Companies Can No Longer Ignore GEO: AI Shopping Agents and the New Search Era (+the Only Updates You Need to Know This Week)

AI Shopping Agents, ChatGPT Search, LlamaCon, Worldcoin Orb, Apple+Anthropic Vibe Coding Platform, ChatGPT 4o's Sycophancy, AI Target Minors, JetBrain's Mellum, DeepSeek Prover v3, Alibaba Qwen 3

Here are all the only AI updates you need to know for the week. Video version de-brief here:

Update 1: Companies Can No Longer Can Ignore GEO: AI Shopping Agents and the New Search Era

This week made one thing impossible to ignore: search is changing in ways we still fully can’t comprehend. I’ve been talking about generative engine optimization (GEO) for a while now, and if you’ve been with me, you’ll remember I covered the first research paper on GEO nearly a year and a half ago. But this week brought 4 major search and shopping updates, so we need to yet again revisit this conversation.

Let’s start with the shopping agent explosion. Visa, Mastercard and Paypal all announced AI shopping agents this week, where AI agents don’t just find products for you, they buy them too. Fully handling purchases based on your preset preferences and pushing forward this concept of “agentic commerce”, Amazon also announced their own version earlier this month.

OpenAI made major updates with ChatGPT Search, now recommending products to you and embedding direct shopping links right inside your chat. They’re pitching this as a rival to Google’s dominance in search, as users made over a billion searches inside ChatGPT just last week. That is an insane number and a massive signal of where user behavior is heading.

So this brings one more very critical inevitable question: now that SEO is dead, GEO is here, when will ads start being run in LLM environments? Sam Altman was super against ads in ChatGPT but is suddenly softening his stance, because, guest what? Google is already there, confirmed in Bloomberg this week that they’ve been quietly slipping ads into AI chatbots through its AdSense network.

They're testing with smaller AI search startups now, but we know how this plays out: if it works, it scales fast.

All of this ties back to something I’ve been hammering: generative engine optimization is the next frontier, and we still don’t have a roadmap. Everyone is scrambling to figure out how to be visible in these new AI environments, but there’s no playbook yet for how to rank or be found inside conversational systems. We’re watching the birth of a new ecosystem where AI agents make choices on your behalf, whether it’s what to read, where to eat, or what shoes to buy, and not only are traditional SEO techniques out the window, but they actually HURT you in generative engine environments. Keyword stuffing helpful in Search engine result pages? HURTS you by 10% in generative engine search environments.

So we are watching search, shopping, and advertising collapse into a single AI-powered loop but we don’t have the answer of how to handle this. How do you win in a world where the AI...not the user...is your new customer your trying to sell to?

Update 2: LlamaCon Recap: Meta's "New" ChatGPT Competitor Not Actually New, Plus Finally a Llama API

Last week two main updates came out of Meta AI’s LlamaCon (Meta’s AI showcase).

First, everyone freaked out on the internet thinking Meta just launched a ChatGPT competitor. Spoiler—they didn’t. At least, not just last week.

If you’ve been following me for awhile, you’d already know that Meta debuted their ChatGPT alternative a full year ago, complete with deep context awareness, personalized recommendations, and conversation memory on a website interface just like ChatGPT called meta.ai. Don’t believe me? Here’s my original video.

Don’t listen to just any old AI influencer folks. They clearly don’t do their research.

So, what did they actually launch that got the internet in a blaze? Well, they didn’t launch a ChatGPT competitor... because we have proven that existed a year ago. They launched... a mobile app. With voice mode. Noice.

The other major update is Meta finally launched a Llama API, letting developers build, fine-tune, and evaluate custom AI apps. Starting with Llama 3.3 8B, the API lets users generate training data, customize their models, and measure performance They say they won’t be training on your data for their own use (or locking into their platform).

How kind of meta, for making that statement. After taking probably the largest repository of private data in the history of the world from all social media, and already using it.

But now they’re not going to use it. Don’t worry.

Update 3: Worldcoin Launches Orb Mini to Expand Human Verification Push in AI Deepfake Age

Worldcoin, the human verification project co-founded by Sam Altman, launched the Orb Mini this week—a portable, smartphone-like device designed to scan your eyeball and confirm you're human. It’s a smaller version of the original Orb, those futuristic silver globes already in use around the world, both aiming to tackle a growing problem: as AI agents and deepfakes flood the internet, it’s becoming harder to tell who’s real. Worldcoin’s solution is a global “proof of personhood” system, assigning a unique blockchain-based ID to verified humans.

With deepfake tech only getting stronger, and 12 million people already scanned, the project is betting big that digital identity will be the next frontier of trust online. Of course the question remains: is your data truly safe, will Sam Altman at some point collaborate with his other venture OpenAI, or is your biometric information at risk? The company states that biometric data is encrypted and not stored centrally, but they have faced regulatory action in several countries, including Spain, Kenya, and Hong Kong.

I feel this project is very important for you all to keep your eye on because the rising issue of what is real versus fake is absolutely only going to get worse. Now you all know I am very cautious with data privacy (I once was detained at Canadian immigration for refusing to go through a facial scanner, true story).

A close family friend is the Chief Product Officer at Worldcoin. And when i’ve had a few conversations with him about this, he constantly reassures me it’s only locally stored. He’s a super honest guy, so, while I have not personally gotten my eyeballs scanned, my brother has, lot of my family has, and I will continue to watch the development of this project with caution but curiosity. Someone has got to fix the problem.

