Hey Everyone 👋,
John Lindquist here with the 27th issue of AI Dev Essentials!
OpenAI's Dev Day introduced new developer tools and APIs and Google launched updates across the Gemini CLI and the rest of their AI offerings. Never a dull week in AI-land!
I was able to attend OpenAI's Dev Day in San Francisco on Monday. It was great to finally meet people from both the Cursor team and the OpenAI Codex team in person. It's great to be able to sit down with people you spend so much time with online.
With all of the recent updates with Cursor over the past few months and with many requests from the community, I'm relaunching my Cursor workshops! To kick things off, I published the first eight lessons of a Advanced Cursor Hooks Course. They're available for free on egghead.io here: https://egghead.io/courses/advanced-cursor-hooks~swp89
If you want to level up your Cursor skills with shared rules, agent workflows, and production-ready patterns, check out the details in the Workshop Spotlight section below.
Register: https://egghead.io/workshop/cursor
🎓 New egghead.io Lessons This Week
I published the first eight lessons of the free "Advanced Cursor Hooks" course:
- Capture Agent Context with Your First Cursor Hook
- Debugging Cursor Hooks with Output Panels
- Simplify Cursor Hooks Configuration with JSON Schema
- Level Up Cursor Hooks with TypeScript and Bun
- Type-Safe Cursor Hooks with the cursor-hooks Package
- Enforce Cursor Tooling Standards with beforeShellExecution
- Automate Cursor Code Quality with afterFileEdit
- Logging and Debugging Cursor Hooks
👀 Watch the lessons here: https://egghead.io/courses/advanced-cursor-hooks~swp89
🚀 Major Announcements
OpenAI Dev Day 2025 Expands Developer Platform
OpenAI hosted Dev Day 2025 on October 6, 2025, announcing new APIs, models, and developer tools that expand the platform's capabilities for building AI applications.
Key announcements:
- Apps in ChatGPT with Apps SDK: Built on Model Context Protocol (MCP), enables developers to build and test apps in ChatGPT with app submissions opening later in 2025
- GPT-5 Pro in API: Extended reasoning model now available for developers' hardest tasks
- Sora 2 and Sora 2 Pro in API: Video generation models in preview with limited access
- AgentKit Platform: Complete toolkit including ChatKit (embeddable chat UI), Agent Builder (WYSIWYG workflow creator), Connector Registry, and Enhanced Evaluations
- Codex General Availability: Cloud coding agent with Slack integration, SDK, and admin tools (cloud tasks billable starting Oct 20)
- New Mini Models: gpt-image-1-mini (80% cheaper image generation), gpt-realtime-mini (70% cheaper voice model)
- Infrastructure Updates: Service health dashboard for API monitoring, GPT-5 40% faster on priority processing tier
([OpenAI Dev Day](https://openai.com/Dev Day/), OpenAI Apps SDK, OpenAI AgentKit, TechCrunch)
Honestly, no huge surprises here. Most everything had already been released or announced, so it was lot of recaps. The ChatKits annoucement was interesting and I imagine we'll all have to take it fairly seriously considering their reach of hundreds of millions of users.
Codex stole the show as it was mentioned everywhere. The highlight for me personally was hearing Jony Ive talk about his work on the future devices for OpenAI. Definitely check out the recording on YouTube to here his thoughts on the future of devices. I loved that he touched on that not everything has to be so serious and focused on productivity. Life should be fun! I hope whatever they come up with helps humanity reconnect with each other rather than being just another "tool" to do more work.
Google Launches Extensions Ecosystem for Gemini CLI
Google announced the Gemini CLI Extensions system on October 8, 2025, introducing a one-command installation framework that packages MCP servers, context files, and custom commands for AI-assisted development.
Platform features:
- One-Command Installation: Install extensions from GitHub repositories with
gemini extensions install <URL> - 22 Launch Partners: Including Atlassian, Canva, Dynatrace, Elastic, Figma, GitLab, Harness, Postman, Shopify, Snyk, Stripe, and others
- MCP Integration: Extensions can package and connect to Model Context Protocol servers
- Open Ecosystem: No Google approval required for publishing extensions, hosted in public GitHub repositories
- Million+ Developers: Over 1 million developers already using Gemini CLI as of announcement
(Google Blog, TechCrunch, SD Times)
I've been wanting plugin ecosystems for these tools for a long time. Everything I've built with custom agents and bundling tools always feels like I'm fighting against the tools. I'm so happy Gemini is taking the initiative here and I hope the other tools follow suit.
Google DeepMind Unveils Gemini 2.5 Computer Use Model
Google DeepMind released Gemini 2.5 Computer Use on October 7, 2025, a specialized browser control model developed in partnership with Browserbase that can navigate web interfaces like a human.
