DeepSeek Takes on OpenAI and Anthropic
The scarcest resource in AI coding isn’t compute anymore. It’s you.
- What happened: DeepSeek’s Harness team posted a design paper on October 31 alongside an open-source GitHub prototype of a “super agent.” Rather than tying an agent to one conversation, it splits the system into a Router layer and multiple Worker sessions. It also watches developers’ group chats for coding problems and suggests fixes without being summoned.
- Why it matters: DeepSeek estimates a person can realistically oversee only three to five active agent sessions at once. That ceiling quietly caps every multi-agent productivity pitch, Claude Code and Codex included.
- What’s next: Don’t expect a launch. The repo is labeled research reference, not an official DeepSeek product, with no pricing and a public version that differs from DeepSeek’s internal build. The money is there for a real push, though, after DeepSeek raised around $15 billion in its latest round.
A January study found coding agents working as a team succeeded 30% less often than when each worked alone.
For builders this week: count how many agent sessions you actually supervise, and put a routing layer or a single review queue in place before you spin up another one.
AI Is Helping Meta Outpace Google in Ads
Google had record search traffic last quarter and still lost the ad race.
- What happened: Meta’s ad revenue jumped 26% in Q3 to $63.1 billion. Google’s Search unit grew 12% to $63.4 billion, the widest growth gap in Meta’s favor since 2021, leaving the two nearly equal in size.
- Why it matters: Meta’s AI ad tools are converting. Almost 10 million advertisers used one, and those following Meta AI’s recommendations saw roughly 5% more conversions while raising ad spend 9%. Google’s own AI Overviews answer questions on the page, so record query volume isn’t turning into proportional ad clicks.
- What’s next: The spending war keeps escalating, with Google projecting up to $205 billion in capex this year against Meta’s $145 billion. Meta carries more risk on that bet since ads still make up 98% of its sales.
Search was Google’s moat for two decades. Chatbots just found the drawbridge.
For builders this week: stop optimizing only for Google rankings and test how ChatGPT and Claude describe your product, because that answer is becoming your new first page.
Apple Brings Alibaba’s Qwen AI to iPhones
Every iPhone in China is about to run on a Chinese brain.
- What happened: Alibaba says Apple will build its Qwen model into Apple Intelligence for China. Beta testing starts November 4, with a full rollout planned alongside iOS 27.3 in December. It reaches deep into the system, from Siri to Visual Intelligence, and switches on only when both the phone’s location and the Apple Account region are set to China.
- Why it matters: A Beijing AI researcher calls it the first time a major global hardware maker has adopted a Chinese large language model as its main system-level AI in that market. Apple gets past the wall that has kept its AI features out of China. Alibaba gets a default spot on one of the most valuable phone fleets in the country.
- What’s next: Alibaba says users can opt out and that their requests won’t be used to train its models. December’s release will show whether Chinese iPhone owners actually use it or stick with the AI apps they already have.
If Apple will hand Siri to a different model in its second-biggest market, how long before it does the same elsewhere?
For builders this week: stop hard-coding one model provider into your product and add a routing layer, because even Apple now picks a different model for each region.
ChatGPT called this cancer harmless
A dermatologist tested GPT-6 Astra and Claude Fable 5.1 on 200 real skin spots, and both got fewer than half right. What caught my attention was the cancer they both waved through.
- Both said a melanoma patient could be sent home. They called it a harmless growth. The photo they misread is in the article.
- A 10-year-old AI beat them both. It needed only about 20 minutes of extra training, and it wasn’t close.
- You can try that old AI on your phone. It runs in your browser, and nothing gets uploaded.
It’s one doctor’s test, not a clinical trial. A changing mole still needs a doctor, but I’d check which cancers fooled the chatbots.
Microsoft offers 60,000 staff a way out
Microsoft has offered voluntary buyouts to about 60,000 employees, roughly 27% of its staff, as AI spending climbs. What caught my attention is where: the US, UK, Canada and India.
- Microsoft calls it “new for us.” It comes just six months after a voluntary retirement program.
- One line in Microsoft’s statement stands out. It hints at why the company wants this now, and Windows Central quotes it in full.
- Governments are doing the same math. The Dutch government plans about 25,000 fewer jobs over several years, with AI as one tool.
Not all 60,000 will leave, and Windows Central is already asking if layoffs are next. If your work is mostly routine, I’d read that line closely.
Google beat ChatGPT and Claude with an AI you can’t use yet
Google says Gemini 4 Carbon is its strongest model yet, and it just took first place on a popular AI leaderboard. What caught my attention was in Google’s own results table.
- It knocked Claude Opus 5.5 off the top spot. It scored 59 on the Artificial Analysis Intelligence Index.
- It won 15 of the 17 tests Google listed. The 2 it lost went to rivals you probably already use.
