AI Dev Essentials #24: Gemini in Chrome, Notion Agents, and OpenAI's ICPC Win

Hey Everyone 👋,

John Lindquist here with the 24th issue of AI Dev Essentials!

This week has been jam packed with devtool news. Gemini slid right into Chrome, Notion went all in on AI, OpenAI shipped a fresh new coding model, and Meta bets on glasses with wristbands. Everything is speeding up without any hint of slowing down, and the best part is that these upgrades are getting real love and proving that AI is becoming more essential to our daily tools. I spent a huge chunk of the week building out my Claude Code Essentials course (it will essentially be the prereq for the Claude Code Power User workshop).

Grab a spot for the Oct 3rd workshop here: https://egghead.io/workshop/claude-code

🎓 New egghead.io Lessons This Week

I published my Claude Code Essentials course this morning. It kicks off with nine lessons packed with the early tips and concepts that knock down that first bit of friction when you pick up a fresh tool. I am already queuing up more lessons, so expect this course to keep growing. Even if you have lived in Claude Code for a while I tucked in little gems in every lesson to keep things interesting. These first nine lessons focus on developer ergonomics and the next set will move deeper into habits, best practices, and sharpening the skills that make you a confident Claude Code developer. Putting this together has been a blast and I am grateful I finally carved out the time to share what I have been learning with you all.

🐬 Dive in to the "Claude Code Essentials" course here:

Claude Code Essentials on egghead.io

🚀 Major Announcements

Google Chrome Receives Major AI Upgrade

Google announced significant AI enhancements to Chrome on September 18, 2025, integrating Gemini AI directly into the browser for Mac and Windows users in the U.S.

Key features:

Gemini in Chrome: Native AI integration for intelligent browsing assistance
AI Mode in Address Bar: Search with context-aware AI for complex queries
Cross-tab Functionality: AI can work across multiple tabs simultaneously
Agentic Browsing: Autonomous task handling for appointments and shopping (coming soon)
Enhanced Google Apps Integration: Deeper connection with Calendar, YouTube, and Maps

I have been on the beta for a while and I absolutely love hitting Control G to pull up Gemini for little tasks. It still feels like a version one because it is missing the keyboard shortcuts and the ways to work on selected text or answer a thread, but it is still an awesome version one. I cannot imagine the scale of shipping AI to everyone using Chrome. Those numbers are unimaginable to me.

(Google Chrome Blog, TechCrunch)

Notion 3.0 Launches with AI Agents

Notion officially launched version 3.0 on September 18, 2025, introducing autonomous AI Agents capable of performing up to 20 minutes of independent work across hundreds of pages.

Agent capabilities:

Personal Agents: Handle complex projects like building launch plans autonomously
Custom Agents: Run autonomously on schedules (coming soon)
Tool Integration: Works with Slack, Google Drive, and GitHub
Latest AI Models: Includes Claude Sonnet 4 and GPT-5
Revenue Milestone: Notion crossed $500 million in annual revenue

I'm not a Notion user but this felt like pretty big news. It's fascinating to see the idea of an agent show up across so many products. I don't think many people have thought about what the user interfaces for agents will look like in a couple years or what the knowledge bases will become. I'm in the mindset that the agent should be the first layer and then it reaches into the application to take you where you need to go. Google's hardware approach with Gemini as a single button on the phone that reacts to the screen feels like the winning formula. I'm curious what AI and agents mean for the consolidation of user interfaces and which products will struggle once agents understand context and sweep away the complexity.

(Notion Official Blog, CNBC)

OpenAI Achieves Perfect Score at ICPC World Finals

OpenAI's general-purpose reasoning model achieved a historic perfect 12/12 score at the 2025 International Collegiate Programming Contest World Finals, outperforming all 139 human teams.