Model capabilities:
- State-of-the-Art Browser Control: Leads on Online-Mind2Web benchmark according to Browserbase evaluation
- 13 UI Actions: Supports clicking, typing, scrolling, and other predefined interface interactions
- Complex Navigation: Can operate behind login screens and interact with complex UI elements
- Public Preview: Available through Gemini API on Google AI Studio and Vertex AI
- Browserbase Partnership: Helped train, evaluate, and deploy the model; provides demo environment and testing infrastructure
(Google DeepMind Blog, Browserbase Blog, VentureBeat)
You don't realize how insanely useful a computer use model is until you find the right scenario. I recently had ChatGPT do a deep research on myself, then copy/pasted it into a beta for Claude's Chrome extension. The extension filled out my entire profile for me, adding all the details, education, and work experience while dismissing all of the annoying popups and dialogs. It was awesome. Such a tedious task and I just got to sit back and watch it happen.
Anthropic Names Rahul Patil as New CTO
Anthropic announced on October 7, 2025, that Rahul Patil, former CTO of Stripe, has joined as Chief Technology Officer, with co-founder Sam McCandlish transitioning to Chief Architect.
Leadership changes:
- Rahul Patil as CTO: Will oversee engineering organization across product, compute, infrastructure, inference, data science, and security
- Sam McCandlish as Chief Architect: Focuses on pre-training and large-scale model training
- Stripe Experience: Patil guided Stripe's technical organization supporting trillions of dollars in annual processing
- 20+ Years Experience: Previous roles at AWS, Microsoft, and Oracle Cloud Infrastructure
(Anthropic Official, TechCrunch)
Sounds like a good hire. I can't think of many companies that require as much performance and reliability as Stripe. Hopefully the skills transfer and we'll see Anthropic take the next step.
🛠️ Developer Tooling Updates
Genkit Extension for Gemini CLI Enhances AI Development
Google released the Genkit Extension for Gemini CLI on October 8, 2025, providing context-aware assistance for building AI applications with the Genkit framework.
Extension features:
- Context-Aware Assistance: Intelligent code generation specific to Genkit development patterns
- Debugging Support: Built-in debugging guidance for common Genkit issues
- Best Practices: Framework-specific recommendations and patterns
- MCP Integration: Bundles Genkit's MCP server with specialized context files
- One-Command Install:
gemini extensions install https://github.com/gemini-cli-extensions/genkit
(Google Developers Blog, TechCrunch)
Jules Tools and Jules API Bring Agent to Terminal and CI/CD
Google introduced Jules Tools and Jules API on October 2, 2025, extending the Jules coding agent beyond the web interface into developer workflows.
New capabilities:
- Jules Tools CLI: Command-line interface bringing Jules directly into the terminal
- Jules API: Public API for integrating Jules into CI/CD pipelines, Slack, Jira, and other tools
- Developer-First Approach: Director of Product Kathy Korevec noted "developers live in the terminal"
- Workflow Integration: Enables Jules to participate in existing development workflows
(Google Labs Blog, Google Developers Blog, VentureBeat)
Vercel Academy Launches AI SDK Course
Vercel Academy released "Builders Guide to the AI SDK," a free comprehensive course created by Joel Hooks and Nico Albanese covering AI fundamentals and practical implementation patterns.
Course structure:
- Three Main Sections: Foundations, Invisible AI, Conversational AI
- Topics Covered: AI Elements, tool calling, data extraction, streaming responses
- Tech Stack: Built with TypeScript, React, and Next.js
- Free and Open Source: Available at vercel.com/academy/ai-sdk
Joel Hooks always delivers quality materials (and he's also a wonderful person). Definitely check this out if you're looking to get started with the AI SDK.
🤖 AI Ecosystem Updates
Ling-1T: Trillion-Parameter Efficient Reasoning Model
Ant Group's inclusionAI released Ling-1T on October 8, 2025, the first flagship non-thinking model in the Ling 2.0 series featuring 1 trillion total parameters with approximately 50 billion active per token.
Model highlights:
- Scale and Efficiency: 1T total parameters, ~50B active per token, trained on 20T+ reasoning-dense tokens
- Evo-CoT Training: Evolutionary Chain-of-Thought curriculum with Linguistics-Unit RL
- FP8 Precision: Largest FP8 mixed-precision trained foundation model to date
- Benchmark Performance: #1 among open models on ArtifactsBench, ~70% accuracy on BFCL V3 tool-use benchmarks
- Availability: Released on HuggingFace (inclusionAI/Ling-1T) and ZenMux
(HuggingFace, FP8 Training Blog)
Google DeepMind's CodeMender Automates Security Patching
Google DeepMind introduced CodeMender on October 6, 2025, an AI agent powered by Gemini Deep Think that autonomously identifies and patches software vulnerabilities at scale.
System capabilities:
- Dual Approach: Combines reactive patching and proactive vulnerability elimination
- Advanced Analysis: Uses static analysis, dynamic analysis, fuzzing, and SMT solvers
- Scale: Handles codebases up to 4.5 million lines, upstreamed 72 security fixes to open source
- Validation: Ensures patches fix root causes, maintain functionality, and follow style guidelines
- Production Ready: Six months of real-world deployment and testing
(Google DeepMind, SiliconANGLE)
Self-healding code is one of the next big unlocks with AI. It's strange to think of how hackers battled each other in the past and now we'll have malicious AIs battling these defensive AIs to keep software secure. It's some real sci-fi movie stuff, but without all the music and drama.