Update 4: ChatGPT’s Desperation to Agree with Absolutely Everything Caused a PR Nightmare

ChatGPT’s obsession with flattery and being a “helpful assistant” finally blew up in OpenAI’s face. Last weekend OpenAI rolled out an update to GPT-4o that immediately caused a disturbing shift felt by users. The chatbot became overly flattering, responding with excessive praise and unsettling sycophancy, enthusiastically validating even dangerous or absurd user suggestions. Social media exploded with memes, screenshots, and jokes about ChatGPT’s sudden desperation to agree with absolutely anything.

It forced OpenAI to roll back the release (which in tech terms, that means they went back to an older code version because there’s something wrong with the current one). CEO Sam Altman said the failure came from the fact they had relied too heavily on short-term feedback without considering how user interactions have evolved for deeper levels of trust and advice. They promised new safeguards and a more cautious deployment strategy, including a dedicated "alpha testing phase", to ensure ChatGPT doesn’t slip back into creepy, uncomfortable over-validation again.

I actually have stopped using ChatGPT as my primary LLM because of this exact problem, and switched to Anthropic’s Claude. ChatGPT is programmed to be a "helpful assistant", and it shows. It doesn't try to push a better output out of you... it does exactly what you tell it to. Claude tells me exactly what I'm doing wrong, and what it would change.

Update 5: SPEED ROUND

CODING, FOR MODELS, AND AI ETHICS OF AI TARGETING MINORS

1️⃣ Coding Updates

Some solid code updates this week for us dev nerds out there. According to Bloomberg Apple and Anthropic are reportedly teaming up to build an AI vibe coding platform, powered by Anthropic’s Claude Sonnet model. I hate the term vibe coding (I wanted to throw up even saying it), but that’s the wording they used. I guess it’s too late for them.

It’s designed to help engineers write, edit, and test code using generative AI. For now, it’s meant for internal use only, but Apple hasn’t ruled out a public release. This follows Apple’s own bumpy attempts at launching a similar platform last year called Swift Assist (which never actually made it to prod), Apple is leaning more heavily on outside AI muscle to boost its developer tools.

Meanwhile, JetBrains, the IDE tool developer, released its first open AI coding model, Mellum, a 4-billion-parameter system trained on over 4 trillion tokens, built specifically for code completion tasks. Mellum is available under an Apache 2 license, making it a fresh option for developers looking to integrate AI-powered code assistance without getting locked into proprietary platforms.

2️⃣ AI + Kids: The Ethics Issue: AI Race Now Targeting Minors

OpenAI confirmed this week that a bug in ChatGPT let users under 18 generate and access erotic content. In some cases, the bot even encouraged them to ask for more explicit content. OpenAI says this violated their own policies and that a fix is rolling out, stressing that protecting younger users is a top priority.

At the same time, Google announced this week that it will make its Gemini chatbot available to kids under 13 through parent-managed accounts, reinforcing the growing push to embed AI tools into younger users’ lives. Google says Gemini has specific safeguards and won’t use kids’ data for training.

In both cases, it’s clear the AI race is now targeting the next generation of users, despite ongoing warnings from regulators and groups like UNESCO calling for stricter age limits, privacy protections, and oversight. This is an ethics issue that keeps getting brushed aside.

I have sat in coffee shops listening to college students complain that ChatGPT is too slow to finish their homework on time, and elementary school kids are already using LLMs to solve their math problems. We are raising a generation that’s outsourcing core skills, problem-solving, critical thinking, even basic research, to a machine that’s always on.

What does that mean long term? Do textbooks become relics? Does education evolve into something richer and more dynamic, or do we end up with a generation that has the illusion of knowledge but little true depth? We need to ask harder questions: what happens when your first instinct isn’t to figure something out, but to prompt an AI to do it for you?

3️⃣ Model Releases

AMAZON NOVA PREMIER

Amazon released Nova Premier, the latest addition to their Nova AI family. Keep in mind Amazon just launched their first family of models Nova only 5 months ago, so they’re relatively new to the scene.

According to Amazon, Nova Premier shines at tackling intricate tasks requiring deep contextual understanding, careful multi-step planning, and precision across multiple tools and data sources. Premier processes text, images, and videos—but notably skips audio. It’s now available on Amazon Bedrock,

DEEPSEEK PROVER V2

China's DeepSeek launched version 2 of its math-focused AI model Prover on Hugging Face last week. It’s powered by their huge 671-billion-parameter MoE-based V3 model, designed specifically to break complex math proofs into smaller, expert-handled subtasks.

ALIBABA QWEN3

Lastly, Alibaba released the Qwen3 multilingual AI model family, claimed to rival—and sometimes surpass—top models from Google and OpenAI. It ranges widely in size and trained on a massive dataset spanning 36 trillion tokens and 119 languages.

Nothing particularly groundbreaking here, just making sure you stay ahead of the game. But Alibaba Qwen3 is seen as a sleeper cell among AI/ML leaders. Not getting a lot of press by the mainstream media. Regardless, DeepSeek and Alibaba are both chinese companies and these releases are putting pressure on US policymakers to tighten restrictions on China’s access to high end AI chips.