- AI with Kyle
- Posts
- AI with Kyle Daily Update 121
AI with Kyle Daily Update 121
Today in AI: Claude pulls up the ladder
What’s happening in the world of AI:
Highlights
News is back. The holiday lull is officially over!
🔒 Anthropic Pulls Up the Ladder on Claude Code
Claude Code is the best coding agent right now. Bar none. I know some people will argue for Codex or whatever, but 95% of the community agrees. And that's created a problem.
Competitors have been using Claude Code to build their own products AND third-party tools have been letting users route their $200/month Claude subscriptions through alternative interfaces.
Anthropic has had enough. Two things specifically happened this week:
1. Third-party tools blocked. Tools like OpenCode had a workaround letting users connect their Claude subscriptions to access the API at effectively subsidised rates. The $200/month subscription is absurdly good value - if you paid per token via the API, the same usage would cost thousands. Anthropic shut that down and annoyed a lot of people.
2. xAI (Grok) cut off. This one's embarrassing for Elon. Remember when he said his team uses Grok internally because it's the best coding model on earth? Yeah, they don't. That was a fib. Turns out xAI's engineers were using Claude Code through Cursor until Anthropic pulled access this week.
The internal memo from xAI's co-founder Tony Wu is hilarious: "Hi team. I believe many of you have already discovered that Anthropic models are not responding on Cursor... We will get hit on productivity. But it really pushes us to develop our own coding product." He tries to spin it as an opportunity for them to build their own Claude Code…
So much for Grok being the best.
Hilariously, OpenAI jumped in immediately: "With the Claude Code shutdown, I'm proud we build Codex in the open... we are 100% invested in supporting a flourishing ecosystem of agentic coding tools."

Kyle's take: Remember it’s very easy to champion openness when you're behind!c When you're on top, things change. All of this drama exists only because Claude Code is the best. That's the real story here.
Sources: Reddit - Anthropic blocks third-party use | GitHub - Open Code issue | Reddit - XAI access cut off | Kylie Bytes on X | Reddit - OpenAI response
🍋 Launch Lemonade Partnership
This is something cool! I'm officially working with Cien Solon on Launch Lemonade.
The problem I keep seeing: there's a massive amount of expertise locked in the heads of normal people - lawyers, accountants, social workers, engineers. They know they could probably build something useful with AI. But when I say "just install Claude Code," that's too much. Too confusing. I get it!
Launch Lemonade makes building AI agents and workflows genuinely simple. No code. You can build tools for internal use or package them up and sell access to others. It hooks into all the major models - Claude, Gemini, ChatGPT.
This fits with my mission of getting 1 million people AI ready by 2030. Shen has built the tool. I'm working on the education. Expect to see a lot more content about this on TikTok and Instagram!
Free account here https://aiwithkyle.com/launchlemonade
🧠 Shopify CEO Builds MRI Viewer in Minutes
When you get an annual MRI scan, they give you a USB stick with the data. Problem: you need expensive commercial Windows software to view it. So for most people, that data is basically useless.
Tobi (Shopify CEO) gave the data to Claude Code and said "make me a viewer." What he got back looked better, felt better, and was functionally better than the archaic commercial software sold to hospitals…
Kyle’s take: Does this mean that MRI software is dead and that the CEO of Shopify just killed the industry? Nope. This is more interesting in terms of what it means for us on the day to day.
Your imagination is the only limitation now. Any idea you have, you can just build it. Genuinely, it’s insane.
I did something similar with my genome data - turned 5.5MB of nucleotides into visualisations. Also built a public speaking coach that analyses my body language, eye contact, filler words, and speaking pace from uploaded videos.
What can you build for yourself, your community, your audience, your customers? We’re free to just go ahead and do the damn thing now.
Source: Tobi on X
👨💻 Linus Torvalds is Vibe Coding Now
Speaking of vibe coding let's get this one straight because people are blowing it out of proportion.
Linus Torvalds - creator of Linux, one of the most important software developers in history - has started vibe coding. That’s awesome. This guy is basically a demi-god in the world of coding.
But not on the Linux kernel. He's building audio software for his guitar pedals. It's a hobby project. For fun.
That said…his commit message was "Is this much better than I could do by hand? Sure is."
So no, the Linux kernel is not being vibe coded. But one of the greatest programmers alive is using AI to write code that's better than what he could write manually. That should still ring alarm bells for anyone dismissing vibe coding as "not real programming."
Sources: GitHub commit | Kimmonismus on X
⛰️ AI Drones for Mountain Rescue
Bittersweet story. A 66-year-old man went missing in the Italian Alps. Drones were sent up and captured thousands of images across a massive area - far too much for humans to sort through quickly.
AI pattern detection was used to identify where the person might be. They found him. Unfortunately, too late.

Nicola Ivaldo's remains were eventually found partially obscured by snow in this gully after the AI spotted his red helmet
The technology works. We just need it to work faster and be deployed more widely. But this is a reminder that we can use AI for genuinely positive things. Amidst all the stories about deepfakes and undressing apps, AI could also save lives.
Source: BBC Future
📚 New Resource: The AI Canon
The AI Canon from a16z is one of the best learning resources out there - foundational papers, courses, explainers. Problem is, their page is a bit impenetrable.
So I rebuilt it.
Pulled all the resources into one place, organised by difficulty. You can track your progress through the material. And I've added "Learn with AI" buttons that pre-fill prompts to explain articles simply, quiz you on the content, or put ChatGPT in tutor mode.
Kyle's response: Depends on your level. If you want to build from scratch and increase your technical expertise, start with Hugging Face's small agents documentation. They have great resources on building agentic systems. DeepLearning.ai also has a good course called "Building Code Agents with Hugging Face Smolagents."
If you want a non-technical approach - build agents without worrying about libraries and code - use Launch Lemonade. You can build, deploy, and sell agents without any technical background. It hooks into all the major models and handles payments via Stripe.
Two paths. Pick based on whether you want to learn the technical side or just get something built.
Sources: Hugging Face | Deep Learning.ai - Smolagents Course | Launch Lemonade
Member Question: "I use ChatGPT between my doctor's appointments with my annual MRIs. It understands my condition and helps me track changes."
Kyle's response: This is exactly why ChatGPT Health is interesting. Most of us have medical data scattered across different doctors and clinics. ChatGPT Health opens up the possibility of a holistic view - MRI scans, blood tests, daily conversations about how you're feeling. It can find patterns that a once-a-year doctor visit never would.
Are there risks in giving OpenAI your private medical data? Absolutely. But there are also opportunities. This is going to be a personal decision for everyone.
Member Question from Lyra: "It was you who promoted me to use Lovable, and now I have two working projects!"
Kyle's response: This is why I keep “bullying” people to just try it! There's a huge barrier between "I could probably build something cool" and actually doing it. Tools like Lovable eliminate that barrier. You don't need to understand gradient descent or back propagation. You just need to know how to describe what you want.
I taught a charity recently that was terrified of AI because some guy with two PhDs had come in and talked about neural network architecture. They thought AI required a maths degree. I showed them how to plan a kid's birthday party in ChatGPT. Suddenly their imaginations took them the rest of the way.
Want the full unfiltered discussion? Join me tomorrow for the daily AI news live stream where we dig into the stories and you can ask questions directly.

