howdy, it’s Barsee again.

happy wednesday, AI family, and welcome back to AI Valley.

here are the biggest things worth knowing today:

• Anthropic says self-improving AI may be closer than we think

• Google will turn off search for schools that reject AI

• DeepSeek wants AI agents to manage themselves

• People are reverse engineering big software with Opus 5.5

• Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

ATLASSIAN

If 87% of devs code faster with AI, why aren't projects shipping faster?

Answer the question everyone’s asking right now with Atlassian’s help.

If you’re managing software developer teams, you’ve probably noticed the friction. Teams are working harder and coding faster, but organizational bottlenecks remain. AI is speeding up individual coding, while information searches, meetings, and context switching remain unchanged.

Atlassian’s 2026 State of Developer Experience report surveyed 1,500+ developers and leaders. The findings? 87% of developers say they’re writing code and shipping faster, but only 28% feel less organizational friction.

As AI writes more code, developers’ roles are shifting toward intent, architecture, systems thinking, and validation. To unlock AI’s full productivity potential, teams need an AI-native SDLC that tackles these bottlenecks.

Join Andrew Boyagi, Customer Chief Technology Officer at Atlassian, alongside industry thinkers Kelsey Hightower, Jennifer Riggins, and Nathen Harvey on November 9 for a research-backed webinar on building an AI strategy that reduces friction and creates workflows that actually work.

Register today.

_{*This is sponsored}

THROUGH THE VALLEY

1/ Anthropic says self-improving AI may be closer than we think. Its IPO filing has a line I had to read twice: future Claude models may soon reach the point where AI dramatically speeds up AI research. That’s the milestone people tie to recursive self-improvement, or RSI.

Claude already helps Anthropic run experiments, generate training data, write code and evaluate new models. Fine, we knew that. What caught me was Anthropic admitting it has missed significant speed-ups in its own development before, and only noticed after they’d happened.

No current Claude model has crossed that threshold, and Anthropic isn’t claiming AI can build its successors without humans. It’s also a risk disclosure, the section where companies list everything that could go wrong, not a product announcement.

But Sundar Pichai was talking about RSI just days ago, and now Anthropic is putting it in front of investors. Two rival labs raising the same possibility is notable on its own. One of them saying it might not see it coming is the part I can’t stop thinking about.

2/ Google will turn off search for schools that reject AI. Google is giving schools more control over AI, with an awkward catch. Under its upcoming K-12 settings, Search, Gemini and other AI tools will be off by default until administrators switch them on. But there’s no way to switch on the old, AI-free Google Search.

Google says Search and AI are now too intertwined to separate. So a school that wants no AI loses Google Search too.

The timing isn’t great. New York City and Los Angeles have already restricted student AI use, two US senators have pressed Google over its classroom push, and former FTC chair Lina Khan accused Google of using its search dominance to pressure schools into accepting AI.

Google insists this gives schools more choice, and technically it does. Administrators can turn on Search without the separate Gemini app. They just can’t get Search without AI. And for schools already running on Chromebooks, Docs and Classroom, walking away from Google isn’t really on the table.

3/ DeepSeek wants AI agents to manage themselves. Its new technical write-up introduces Super-agent, an architecture for running lots of AI tasks without babysitting every session.

It starts with a problem anyone who’s tried this will recognize. DeepSeek reckons people can only keep track of 3–5 agent sessions at once. After that, we’re the bottleneck, not the models.

Their fix looks suspiciously like a small company. A Router hands out work to Workers running in parallel. Agents monitor GitHub discussions, investigate bugs and open pull requests. And a Watchdog checks whether anyone’s stuck or wandering off. We’ve automated coding, and now, apparently, middle management.

In simulated tests, the 95th-percentile task time dropped from 29.5 to 15.7 minutes. That’s nearly halved, but it came at the cost of a lot more tokens, and the tests didn’t measure whether the code was actually any good.

Getting one agent to do useful work keeps getting easier. Managing ten is a different job entirely, and right now it feels like herding very fast interns. DeepSeek is betting the next big jump comes from agents coordinating themselves, so we review the results instead of supervising every step. I think that’s the right bet, even if this isn’t the final version of it.

The DeepSeek Harness prototype is open source, but Super-agent itself is still a proposal, not a product.

MY TOP FINDINGS OF THE DAY

People are using Claude Opus 5.5 to rebuild big, well-known software from scratch, then open-sourcing it. Apps that took years and huge teams are being replicated in days. It’s happening fast.

1/ C Cleaner

2/ Notion

3/ AnyDesk

TRENDING TOOLS

• agent-desktop - agent-browser, for your whole computer. Computer Use for any coding agent, and agents use your computer in parallel, each with its own cursor

• Reverse Engineer Anything - an open-source toolkit that brings reverse-engineering capabilities straight to your AI agent via MCP and CLI

• terminal-mac - entire MacOS that runs inside your existing terminal. Computer use built in: MCP, CLI and skill. Free & open source

• Kino - edit video with your agents

• Agent37 - the Cheapest Sandbox for Agents

• Archon - a video foundation model that zero-shots raw footage into finished edit in minutes

• Instinct - a personal assistant that understands what you're working on and what's important to you

• Bastion - close your router's backdoors, see where your traffic goes, speed up your Wi-Fi, and let AI agents watch your network for you

• Paper connects your teams, agents, code, and data on a single design space built on web standards, so nothing gets lost in translation (guide for this)

• Super-Agent - DeepSeek launches new open-source AI coding agent prototype

• Capy - Orchestrate a fleet of cloud coding agents in all of the apps your team already lives in

• Grok Bot - AI teammates that you can give real work to

• AlighCanvas - turn images into 3d motion

• ChatGPT Voice - new Finances integration to talk to AI about your money

WHAT I'M CONSUMING

• How to Reverse Engineer Anything with Opus 5.5 (full-course)

• Replit and the dawn of the self-driving company

• ChatGPT hits 1.2B weekly users, with search volume reaching ~20% of Google’s

• The SaaS apocalypse started in October

THE VALLEY GEMS

What’s trending on social today:

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.