Welcome to Today’s Update!

Hey everyone,

Today, I’m shaking things up a bit—because why not? Let’s dive in!

(I promise to fix the aspect ratio next time; my ultra-wide monitor is being a bit uncooperative.)

PS:
My talk from Stanford is now available on YouTube
. I had an early start at 2 AM, so only kind comments, please! You can expect early access to the course preview next week…

If you’re interested in discussing early experiments or sharing feedback on my projects, feel free to join my
new Substack group chat
.

Ben’s Bites is brought to you by
Shopify

Personal shopping capabilities are becoming available for all agents. Developers can access a single API at

shop.dev

for over a billion products featuring one-tap checkout. Shoppers may add this skill at

shop.md

, ensuring that any agent is aware of their size and order history while being able to check out directly in the chat—with no purchases without their consent.

Google has introduced its latest model, Gemini 4 Argon
, which is now accessible for everyone.
Dan has integrated it into his writing routine
, but I found it lacking—particularly in its design aspects. My tests revealed it to be slow and visually uninspiring, and
Theo deemed it unusable
.

I
tried it in Droid
to edit a short video: it excelled at analyzing reels and planning cuts but faltered in the UI, leading to strange
hallucinations in the final edit
.

— Keshav

A new version of Jev
is here, featuring multimodal inputs, extended context, and a new decision type: bounding boxes.

Claude Code desktop app
has received a design upgrade alongside numerous new features, such as multi-folder environments, child sessions, and side questions in the
/btw
tab. Additionally, they’ve developed their own version control system to replace Git.

Grok Bot 0.70.2
can now utilize Figma, Notion, and Canva as live cards in chat. Bots have improved functionality to search X without the need to connect an account, and they will confirm with you before executing any irreversible actions. Moreover, founders are now able to
submit their product as a connector
.

Are you closing your business? You might still have substantial value in your operational data—worth over $1M+. micro1 offers financial incentives to qualifying U.S. companies with 50+ employees for licensing records for AI training. If you refer a company, you could earn up to $50,000.
Find out if you qualify.
*

Share Ben’s Bites

[From Keshav]
I maintain two files that every agent can access:
chat-index.md
(a record of past chats, updated daily via automation) and
folder-map.md
(which indicates where each project is located). Instead of resuming a session, I typically
initiate a new one
by saying:

Read the chat where I was working on [project]. We’ll continue that today.

This method works with any agent capable of accessing your files.

Video mentions:

Alexandr Wang@alexandr_wang

you cannot out-shitpost me (with the help of muse wonder full)

5:18 PM · Oct 13, 2026 · 1.13M Views

397 Replies · 305 Reposts · 9.29K Likes

Sam Wilhoit@samwilhoit_

how I scaled my design workflow for a global billboard campaign that is live in NYC, Tokyo, London, and cities around the globe using coding agents in @FactoryAI 🧵🧵🧵
this thread is focused on using coding agents to reformat design assets for dozens of billboard panels that …

8:33 PM · Oct 14, 2026 · 7.62K Views

21 Replies · 10 Reposts · 92 Likes

greg@greg16676935420

@bertgodel Sorry but I’ve already given the Chinese all my personal information. Feels rude to switch now

5:08 PM · Oct 14, 2026 · 5.92K Views

7 Replies · 2 Reposts · 213 Likes

Ben Koska@BenKoska

while benchmarking our AutoResearch harness we noticed that there was a ~1.1pt jump in benchmark performance when the agent’s name is Ilya. it’s kind of funny to think about the set of circumstances that made that happen