Sunday Rundown #107: Agentic Coding & Tiny Shoe
Sunday Bonus #67: Explore what GPT-5 can do with my Custom GPT.
Happy Sunday, friends!
Welcome back to the weekly look at generative AI that covers the following:
Sunday Rundown (free): this week’s AI news + a fun AI fail.
Sunday Bonus (paid): an exclusive segment for my paid subscribers.
Read the Thursday post to get up to speed on the past month of AI news:
Let’s get to it.
🗞️ AI news
Here are this week’s AI developments.
👩💻 AI releases
New stuff you can try right now:
Anthropic news:
Claude finally got a memory feature that lets it reference past chats for relevant details.
Claude Sonnet 4 now has a 1M-token context, so it can handle entire codebases or hundreds of documents in one chat.
ElevenLabs launched Video-to-Music that can automatically generate a custom soundtrack based on the content of the video. (Try it here.)
Google news:
Gemini is also getting a memory to learn from your past conversations and a Temporary Chat option that doesn’t track your chat.
Gemma 3 270M is a fast, cheap, open model that can be fine-tuned for specific high-volume tasks and used on edge devices.
Flight Deals is an AI search tool that finds flights based on open-ended requests. Rolling out gradually in US, Canada, and India. (Try it here soon.)
Higgsfield rolled out a Draw‑to‑Video feature that lets you inpaint products, outfits, characters, props, and more directly into the scene using visual pointers.
Liquid AI open-sourced LFM2‑VL, compact, low-latency vision-language models for edge devices that can handle high-resolution images. (Try for free.)
Microsoft launched Copilot 3D, a free tool that turns 2D images into 3D models right in the browser without further prompting. (Try for free.)
Mistral AI introduced Mistral Medium 3.1, offering a performance boost, improved tone, and smarter web search capabilities. (Try in Le Chat.)
Skywork AI launched Matrix-Game 2.0, a real-time world model similar to Genie 3 that generates high-fidelity interactive videos from text prompts.
Vercel rebranded v0.dev as v0.app, turning it into an agentic AI builder that turns ideas into fully deployed apps with UI, backend, logic, etc.
Windsurf added Devin features that help explain code and handle bigger coding tasks agentically.
xAI temporarily made Grok 4 free worldwide to let anyone try its latest model without a paid plan.
🔬 AI research
Cool stuff you might get to try one day:
Pika Labs previewed an audio-driven model that animates realistic facial expressions from voice in near real-time.
Suno teased an upcoming Suno Studio platform with multi-track creation, MIDI export, and more. (Join the waitlist.)
🤦♂️ AI fail of the week
“Stupid levitating footwear!”
(ChatGPT’s one-panel meme about modern society…yeah, me neither.)
Send me your AI fail for a chance to be featured in an upcoming Sunday Rundown.
💰 Sunday Bonus #67: Test drive GPT-5’s capabilities with “GPT-5 Explorer”
The launch of GPT-5 wasn’t exactly a smooth ride.
But for all the disappointment and the well-deserved criticism of OpenAI’s messy release, GPT-5 is still the most capable LLM to date.
Yet it’s hard for the average person to figure out exactly what it can do for them.
That’s why I made GPT-5 Explorer: a Custom GPT that interviews you about your task and proposes several ways to use GPT-5’s capabilities to help you.
Here’s how it works:
You tell it what you’re trying to do right now.
It asks you a series of clarifying questions.
It suggests at least three varied, tailored ways it can help.
You pick one, and it executes it right inside the chat.
It can be a bit of a hit-or-miss, since ChatGPT occasionally suggests something beyond its capabilities, but it’s a quick, hands-on way to see how GPT-5 might be useful to you.
Try it here: