- AI with Kyle
- Posts
- AI with Kyle Daily Update 145
AI with Kyle Daily Update 145
Today in AI: Gemini 3.1 Pro's really good, but pretty useless...
What’s happening in the world of AI:
Highlights
🧠 Gemini 3.1 Pro: Smart Brain, No Hands
Google dropped Gemini 3.1 Pro and it's a great model. Scores 77.1% on ARC-AGI-2, blowing past Opus 4.6's 68.8%. But after a morning of testing, the verdict? A bit meh. Not because the model is bad, but because it's still just a chatbot. You type, it replies. That's it. That’s no longer enough.
Kyle's take: I gave it my standard test: help me plan a content workflow based on my actual content. It didn't bother looking at my website, my socials, anything. Just spat out a generic waterfall strategy.
The model isn't the problem. The harness is. Claude has Claude Code and Cowork. OpenAI has Codex. These don't just talk about work, they do the work.
When I ask Codex or Claude Code to help me with a content strategy they’ll go off and research who I am, what I publish, how it all fits together. In fact when I tested with the Codex app recently it ran a full scrape on my past newsletters. It got to work.
In 2026, the model layer has converged. GPT 5.3, Opus 4.6, Gemini 3.1. All brilliant. The differentiator is what wraps around the brain. Until Google sort that they’ll be behind Claude and ChatGPT. Chatting to AI is very 2025! How fast we move on!
That said, don't bet against Google. They build their own TPUs, run their own cloud, own Chrome and Android, Gemini's going inside Siri, and they pulled in $402 billion revenue last year. They are doing a’ight. Once they crack the harness layer, watch out.
Source: Google Gemini announcement
😬 Anthropic Keep Burning Goodwill
Anthropic have now also sent legal notices to Open Code, explicitly banning external tools from using or orchestrating Claude Code.
This follows the whole OpenClaw saga where Peter Steinberger's fastest-ever-growing open source project got hit with a cease and desist, he panic-renamed it at 5am, someone grabbed the domain for a crypto scam, and he eventually got hired by OpenAI. Anthropic came out looking like the corporate bully.
Kyle's take: Legally, a hundred percent within their rights. You have to defend your trademarks or they get diluted. And third-party tools have been against ToS since the beginning. But the optics are terrible.
And OpenAI are capitalising on this. They announced they are allowing subscription use in external tools. Developers paying $200 a month are the exact people being alienated here. Anthropic were the goodies. The safety-focused, ethical alternative. Now they're stomping on indie devs and the community is pissed.
This comes at the same time as there now being a credible Claude Code competitor. Codex 5.3 is good enough that switching is a credible threat. It’s been a rough 2026 for Anthropic so far.
📸 Google's Free Product Shots vs a Thousand Startups
Google released "Photo Shoot" inside Pomelli (Google Labs). Upload one product image, get high-quality customised product shots for e-commerce. Free. Dozens of startups built their entire business around doing exactly this, mostly wrapping Imagen with system instructions and charging a subscription. Google just made their product free overnight.
Kyle's take: This is life in 2026. Wake up, pray a big lab doesn't one-shot your company with a single launch. I’m seeing a lot of “your startup is dead “ type narratives around this.
But most people have never heard of Pomelli. Your company is not just the product. It's the distribution, the brand, the existing customers.
Remember that Google release these things to flesh out their ecosystem, not to build niche businesses. They aren’t bothered about competing with startups in the millions of dollars range. It’s just too small a target - chump change for Google, a rounding. This is why they won’t actual market this. They’ll announce this (like all their AI products) and go onto the next thing.
If you've built a following and your app is easier to use, you've got breathing room. This applies not just to product-shot apps but ALL businesses you build using AI.
Source: Google Labs
🔧 Karpathy's Vision: The App Store Is Dead
Andrej Karpathy vibe-coded a custom cardio experiment tracker in about an hour. Had Claude reverse-engineer his treadmill's API, pull data, build a dashboard. Bugs? Sure. But 300 lines of code, perfectly tailored to his needs. His conclusion: the concept of an app store full of discrete apps feels increasingly outdated when an LLM can build one on the spot. Recommend reading the whole essay:
Kyle's take: Same experience here. Needed a remote control to paste text into my TV. Every app was buggy, ad-filled, or trying to nick money. Had an LLM build one instead. Done.
Karpathy's bigger point is that every device should expose an API for AI agents, not a human-readable front end.
Increasingly we’ll be building apps for AIs. NOT for humans. That’s the shift here.
Karpathy has (correctly!) spoken about these broad developments in software before. If you haven't read his "Software 2.0" essay, it's on the AI Canon at aiwithkyle.com. Still one of the most important things written about what is going on nowaday.
Source: Andrej Karpathy on X / Software 2.0
Member Question from Jacob: "Which AI is best for creating courses?"
Kyle's response: Claude for the content. Its writing is a cut above and it pushes back on your ideas instead of telling you everything's brilliant (looking at you, ChatGPT!). Nano Banana Pro for images. For video, record it yourself! People want access to you, not an AI avatar.
Bigger picture: information is free now. I give away all my courses at aiwithkyle.com. The value is in guiding people through the information and making sure they act on it. Get Claude to interview you about your subject, draft material you then edit, layer in interactive AI tools, then wrap it in a cohort for accountability. That's the future of education.
Streaming on YouTube (with full 4k screen share) and TikTok (follow and turn on notifications for Live Notification).
