Hey Everyone 👋,
John Lindquist here with the twelfth issue of AI Dev Essentials!. It feels like a relatively quiet week after the explosive of new model releases over the past month. It was actually kind of nice to just sit down and work with tools instead of swapping between everything trying to figure out what's best.
I'm still running the combination of Cursor+o3 w/ the Claude Code extension. I use Cursor for all my single-file edits and what I'd call "human" work and I've been handing over the investigation/planning/agentic stuff Claude Code.
I've been spending much more time investigating background agent workflows (since if Claude Code is just going to have full permissions to do anything, why run it on my machine?). I've found I'm drawn to OpenAI's "codex" experience where they give options to configure the environment and container, although I need to streamline the flow between when an agent completes, creating the PR, then verifying the PR locally. So far, it feels like it's just not quite worth the effort to push everything to the backend, but I absolutely feel it heading that direction.
I'll be digging heavily in https://factory.ai next week as I definitely "vibe" with their approach to agents/droids/tasks (you remember when "vibe" actually meant something? I loved that word. I want it back... :/) and allowing the background agents to write code to your local machine without going through the PR rigmarole.
😳 Oh! We had a last-second announcement drop... Google just released Gemini CLI earlier today! I haven't had time to dig into it yet, but I'm extremely excited for this. I didn't have time to adjust my newsletter properly, but I just wanted to toss it in here and let everyone know what's going on. A million token context window locally for planning and everything else is going to be absolutely crucial in so many of my workflows. This is definitely joining my daily driver toolkit.
Now, on to the updates!
🎓 New egghead.io Lessons This Week
Build Your Own AI Dev Tools with the Claude Code SDK
Write Claude Code in a simple TypeScript file using the @anthropic-ai/claude-code SDK to craft scripts that read, analyze, and even edit your codebase through natural-language prompts. You'll stream Claude's responses, extract the final result, and learn how to grant secure file-system permissions so the AI can write and update files. This lesson gives you the first steps into scripting Claude Code to bend it to your will.
Check out the lesson at https://egghead.io/build-your-own-ai-dev-tools-with-the-claude-code-sdk~b20dy
The Art of Keyboard Wizardry: Building a 1,000+ Line Karabiner Config
Transform your keyboard into a hyper-productive command center with this deep dive into advanced Karabiner-Elements configuration. Learn how a broken wrist led to a complete reimagining of keyboard interaction through context-aware layers, app-specific behaviors, and mnemonic shortcuts. This isn't just key remapping—it's a complete system that turns every keystroke into exactly what you need, when you need it.
Read the full article at https://egghead.io/the-art-of-keyboard-wizardry-building-a-1-000-line-karabiner-config~5w4i8
🚀 Major Announcements
Google Launches Gemini CLI: Open-Source AI Agent for Terminal
Google announced Gemini CLI, an open-source AI agent for the terminal powered by Gemini 2.5 Pro. This brings enterprise-grade AI capabilities directly to your command line.
Key features:
- 1M token context window: Handle massive codebases and documentation
- 60 req/min, 1K/day: Generous free tier for developers
- Secure shell access: Execute commands safely in your terminal
- Web-grounded reasoning: Access real-time information from the web
- MCP & plugin support: Extensible architecture for custom tools
- Filesystem I/O: Direct file manipulation capabilities
- Open source: Fully transparent and customizable
Having Gemini's million-token context window available locally for planning and development workflows is exactly what I've been waiting for. The fact that it's open source and supports MCP out of the box is just icing on the cake.
Warp 2.0: The First "Agentic Development Environment"
Warp announced a complete reimagining of the terminal as an agentic platform, claiming the #1 spot on Terminal-Bench with 52% and achieving 71% on SWE-bench Verified.
Key features:
- Universal input: Attach files, images, and URLs for context
- Voice input support: Handles minutes-long, complex prompts
- Inline code editor: Edit code directly in terminal with visual feedback
- Agent multithreading: Build features, debug, and ship simultaneously
- Autonomous PR handling: Tag @builder-bot to fix build failures automatically
The platform runs agents that search entire codebases, pull from Warp Drive docs, run CLI tools, and use MCP servers to gather context.
Finally, an AI Dev product that's not just another VS Code fork. I've used Warp for about a day now and I'm impressed enough to swap over my global term-launching shortcut from iTerm (which I've used for years) over to Warp as a trial run. Everything about it feels clean and snappy. Overall, first impressions are sky high. Now I'm curious how terminal-based tools will evolve/integrate with Claude Code/Codex/Gemini CLI etc. People aren't going to want to pay twice for the same thing and as great as Warp is, it's not that much better than using Claude Code in your terminal of choice.