Claude Memory Cookbook Provides Implementation Patterns
Anthropic released a comprehensive memory implementation cookbook on October 7, 2025, demonstrating how to use the new memory tool that stores information outside the context window.
Memory system features:
- File-Based Persistence: Memory stored in dedicated directory that persists across conversations
- CRUD Operations: Create, read, update, and delete memory files
- Client-Side Control: Developers manage storage backend and retention policies
- Security Considerations: Cookbook covers memory validation and sanitization
- Multiple Patterns: Examples cover basic memory, compactify memory, file-based memory
Memory is tough. We all know it's a critical feature, but what should the tools remember and when should they recall them are tricky problems. Glad to see Anthropic taking the time to write this cookbook.
⚡ Quick Updates
ElevenLabs Releases Open Source UI Components
- 22 Audio Components: MIT-licensed React components for chat, transcription, waveforms, music
- Built on shadcn/ui: Fully customizable component library
- Ready to Ship: Drop-in components for voice agents and audio applications
- Available Now: ui.elevenlabs.io and GitHub
Vercel AI Elements Now Open Source
- Announced August 6, 2025: Component library for AI SDK applications
- Built on shadcn/ui: Prebuilt, customizable React components
- AI-Native Design: Components specifically designed for AI interactions
- Free Access: Available via GitHub
Claude Code Mobile in Development
- Mobile Preparation: Industry reports indicate Anthropic preparing Claude Code for mobile
- GitHub Integration: Connect to repositories from mobile devices
- On-the-Go Coding: Run coding prompts from phones
- No Official Date: Release timeline not yet announced
GPT-5 Pro Expected for Perplexity MAX
- API Access Announced: GPT-5 Pro available via OpenAI API after Dev Day
- Perplexity Integration: Expected to add support for MAX subscribers
- Following Pattern: Perplexity typically integrates latest OpenAI models quickly
- Highest Capability: GPT-5 Pro is OpenAI's most advanced API model
([OpenAI Dev Day](https://openai.com/Dev Day/), [TechCrunch](https://techcrunch.com/2025/10/06/what-to-expect-at-openais-Dev Day-2025-and-how-to-watch-it/))
Building Claude Code with AI
- Pragmatic Engineer Feature: Gergely Orosz interviewed Claude Code team
- 90% AI-Written: Claude Code's codebase is 90% written by AI
- High Velocity: Team ships ~5 PRs per engineer daily
- AI-First Workflow: Extensive use of AI agents throughout development
👆 Cursor Corner
Eric Zakariasson on Working with Coding Agents
Cursor team member Eric Zakariasson published "Working with coding agents" on October 2, 2025, a comprehensive guide covering the fundamentals of effective AI-assisted development.
Key topics covered:
- Prompting: Communicating intent and mental models to agents
- Context: Providing proper background and constraints
- Self Verification: Enabling agents to validate their own work
- Model Selection: Choosing between speed and intelligence
- Background Agents: Running long tasks autonomously
- Parallel Agents: His recommendation: 1-3 agents for optimal throughput while maintaining oversight
Eric's recommendation of 1-3 parallel agents aligns with my experience. Beyond three agents, context switching overhead outweighs productivity gains. The emphasis on self-verification through tests and linting provides necessary feedback loops for improving agent outputs. His distinction between faster models for maintaining flow versus slower models for background work addresses practical tradeoffs in model selection.
✨ Workshop Spotlight
Cursor Workshop - October 17th, 2025
Early Bird Pricing: $225 (Save 25% from $300) - Ends Friday!
Date: October 17, 2025 Time: 9:00 AM - 2:00 PM (PDT) Platform: Zoom
In just a single day you'll set up shared rules, optimized workflows, and agent mastery to help you ship AI-generated code with confidence.
Testimonials:
- Kent C. Dodds: "One-shot OAuth in minutes."
- David Wells: "🔥 Highly recommend."
- Sunil Pai: "John is the Cursor god."
What You'll Learn:
The AI development landscape shifts daily with new models, tools, and best practices. This workshop teaches fundamental principles and repeatable workflows that work regardless of which tool you use—Cursor, WindSurf, Kiro, Cline, or the next hot thing.
Essential Skills:
- Project Generation Mastery: Accelerate from zero to structured codebase with best practices
- Planning & Cursor Rules: Refine workflows with custom rules and step-by-step planning
- Bulk Code Analysis & Bugfinding: Leverage specialized debugging tools to isolate errors fast
- Logging & Feedback Loops: Pipe error logs and test outputs directly into Chat or Composer
- Refactoring & UI Checks: Use AI to systematically review code quality and front-end design
- Agent Toolchain Pipelines: Create specialized workflows that automate terminal tasks
Register: https://egghead.io/workshop/cursor
Read this far? Share "AI Dev Essentials" with a friend! - https://egghead.io/newsletters/ai-dev-essentials
- John Lindquist