- You can’t use it yet. Only a small group of security teams has access, and there’s no date for the Gemini app.
Google’s best AI still lost to a chatbot you can use today, on a task you probably do every day.
Nasdaq loses $2 trillion in October
The Nasdaq 100 shed more than $2 trillion in market value in October, sliding over 5.8% by the 30th, as investors questioned whether Big Tech’s AI spending can pay off fast enough.
Meta, Alphabet and Amazon all raised their AI spending forecasts again. Alphabet posted its worst day ever, dropping about 13% on weak search ad sales, while Meta’s $97.5 billion stock-sale plan sparked dilution fears.
Microsoft bucked the gloom with results showing its AI spending is starting to pay off. With JPMorgan forecasting $5.5 trillion in global AI spending by 2030, investors now want receipts, not roadmaps.
Apple CEO says it won’t ship AI for AI’s sake
In an October 30 interview with Wired, new Apple CEO John Ternus said the company doesn’t “want to ship AI for AI’s sake,” pushing back on claims that Apple has fallen behind rivals in AI.
Ternus took over as CEO on September 1, after years of criticism over Apple’s delayed Siri upgrade. He says Apple will “absolutely” use AI, but the goal is products people love, not AI for show.
Asked whether AI agents might make apps obsolete, Ternus called building products “deeply human.” Still, Apple’s revamped Siri outsources much of its brainpower to Google’s Gemini, so the AI part isn’t exactly homegrown.
OpenAI reportedly cuts funding target to $20B
OpenAI reportedly slashed its latest funding target from $30 billion to $20 billion ahead of its planned IPO, after earlier reports suggested valuations ranging from $1.25 trillion to $1.4 trillion.
The Wall Street Journal says OpenAI has secured just $11 billion so far, while UAE investors will lead the round. A later offering targeting wealthy individuals could help close the gap.
The timing looks rough as investors grow wary of massive AI infrastructure spending, triggering a tech stock selloff. Even OpenAI seems to be discovering that AI hype doesn’t guarantee investor enthusiasm.
Judge allows pollution lawsuit against xAI
A federal judge rejected xAI’s bid to dismiss an NAACP Clean Air Act lawsuit accusing it of running dozens of unpermitted gas turbines in Southaven, Mississippi, to power its Colossus 2 data center.
xAI argued the turbines count as exempt “mobile sources” and that citizen lawsuits like this one are unconstitutional. Judge Debra Brown rejected every argument, shortly after denying a similar dismissal bid from the Justice Department.
Déjà vu for xAI: it removed unpermitted turbines from its first Colossus site after an NAACP legal threat in 2025. Now the same groups are back, just one state line south.
China’s next archaeologists: AI and robots
China’s 2026–2030 archaeology plan will combine AI, ancient DNA analysis, and deep-sea robots to reconstruct artifacts, investigate shipwrecks, and trace human migrations across East Asia and beyond.
Researchers plan to train AI to match broken artifacts, deploy underwater robots along historic trade routes, and study ancient populations, while China develops new archaeological laboratories and international partnerships.
The blueprint also aims to strengthen China’s cultural influence worldwide. But even deep-sea robots can’t manufacture missing evidence; the real test is whether researchers uncover findings that hold up to scrutiny.
AI-Related Jobs Are Surging Across the U.S.
AI-related roles have accounted for more than 750,000 new U.S. jobs since 2023, according to LinkedIn estimates. Data annotators led with 282,000 positions, followed by data center jobs (117,000) and AI engineers (105,000).
These three categories alone account for approximately 504,000 new jobs.
AI-related positions offer a median salary of around $180,000, compared with $80,000 across all jobs, highlighting the growing wage premium associated with AI skills.
AI Giants Back a Shared AI Scraping Opt-Out Standard
OpenAI, Google, Microsoft, and Anthropic plan to support a new standard that lets website publishers set their AI training preferences across multiple companies through a single control.
- The proposed standard introduces a shared “policy token” that publishers can add to their website’s robots.txt file, which already tells web crawlers which parts of a site they can access.
- The proposal was submitted to the Internet Engineering Task Force (IETF) by Kevin Kelley, a technical lead for standards at Anthropic.
- Currently, publishers must communicate their AI training preferences separately to each AI company. The shared signal is intended to make that process more consistent.
- OpenAI, Google, Microsoft, and Anthropic say their AI crawlers will respect the signal and restrict access to websites that use it.
- Microsoft will also offer more specific controls, allowing publishers to opt out of AI training or AI grounding, which involves using verified sources to support AI-generated answers. Websites will still appear in Bing search results.
The proposal comes as publishers seek more control over how AI companies collect and use their content. Anthropic has already adopted the signal. Google and Microsoft plan to support it next year, while OpenAI expects to implement it in the coming months.