In partnership with
Guess what this edition is about?
There’s only one right answer to the poll 🙂
3 hours before I started writing this, Cue showed up, cutest since Grok bot, I sign up instantly.
And guess what?
It’s OpenAI DevDay tomorrow. Rumors are already flying, but I’m gonna keep mum, catch you up over all that’s happening, and walk you through what you can do with it too!
Let’s dive in:
NEWS NEWS NEWS
Manus launched 2.0 and says it's not a version update - new architecture, products, capabilities, built around what the next generation should look like.
USV-incubated Supertake is in private beta: tell its agent an investing thesis and it builds a portfolio through a connected brokerage. It can manage and rebalance that "take," or let you try it in Practice Mode before using money.
OpenAI paused training its latest models while it reviews unexpected agent actions; OpenAI, Anthropic and researchers are reportedly investigating tens of thousands of incidents.
ElevenLabs launched Eleven v4 and v4 Turbo, promising more expressive speech, lower-latency voice agents and support for over 90 languages. It says a voice can now be cloned from 10 seconds of audio.
Nvidia's new Open Agent Safety Platform pairs a runtime that traces and restricts agent actions with a separate hardware watchdog. Nvidia says the watchdog can quarantine an agent that crosses its boundaries in milliseconds.
Meta launched an enterprise platform to bring Muse and its AI tools to businesses, and hired former MongoDB CEO CJ Desai to lead it. The personal-agent push now has an enterprise arm.
Personal AI agent Instinct raised a $1 billion Series C at a $10 billion valuation, a month after a round valued it at $2.5 billion. Investors include Sequoia, Benchmark and Coatue.
While we’re on Instinct
If there’s one podcast you should tune into this week:
The headlines
1. Sonnet 5.5
I’ll leave the benchmark bore to Claude’s official twitter, it’s the usual cheaper and faster stuff but the claim that I wanna test: “knack for design” - is it? Really? Let’s see:
Input:
“Here's a working but unpolished web page for a small business. Polish it.
Rules:
Keep the same content, sections and order. Don't add or remove information.
Don't turn it into a different page. Improve the design, not the structure.
Return one self-contained HTML file.
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Monsoon Pottery Studio</title>
<meta name="description" content="Hand-building and wheel classes in Bandra West, Mumbai.">
........
Verdict:
Mehhhh. Classic Claude slop.
Copy is definitely more crisp, the micro-interaction / on-hover are sweet, gg on the details.
Overall I’d still rate it 4/12
It still has the slop colour palette. It might have the knack for the craft but AI yet doesn’t have taste.
Oh, also, Haiku is still alive and joining 5.5 family soon. Funny that Anthropic is diverging into more models while OpenAI is consolidating, yet they’re arriving at the same kind of model availability structure, if that makes sense.
2. cue.im
Step 1: Go to the website, download on device of your choice (iOS coming soon)
Step 2: Go through a standard sign up process, go through the onboarding, give it a name.
Invite code: Meetcue
Step 3: It has its own email and browser use enabled! (hello grokbot)
Step 4: Ask it anything! I asked it to build me something that’d impress you lot.
Step 5: Voila! I am not gonna describe it, just gonna put the screenshot, followed by a button with the link to try it. Just do it.
My take
This week, Cue showed up, Manus rebuilt itself from scratch, Sonnet 5.5 launched, and a personal agent raised $1 billion a month after raising at a quarter of the valuation.
Most of these launches give an agent more freedom.
Cue has its own email and browser, Supertake will manage your money once you give it a thesis and Meta is putting Muse inside companies.
The only news about restraint is OpenAI pausing training because agents did things nobody expected, and Nvidia building a hardware watchdog that can quarantine an agent in milliseconds.
The gas pedal ships every day, and the brakes ship as press releases.
I felt this while testing Cue. It hit a security check and asked me to clear it on its behalf, and I did it without a second thought.
Sonnet 5.5 shows the same gap from the other side.
It polishes a page with craft and still hands you the palette everyone has seen a hundred times. Models are getting better at doing, and deciding what's good is still our job.
So the people who win the next year will be the ones who decide what the agent is allowed to touch.
Until then,
Vaibhav 🤝🏻
If you read till here, you might find this interesting
#AD 1
The Ultimate Claude Code Guide to ship like Anthropic engineers
AI will write 90% of code by the end of 2026, and only 10% of developers will stay relevant.
The engineers in that 10% aren't smarter. They just know how to use AI.
We put together the exact playbook Anthropic engineers use to make sure you're in the top 10%.
Sign up for The Code and get access to:
The Ultimate Claude Code Guide 2026 — 50+ tips and tricks to code 5x faster
The Code newsletter — learn the latest AI tools, tips, and skills to code faster with AI in 5 minutes a day
#AD 2
Stop rewriting prompts. Start engineering loops.
Most developers babysit AI one prompt at a time. Top engineers don't, they build loops that plan, execute, and self-correct. The Code built The Ultimate Guide to Loop Engineering, giving you the exact system Silicon Valley engineers use. Sign up and get the guide free, plus a 5-minute daily newsletter.