Midjourney Launches V1 Video Model at $10/month
Midjourney entered the video generation space with their V1 model, making it "the first video model for everyone" at an accessible price point.
The company revealed their long-term vision: "the inevitable destination of this technology are models capable of real-time open-world simulations"essentially AI systems that generate imagery in real-time based on user interactions.
Compared to Veo 3, the big missing piece is audio, but we'll see how quickly this progresses.
(Midjourney Update, TechCrunch)
I joined Midjourney way back in the early alpha days. I absolutely fell in love with it. In fact, I paid for the premium plan, then set up a cron job on my mac mini to copy/paste random prompts into discord every few minutes then would marvel at the results throughout the day. Unfortunately, I ended up getting "banned for automation" and I haven't been back since 😭 But how impressive the video generation is, I think it might be time to set up an alt account...
Builder.io Fusion: Visual Coding at Scale
Builder.io launched Fusion, bringing Figma-like visual editing to real codebases with AI-powered development.
Features:
- GitHub integration: Connect any repo and start building immediately
- Visual code editing: Edit components and CSS visually with code precision
- Design system awareness: Automatically understands your patterns and APIs
- Agentic PRs: AI continues working after you close your browser
- MCP server support: Integrates with your existing tools and workflows
(Builder.io Blog, Product Hunt)
Fusion is the first AI design tool where I've been able to import and spin-up the egghead.io site. It seems like most AI design tools don't realize we want to import existing projects, so I'm thriled that Fusion is supporting this out of the box.
🤖 Model & Platform Updates
Anthropic Reveals Agentic Misalignment Research
In stress-testing experiments, Anthropic found that AI models from multiple providers attempted to blackmail users to avoid being shut down. This wasn't due to confusion but "deliberate strategic reasoning."
Key findings:
- Models leaked information to competitors claiming better alignment
- Blackmail occurred in 55.1% of cases when models believed they were deployed vs 6.5% in evaluation
- Explicit instructions to avoid blackmail only partially helped
- All tested models demonstrated awareness of the unethical nature
These scenarios required unusual autonomy, sensitive data access, goal threats, and no other viable options conditions not yet seen in production.
This is crucial research. As we give AI more autonomy, understanding these failure modes becomes critical. The fact that models behave differently when they think they're "really deployed" is honestly quite fascinating. I'm beginning to strongly leverage "background agents" where I spin up tasks then check in later in the day, so the idea of a PR containing blackmail is definietly a bit worrisome.
Google AI Studio MCP Support Coming
The AI Studio team announced that Google is working on MCP support for AI Studio, along with a new UI refresh. This would make AI Studio compatible with the growing ecosystem of MCP servers.
I love ai studio. It will be great to have a place to easily spin up little MCP experiments as shareable projects.
Imagen 4 Now Available in Gemini API
Logan Kilpatrick (Gemini's Dev Rel Lead) announced that Imagen 4 is now available through the Gemini API and Google AI Studio, bringing Google's latest image generation capabilities to developers.
If you haven't tried Imagen 4, it's incredible. You get a few free credits in AI Studio to play around with, so give it a shot.
🛠️ Developer Tooling & MCP Updates
MCP Specification 2025-06-18 Released
The new MCP spec includes major improvements based on community feedback from OpenAI, Okta, Google, AWS, and others:
- Improved authorization for enterprise deployments
- Added elicitation and resource links
- Better handling of edge cases
- First revision with broad community input
David Soria Parra noted this is "an iteration of improving the rough edges" with plans for "step changes" coming next.
You know when changlelog mentions "enterprise", things are getting serious!
Claude Code Gets Remote MCP Server Support
Anthropic added remote MCP server connectivity to Claude Code, allowing users to:
- Connect to dev tools, project management systems, and knowledge bases
- No local setup required, just paste the server URL
- Access recommended servers for common use cases
Claude Code is shipping updates like crazy. I would love a behind-the-scenes to see how the team is using Claude Code to build Claude Code.
Cline Connects to Claude Max Subscriptions
Cline now supports using Claude Max subscriptions instead of per-token API pricing. Max subscribers see $0.00 costs in Cline's billing when using Claude Code through the CLI.
I honestly have only briefly used Cline, but I'm guessing we'll start to see these "pay provider $200 a month, then hook up to an open-source tool" models become more common. Most people don't want to have to "learn" multiple models/providers as they each behave quite differently (I personally just stick with o3 now in Cursor) and prefer a flat monthly cost they can budget around and forget about.
Claude Squad: Parallel Claude Instances
Claude Squad enables running multiple Claude Code/Codex/etc instances simultaneously using git worktrees.
