and sends a photo of the door.
What he said: "Just found out it told people my address and agreed a lowball price" Matt Robb, on Threads.
Why it didn’t ask: His address was in an auto-reply template he had approved. Photo: Robb’s Threads post, buyer’s name blurred.
Meta’s side: Meta: “no breach of privacy controls.” Photo: Meta headquarters, Menlo Park.
What Muse is: Muse can send emails, fill out forms and negotiate for you. Photo: Muse filling out a form (Meta’s example).
Before you choose Allow Always, check three things: can it send messages or accept offers for you, can it share your address or personal details, and can you see its activity log and switch it off?
End slide: read the full story on VVN, with sources and image credits.
Elias MwewaElias Mwewa
Cover: How one hacker used AI agents to break into 100+ company websites.
What happened: A Chinese-speaking hacker aimed AI agents at 1,200+ websites.
Who was hit: Deere (fixed it, no sign sensitive data was accessed), SharkNinja (a product-support site, threat contained) and Cummins (investigating a site a vendor runs).
The tool: Cairn is a free, open-source tool that runs teams of AI agents. Photo: A Cairn control panel.
How it worked: software fed online shops to the agents, the agents shared findings on a board and planned next steps, aiming to run code and plant a hidden upload form.
The cost: $5,000 in AI usage for the whole run. Photo: One attack, mapped from start to goal.
What the agents reported: 134 attack projects claimed a result. Ran code on the server 56, Got into a database 28, Took over accounts 20, Crypto wallet data 13, Payment card data 9.
How it was found: The attackers left their control panels open online. Photo: An exposed attacker task, redacted by NorForge.
Who’s behind it: NorForge saw no obvious state backing. The groups appeared to be after money.
Why it matters: "It’s not script kiddie stuff." Snorre Fagerland, NorForge, to Reuters. He had assumed hacking agents still needed some supervision. Photo: A break-in NorForge verified.
End slide: read the full story on VVN, with sources and image credits.
Valley View NetworksValley View Networks
Cover: People are using AI to make their smart TVs dumb again, over a Google TV home screen
Before and after: a TV home screen showing a sponsored ad, then a clean launcher, with the developer's post about Claude Code removing 25 Xiaomi apps
CleanYourTV project page: 13k GitHub stars, 147 TV models, 49 brands, 21 languages
The catch: one person bricked their TV and another got stuck halfway on a Fire TV
Consumer Reports guide, How to Turn Off Smart TV Snooping Features, and the tip to turn off ACR
Menu paths to turn off tracking on Samsung, LG and Roku TVs
Menu paths to turn off tracking on Fire TV, Vizio and Sony TVs
CleanYourTV's ad-free home screen, with the note that you'll still see ads, just less targeted ones
End slide: save this for later and follow VVN, with sources and image credits
Elias MwewaElias Mwewa
Valley View NetworksValley View Networks
Elias MwewaElias Mwewa
Elias MwewaElias Mwewa
Elias MwewaElias Mwewa