Performance highlights:

ICPC World Finals: First AI to achieve perfect score, would have placed 1st
GPT-5 System: Solved first 11 problems with experimental model for final problem
GPT-5-Codex Release: Optimized for agentic software engineering
7+ Hour Autonomy: Can work independently on complex tasks
Multi-platform: Available in CLI, IDE extensions, web, mobile, and GitHub

I was lucky enough to join the beta for the Codex model and it has handled some pretty complex tasks I tossed at it. I feel much more comfortable handing it multi step features or fixes that take time to think through, stepping away from the computer, and coming back to real progress. Plenty of folks are already using the Codex CLI as their main development tool. It will be interesting to see how Anthropic responds because I keep hearing whispers about new models dropping next week. The competition in the coding space is a huge win for all of us because it pushes every big provider to ship better models more often and at better prices.

(OpenAI Codex Announcement, The Decoder)

Meta Unveils Ray-Ban Display Glasses with Neural Interface

Meta announced the Ray-Ban Display glasses with Meta Neural Band at Connect 2025 on September 17, 2025, priced at $799 and available September 30.

Revolutionary features:

Full-Color Display: Private viewing of messages and navigation
Meta Neural Band: EMG wristband translating finger movements to commands
18-Hour Battery: Long-lasting wristband with 4 years of research backing
200,000 Participants: Extensive testing program for accuracy
September 30 Launch: Available for $799

The EMG wristband is the sleeper feature here. Being able to control interfaces through subtle muscle movements opens up entirely new interaction paradigms. The quality and reliability will be crucial because if Meta can nail the input accuracy this could genuinely change how we interact with computers. My big hesitation is wearing glasses when everyone knows the cameras are always watching. I don't know if we need a full generation to get over that idea. I'm starting to think I should accept it and maybe even embrace it for myself because the upside looks so useful.

(Meta Official Announcement, TechCrunch)

🛠️ Developer Tooling Updates

Vercel Agent Enters Public Beta with AI Code Reviews

Vercel launched Agent in public beta for Pro and Enterprise customers in September 2025, offering AI-powered code reviews that analyze correctness, security, and performance.

Agent features:

Codebase-Aware Reviews: Examines beyond just the diff
Sandbox Validation: Patches tested in Vercel Sandboxes before suggestions
Framework Support: Next.js, React, Nuxt, and Svelte
Multi-language: TypeScript, Python, and Go support
Usage-Based Pricing: $100 Vercel Agent credit included

I'm always interested in agents that are tightly integrated with the tooling and stack they live in. A Vercel agent tuned for the Vercel platform and the Next.js framework should guide you from a smelly PR straight to the right tool, fire up whatever analysis is needed, and knock out the fix for you. That kind of deep integration could really stand out if you are already in the Vercel ecosystem compared to the generic code review tools floating around.

(Vercel Community Announcement, Vercel Changelog)

GitHub Launches MCP Registry for AI Tool Discovery

GitHub released the MCP Registry on September 16, 2025, as a centralized hub for discovering Model Context Protocol servers with one-click installation.

Registry benefits:

One-Click Installation: Direct VS Code integration
Community Metrics: Servers sorted by GitHub stars and activity
Launch Partners: Figma, Postman, HashiCorp, and Dynatrace
Cross-platform: Works with Copilot, Claude, and MCP-compatible tools
Open Contribution: Community-driven server submissions

Discovering MCPs has been a challenge and I agree this registry helps, but I'm extremely hesitant to install MCPs even when friends recommend them because there are still big security issues and the context cost can hit your tooling. So while I think the registry is a great idea I'm still going to do my own homework before installing anything that looks interesting. My concern is that people will install MCPs willy nilly and run into the traps that pop up when you do not fully understand what an MCP is doing to your project.

(GitHub Blog, GitHub MCP Registry)

Claude Developer Platform Unveils Tool Helpers for SDKs

Anthropic launched tool helpers in beta for Python and TypeScript SDKs through the Claude Code SDK, simplifying tool creation and execution in Claude applications.

