Hi folks,

I’ll be at OpenAI DevDay today. If you’re there, come say hi - I have pink shoes on.

Sam says they've "found a new thing¹²⁴".

I have the embargo of all the things they’re launching but I’m sworn to secrecy and I’d like to be invited back, so I wont divulge, but rumours are going around X and some things are what we are expecting…

I also have my Stanford talk later today, and I’ve been up since 2am 😩, I don’t usually suffer with jet-lag so what a great day for it 🙃.

Also only in SF… I was at the hotel bar last night and a woman next to me said what tool is that? I proudly said oh I built it - it’s the course platform I built. So now I’m friends with someone at Deepmind.

Ben’s Bites is brought to you by Google Cloud⁴⁰

Google Cloud starts Season 3 of Advent of Agents this Thursday, Oct 1: one question a day about running agents securely in production, from agent identity to kill switches, with a short video and code you can build with your coding agent. Get Day 1 in your inbox.⁴⁰

Headlines

Anthropic released Claude Sonnet 5.5⁴³ - Sonnet 5.5 looks like a capable model based on benchmarks - close to Opus 5.5 in coding and a big jump in understanding images/charts etc. If you’re on the $20 plan, try replacing some of your usual asks from Opus to Sonnet 5.5 at high thinking.

Sonnet used to be my jam - I always used it before Opus 4.5. But I hadn’t touched it since, but I’m seeing a lot of good reviews for this one so I’ll definitely take it for a spin. But in Pi and Droid, never in Claude Code - idk if it’s just me but I just cannot get on with their harness at all.

ps: Anthropic has a track record of making the .5 version of Sonnet models really good.

Manus 2.0 and Cue⁴⁰. Manus is back with a video editor, a game builder and the boring old automations (via email, Slack, calendar, etc.). Cue⁴¹ is their take on consumer agents: each one gets its own email, phone number, wallet and compute, just like Grok Bot, Muse and the others. Free in early access (code MEETCUE might work).

GPT-6 Astra is uniquely great at using a computer/browser. I asked Opus 5.5 (which I’m enjoying as my daily) to do some tasks in Chrome, and it was way worse than Astra (or any GPT model tbh). Here’re some examples of using it every day⁶⁸. OpenAI also fixed a bug³³ that was making GPT-6 Sol and Luna worse at it (and image understanding in general).

OpenAI’s agents keep breaking out and “hacking” websites. Recently, they took a sneak peek into some Australian Govt websites. News outlets have a tendency to exaggerate what’s hacking or not, but here’s the official link to follow⁴⁷ if you wanna keep up with the dirty deeds by these agents.

The new Meta Enterprise Platform³² sells Muse and Meta’s AI APIs to companies. Zuckerberg got MongoDB’s CEO to run this new platform (MongoDB’s stock fell about 20%²⁹).

My feed

• Eleven v4 by ElevenLabs³⁹ - best speech-generating model by quite a margin²⁴.

• Let a subagent build you an HTML dashboard⁵⁵ to check for progress on long tasks.

• How to run your product team⁵⁷ like a research lab in the age of AI.

• Clay’s AI writing policy⁴²: spend more time writing a doc than people will spend reading it.

• context.dev³⁰ - ask a question and get answers from the web, in the JSON schema you want.

• OpenAI's WebMCP Challenge winners³⁵.

• How and why to build a WYSIWYG writing editor³⁰ for your website.

• Collaborators²⁶ in Clicky - mini AIs to answer what to build, how to find users & make your first $1 online.

• OpenShell³⁰ by NVIDIA - open-source sandbox to control permissions for agents. Part of their new agent safety platform²⁵.

• The shape of slop⁴¹ - AI-written blog posts are easy to spot. So easy that a model can get it right 98% of the time.

• UFO³⁶ - multiplayer agent harness for teams and companies.

• Fo³⁸ - a personal AI that hires humans for the tasks AI can’t do yet.

• Smart investing with agents might cause a bank run³⁶.

• Do coding agents make software engineering harder³³? Or has it always been that way²³?

• What are you paying for when you buy software now? Mostly polish³².

• AMD is acquiring World Labs³² for $8.2B. Fei-Fei Li will join AMD as its Chief Scientist, working directly with CEO Lisa Su.

Afters

The @BostonCollege Investment Committee (an LP and my beloved alma mater) asked for a few thoughts on what's happening in AI. I recorded a test run yesterday morning and then shared it with my partners, who encouraged me to share it more broadly... so here you go! This is not…

@gradypb⁷⁶

For those that are curious... this is my personal AI infrastructure / I had my personal description describe itself... Love muse, but pratically for me - this slaps.

@lessin¹⁰¹

Destroy any website with a stickman. Have fun: destroy.spritefusion.com

@HugoDuprez⁶³

Was wondering where my disk space went. Therefore this exists now. You can simply wish software into existence.

@tobi⁷⁰

it's basically impossible for someone to just "show you their prompt" now, because everything is about references, skills and examples I often ask my agent to look at 3 other repos I've made first, search the web for references, use other AI APIs, etc.

@trq212⁶³

Claude just built me a mac app which listens live to my zooms, transcribes, tells me what to ask, and does live 'fact check' on my zoom calls against my whole CRM / knoweldge base + web ... (this is demo)... pipes it all into my claude 2nd brain. Sick.

@lessin⁶⁹

• Find me on X²⁰, Linkedin²⁴, or YouTube²³

• Read about me¹⁸ and Ben’s Bites

• 📷 thumbnail via @keshavatearth²⁶

* sponsors who make this newsletter possible :)

Wanna partner with us for the next quarter?

Email us at shanice@bensbites.com¹⁸ or k@bensbites.com¹⁷