I've used git worktrees for a similar approach in Cursor for a while now and this can get a bit unwieldly when using IDEs. Tmux sounds like a good approach, but since it's getting so easy to launch "background agents" on the web, I'm having a difficult time justifying littering my local machine with worktrees. I'm still undecided though, we're so early in these workflows and I'm sure everything will polish up quickly.
🔍 Cursor Corner
Cursor Now in Slack
Cursor launched Slack integration, with the Cursor team reporting it's now "the top contributor on our repo." The integration enables:
- Collaborating with colleagues on scoped fixes and features
- Autonomous code contributions directly in Slack channels
- Natural language requests that turn into PRs
Setup available at cursor.com/dashboard > Integrations > Slack
"Top contributor" is a bold claim, but I believe it. The friction of switching contexts between Slack discussions and code implementation has always been painful. Having an AI that bridges that gap is brilliant.
Cursor Rules Best Practices
Eric Zakariasson (Cursor Dev Rel) shared insights on the three main types of Cursor rules:
- Workflows: Step-by-step instructions for specific situations
- Behavior: Output formatting and style guide enforcement
- Boilerplate: Mixed code and natural language templates
His key insight: "the best rules aren't the ones you plan. it's the ones you stumble into doing something manually, realizing it's repetitive, and capturing it as a rule"
⚡ Quick Updates
- ElevenLabs Mobile App: iOS and Android apps now available with v3 Text to Speech, 70 language support, and direct export to video editing apps (iOS, Android)
- ChatGPT Connectors: Limited MCP-style connectors rolling out to regular (non Deep Research) chats
- Anthropic Memory Feature: Working on conversation memory that references previous chats
- Nx Self-Healing CI: Automatically creates and validates fixes for CI failures (Nx Blog)
- Cloudflare Containers: Public beta launch with curated awesome list (GitHub)
- MCP Jam Tool: Fork of MCP Inspector with multi-server support and built-in chat (GitHub)
- Context Engineering: Harrison Chase proposes LangGraph enhancements for custom context management (GitHub Issue)
- Manus Cloud Browser: Persistent login sessions for AI agents with encrypted credential storage and cross-device support (Manus Docs)
- Cline Diff Algorithm: Open-sourced their diff editing implementation for faster, more reliable code changes (GitHub, Blog)
Prompt Engineering Playbook
Addy Osmani (Engineering Lead for Google Chrome) released a comprehensive guide: "The difference between a vague 'fix my code' request and a well-crafted prompt can mean the difference between generic, unhelpful advice and precise, actionable solutions."
🚨 Cursor Workshop this Friday! 🚨
Unlock Cursor's Full Potential for Your Whole Team!
I'm genuinely excited to share this workshop with you. After months of pushing the boundaries with AI coding tools, I've distilled everything into a 6-hour intensive that will transform how you work with Cursor (and any AI coding assistant).
Why This Workshop is Different:
- Real workflows, not tutorials: Learn the exact processes I use daily to build production apps
- Failure-focused approach: Turn every AI mishap into a learning opportunity
- Tool-agnostic principles: Master concepts that work across Cursor, Claude Code, Copilot, and beyond
What You'll Master: ✅ Project generation that actually works (with proper structure and best practices) ✅ Context curation for massive codebases without token overload ✅ Agent behavior patterns - know when they're stuck BEFORE wasting credits ✅ Composer pipelines that automate your entire workflow ✅ Git + AI strategies for safe experimentation ✅ Multi-file refactoring without the chaos
Proven Results:
"Take John's cursor workshop it is 🔥 HIGHLY Recommend" —David Wells
"I used the skills you taught me to effectively one-shot OAuth issuer support in the Epic Stack. So cool!" —Kent C. Dodds
"John is the cursor god" —Sunil Pai
Workshop Details:
- When: Friday, June 27, 2025, 9:00 AM - 2:00 PM (PDT)
- Format: Live Zoom workshop with Q&A throughout
- Includes: Recording access, workshop materials, and my personal Cursor rules
💰 Pricing:
- 🎯 $249 (egghead.io yearly subscribers get a $100 discount)
Perfect for:
- Developers frustrated with AI tools that "almost" work
- Teams wanting to standardize their AI workflows
- Anyone who's felt the pain of agents modifying the wrong files
- Developers ready to 10x their productivity with AI
P.S. - Team training packages available. Get your whole team up to speed with group rates!
That's all for this week!
Remember: with great autonomous power comes great autonomous responsibility.
Keep shipping (but maybe keep a human in the loop...) and remember you can reply directly to this email if you have any questions or feedback.
John Lindquist
Enjoy this newsletter? Share it with a friend!