SDK improvements:

Automatic Input Validation: Type-safe parameter checking
Tool Runner: Automated tool handling in conversations
Package Availability: claude-code-sdk on PyPI, @anthropic-ai/claude-code on NPM
Simplified Creation: Reduced boilerplate for tool definitions
Cross-language Parity: Consistent Python and TypeScript experience

My Power User workshop essentially forced me to build out a custom SDK to make advanced agents easier to build (which I have no desire to maintain) so I'm thrilled whenever the official Claude Code SDK ships a new update. Consistent tool calling is a huge piece of making AIs more productive and useful when you build out automations and brings me one step closer to dropping my own SDK 😇.

(Claude Code SDK Documentation, Claude API Release Notes)

Cursor 1.6 Ships Custom Commands and MCP Resources

Cursor released version 1.6 with custom slash commands stored in .cursor/commands/[command].md files, MCP Resources support, and 26 stability patches.

Version 1.6 features:

Custom Commands: Reusable prompts in .cursor/commands/ directory
MCP Resources: Share contextual data like files and database schemas
/summarize Command: On-demand context summarization
Variable Interpolation: Enhanced MCP configuration options
26 Stability Patches: Significant reliability improvements

I love custom commands. I'm a script heavy automation person and these commands are an amazing way to pair an AI with CLI tooling and wrap it behind a simple English word or two. Commands started in Claude Code and I love seeing them show up in other IDEs because I think more of us will start using them.

(Cursor 1.6 Changelog, Cursor MCP Documentation)

🤖 AI Ecosystem Updates

Anthropic Publishes Detailed Postmortem on Recent Claude Issues

Anthropic released a comprehensive technical report detailing three overlapping infrastructure bugs that caused intermittent degradation of Claude's response quality between August and early September 2024.

Key findings:

Three Infrastructure Bugs: Context window routing errors, token generation corruption, compiler/top-k bugs
Peak Impact: 16% of Sonnet 4 traffic affected on August 31
Complete Resolution: All bugs fixed with preventive measures implemented
Transparent Communication: Full technical details shared publicly
Enhanced Monitoring: Improved systems to catch similar issues faster

This level of transparency from Anthropic is refreshing. I can't imagine trying to hunt down non-deterministic, overlapping bugs spread across different infrastructures. Hopefully this puts an end to all the speculation and drama around recent quality issues.

(Anthropic Engineering Blog, Hacker News Discussion)

Microsoft Announces $30 Billion UK AI Investment

Microsoft CEO Satya Nadella announced a $30 billion investment in UK AI infrastructure from 2025-2028, representing Microsoft's largest financial commitment ever in the UK.

Investment details:

Record Commitment: $15.5B capital expansion, $15.1B operations
UK Supercomputer: Over 23,000 NVIDIA GPUs in new facility
Stargate UK: Partnership with OpenAI and NVIDIA for sovereign AI compute
Timeline: 2025-2028 investment period
Job Creation: Thousands of high-skilled AI positions

Big numbers keep getting bigger. Would sure be nice if a couple of those GPUs and 0.1 of the Bs were accidentally sent to my address...

(Microsoft Official Blog, OpenAI Stargate UK)

Groq Raises $750M at $6.9B Valuation for AI Chip Development

AI chip startup Groq secured $750 million in funding at a $6.9 billion post-money valuation in September 2024, more than doubling its previous $2.8 billion valuation.

Funding highlights:

Lead Investor: Disruptive, with BlackRock, Neuberger Berman, Deutsche Telekom
Doubled Valuation: From $2.8B (August) to $6.9B (September)
LPU Performance: 241 tokens/sec on Llama-2 70B, 640 tokens/sec on Mixtral 8x7B
Enterprise Focus: Targeting large-scale deployment needs
Manufacturing Scale: Ramping production to meet demand

My experience with Groq has been pretty limited to experiments. Usually running into call limits and prohibitive pricing. But the speed! It's so hard to go back to other models after you see the instant responses from Groq. I'm cheering for them and glad to see they have more money in the bank.

(Bloomberg, TechCrunch)