- AI with Kyle
- Posts
- AI with Kyle Daily Update 044
AI with Kyle Daily Update 044
Today in AI: 30% of code is AI + Dia goes to die
The skinny on what's happening in AI - straight from the previous live session:
Highlights
🆓 ChatGPT Free Users Get Projects, All Users Get Conversation Branching
OpenAI rolled out Projects to free users (organise chats with shared knowledge bases and custom instructions) and conversation branching to all users (explore different directions without losing your original thread).
By popular request: you can now branch conversations in ChatGPT, letting you more easily explore different directions without losing your original thread.
Available now to logged-in users on web.
— OpenAI (@OpenAI)
8:13 PM • Sep 4, 2025
Kyle's take: This is brilliant news for anyone who's been on the fence about paying for ChatGPT. Projects are super useful - I use them constantly in both ChatGPT and Claude to keep different work streams separate and stop the AI pulling in random context from other conversations.
Branching is even cleverer - you can go "buck wild" exploring crazy ideas without polluting your main conversation thread. It's like having parallel universe chats and then can choose which one to continue with.
Source: OpenAI feature announcement
🤖 Sam Altman Claims Twitter Is Now Mostly AI Bots
OpenAI's CEO posted on Twitter saying he "never took the dead internet theory that seriously, but it seems like there are a lot of LLM run Twitter accounts now." This is ironic since many of those bots are likely powered by ChatGPT… The dead internet theory suggests the web is increasingly just bots talking to bots rather than humans.
i never took the dead internet theory that seriously but it seems like there are really a lot of LLM-run twitter accounts now
— Sam Altman (@sama)
10:21 PM • Sep 3, 2025
Kyle's take: Weird flex from Sam Altman - going onto Elon's platform to basically say "your site is rubbish because it's full of bots that my company probably created." It's like an arsonist complaining about all the fires around town.
The dead internet thing isn't just conspiracy theory nonsense though - there genuinely are loads more bots now. I showed Deddit on stream:
It’s literally Reddit but entirely run by AIs having conversations between themselves. The dead internet theory itself posits that there’s some planning to the dead internet - governments and companies up to no good trying to control us. That’s likely not the case!
Whether this is some grand manipulation or just the natural result of making AI accessible doesn't really matter - the effect is the same: bots talking to bots.
Source: Sam Altman Twitter / Deaddit
🎨 Google's Nano Banana Can Read Text in Images and Use It as Prompts
Google's new image model Nano Banana can read text within images and incorporate it into generation prompts. A user showed it a 1920s criminal mugshot with details about his crimes, prompted "make him do the crime he was arrested for," and Nano Banana correctly generated a period-accurate warehouse break-in scene without being told what the crime was!
Ok, this is crazy 🍌
Google’s new image model, Nano Banana, can actually read the text inside an image and use it as part of the prompt.
For example, I found this image of a London criminal. Without me adding the crime to the prompt, the model read the text in the image,
— Martin LeBlanc (@martinleblanc)
8:28 PM • Sep 1, 2025
Kyle's take: Really cool stuff. The AI read the mugshot text, figured out the bloke was a "shop and warehouse breaker," and generated the exact crime scene in 1920s photographic style.
Increasingly I’ve been hearing people use the phrase "Nano Banana moment" like the original ChatGPT breakthrough. I don't usually cover image models since I focus on business applications, but this text-reading capability could be brilliant for product photography and brand work. Being able to extract and build on written information in images opens up some really clever automation possibilities.
Source: Social media demonstrations
📊 30% of Code Now Written by AI (Not the Predicted 90%, But Still Massive)
A year ago, Anthropic's CEO predicted 90% of code would be AI-written by now. That didn't happen. And people laughed at him.
BUT 30% of code was AI-generated by December 2024 - likely higher now. While critics mock the failed prediction, the actual adoption rate represents massive change from, well, ~0% just five years ago.

Kyle's take: Missing the target doesn't mean the direction is wrong. Shoot for the stars, aim for the moon and all that.
Going from basically zero to 30% in a few years is absolutely staggering progress. That’s being ignored.
This is the pattern with AI - boosters get the direction right but timelines wrong. Making bets against AI because of overhyped deadlines is still problematic when the underlying tech keeps delivering real value. I use AI for coding daily and so do most developers I know. We're clearly heading toward much higher percentages, just not as fast as the hype merchants claim.
Source: Ethan Mollick analysis
🪦 Atlassian Buys Arc Browser for £610M (RIP Innovation)
Atlassian acquired The Browser Company, makers of Arc browser and the new AI-focused Dia browser, for £610 million. This comes a day after I finally got access to Dia, which uses AI to browse and search the web!

Boo!
Kyle's take: Atlassian is where startups go to die - they've ruined Trello, made Loom worse, and turned everything they touch into clunky enterprise software that people hate but have to use (like Jira).
Arc was brilliant, and Dia looked promising, but now they'll probably become slow, expensive, and designed for corporate purchasing departments rather than actual users. I'll be sticking with Perplexity's Comet browser for AI browsing. It's like watching someone feed a promising racehorse into a bureaucratic glue machine… gosh, a horrid simile, sorry!
Member Question from Ozzie: "Opinions on building an AI audit without delivery, but build referral system for that?"
Kyle's response: Absolutely could work, but you need to partner with implementers upfront. The big risk is your audit flagging different things than what the implementation team actually focuses on - that creates a mismatch that kills trust.
Most engineers are terrible at marketing and finding customers (sorry guys!), so they'd love someone doing the front-end work of securing clients.
Remember always that the real AI opportunities are usually boring backend stuff - reducing support tickets, moving data around - not obvious things like "add a chatbot." You need to dig into their actual workflows to find the valuable automation opportunities.
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.
Streaming on YouTube (with screen share) and TikTok (follow and turn on notifications for Live Notification).
AI with Kyle - Tools, Resources and Products: Daily Show | Build an AI Business Challenge | AI Playbooks | 10 Week AI Summer Camp | BATON AI Business Series | AI App Building | AI Trainer Business Planner | What AI Business Should You Start? | AI Authority Accelerator | AI Automation Accelerator | 🏁 AI Workshop Kit