AI with Kyle Daily Update 067

Today in AI: Claude Code app + Robin Williams Deepfakes

Author

Kyle Balmer
October 09, 2025

Catch Kyle’s latest video guide on Youtube.

Walking you through step by step:
How To Make $2,000 An Hour With AI Without Coding

The skinny on what's happening in AI - straight from the previous live session:

Highlights

📱 Claude Code Coming to Mobile: The End of Work-Life Balance?

Anthropic is preparing to launch Claude Code on mobile with GitHub integration, repository browsing, and the ability to run coding tasks directly from your phone. Very likely only available for Max subscribers ($200/month) when it rolls out - unconfirmed as of yet.

Kyle's take: This is exciting but terrifying. Last night I was vibe coding until 11pm and couldn't sleep because I was still running through problems. Now I won't even be able to step away from the computer - I'll be lying in bed thinking "actually, maybe I could fix this" and pulling up Claude Code.

When you can build anytime, anywhere suddenly any down time seems very wasteful. Gulp.

Also interesting to see the differing strategyt. While OpenAI spreads themselves thin with video, shopping, and agents, Anthropic is laser-focused on being the best at one thing: code. Smart strategy, but RIP my mental health.

🎭 Robin Williams' Daughter: "Stop Sending Me AI Videos of Dad"

Zelda Williams posted on Instagram begging people to stop sending her AI-generated videos of her late father after Sora 2's release triggered a flood of Robin Williams deepfakes. OpenAI has since restricted celebrity and cartoon character generation but not after upsetting instances like this.

Kyle's take: People likely thought they were being thoughtful sending her hundreds of AI videos of her dead father. Well meaning but…misplaced.

She called it "gross" and said AI is just "badly recycling the past... the human centipede of content."

This always happens - new models launch unrestricted for viral marketing (remember Studio Ghibli selfies?), then get locked down a week later. OpenAI very likely knew this would happen and probably had the restrictions ready to flip on but wanted their viral week first. Let’s be honest, even this controversy is more publicity for Sora.

Source: Variety

🛠️ Google Opal: The "Mini App Builder" That Could Kill Every Vibe Coding Tool

Google's expanding Opal to 15 more countries. Unlike OpenAI's agent builder, it starts with chat - you describe what you want in plain language, then it builds the workflow. They're calling it a "mini app builder" not a vibe coding tool.

Looks nice!

Kyle's take: This is what OpenAI should have done! Chat first! Then a drag and drop interface.

OpenAI’s agent builder falls between two stools - too complex for beginners (no chat interface, huh?), too simple for developers (drag and drop? Really?). Opal seems to gets it right: start with conversation, then build the framework.

What I’m interested in is why call it "mini apps"? This feels like they want to stay away from the “vibe coding” market - maybe for fear of it not being taken seriously? But this to me looks like a potentially free and very dangerous competitor to tools like Lovable and Bolt.

📊 Who's Burning the Most GPT Tokens? The Leaked List

At Dev Day OpenAI showed a "thank you" screen with developer names ranked by how many tokens they’ve used. Including those who have used over 1T tokens! Internet sleuths traced them back to companies: Duolingo, Salesforce, Shopify, Notion, HubSpot, Canva, and more.

Kyle's take: The chart is a little misleading because it’s ranked - suggesting that Duolingo use the most. Actually the names are just alphabetised by token usage tier (1 trillion, 100 billion, 10 billion) but people are misreading it as a ranked list.

ie. Isaac Anderson from Duolingo just has a surname starting with 'A'.

Still, fascinating to see who's burning through tokens. Anyone surprising in there?

Member Question: "Do you know the exact reason why Sora is not released in the UK?"

Kyle's response: It's the EU AI Act - strictest regulations globally, especially around deepfakes and using private information for digital avatars (face scanning, voice recording - exactly what Sora's cameo feature does). Post-Brexit, the UK doesn't really have its own AI rules yet, so we default to EU standards.

That's why we tend get AI releases at the same time as Europe. Sora 1 launched December 2024 in the US, we didn't get it until February - three month gap. Might be similar this time. But you can use a VPN to access the browser version at sora.chatgpt.com right now. More details on workarounds here in a guide I wrote: