AI Dev Essentials - Issue #22
Hey Everyone 👋,
John Lindquist here with the 22nd issue of AI Dev Essentials!
This past week I actually spent some time stargazing in Bryce Canyon with my beautiful wife. It honestly felt amazing to step away from the noise of the world and appreciate the beauty of the Universe. When I got back home, I sat down to focus on planning out a Claude Code intro course which I'll be releasing on egghead.io for free. This will essentially serve as a prerequisite course for people who are planning to attend my Claude Code Power User Workshop. I'm starting recording today and hopefully you'll have something to share next week.
Sign up for the next workshop here: https://egghead.io/workshop/claude-code
🚀 Major Announcements
OpenAI Expands ChatGPT Projects to Free Users
OpenAI has made ChatGPT Projects available to all free users globally, following their pattern of gradually expanding paid features to broader audiences.
Key features for free users:
- Project organization: Create unlimited projects for organizing conversations by topic
- Enhanced file uploads: Up to 5 files per project (vs. 25 for Plus, 40 for Pro)
- Custom project icons: Choose colors and icons for better organization
- Project-only memory: Separate memory contexts for each project
- Cross-platform access: Available on web and Android app, with iOS rollout coming soon
This represents a major democratization of ChatGPT's advanced features, previously exclusive to paid tiers.
(OpenAI Help Center, Engadget)
I love the fact that the memory will be separate from other projects. One of the biggest pain points of using ChatGPT is memory leaking from one thing to another. In one chat I might be talking about an Electron app, and in another I might be talking about a Tauri app and the suggested solutions would obviously be for the wrong technology. Really happy to see these siloed separations taking place.
Google Intensifies AI Chip Competition Through Strategic TPU Expansion
Google has escalated its competition with NVIDIA by approaching smaller cloud providers to host TPUs in their data centers, striking deals with providers like London-based Fluidstack.
Strategic developments:
- TPU v5p Performance: 2.8x faster at training large language models compared to TPU v4
- Market Impact: Google's TPU sales estimated at $6-9 billion in 2024, potentially reducing NVIDIA's market share for the first time
- Next-Gen Trillium: Sixth-generation TPUs promise 4x boost in training performance and 3x increase in inference throughput over TPU v5e
- Distribution Strategy: Partnering with smaller cloud providers to expand TPU availability beyond Google Cloud
This move represents Google's most aggressive challenge to NVIDIA's AI chip dominance to date.
NVIDIA needs the competition since they're absolutely dominating the AI chip market. The competition is good for the industry and for us as the end users. Hopefully this simply results in better products and better pricing for everyone.
🛠️ Developer Tooling Updates
Zed Editor Integrates Claude Code via Agent Client Protocol
Zed launched Claude Code integration in public beta through their Agent Client Protocol (ACP), bringing AI-powered coding directly into the high-performance editor.
Integration features:
- Native Claude Code support: Full integration with existing Claude Code subscriptions
- Agent Client Protocol: Open standard specification for connecting any AI agent with any compatible editor
- Performance optimization: Maintains Zed's speed while adding AI capabilities
- Multiple agent support: Framework supports various AI agents simultaneously
- Public beta: Currently in beta testing phase for user feedback
The integration leverages Zed's performance advantages while providing access to Claude's coding capabilities.
(Zed's Official Blog, Agent Client Protocol GitHub)
I always keep a close eye on Zed. And I know the performance versus convenience conversations are always difficult ones to have. I wouldn't call myself a Cursor fanboy by any means, but anytime I try another product, I just keep on coming back to it. It always feels like it's just a little something here or there that's missing that's already a major part of my workflow. But I'm cheering Zed on, and I hope that one day it's strong enough to pull me away. Claude Code integration is definitely a massive plus.
into.md Tool Simplifies Web Content Conversion
The into.md tool converts any web page into clean, LLM-friendly markdown by simply prepending "into.md/" to any URL, created by Timo Lins.
Tool capabilities:
- Simple URL prefix: Add "into.md/" before any URL to convert content
- Clean markdown output: Removes navigation, ads, and irrelevant content
- LLM optimization: Formatted specifically for AI processing
- No signup required: Works immediately without registration
- Open tool: Available for public use
This addresses the common need to feed web content to AI models in a clean, processable format.
I love this new into.md tool because copying the content of a web page into an AI prompt is turning into a common everyday task that I pull off multiple times. I use a Chrome extension called Mark Snip for this most often. But I've run a few experiments with using into.md with curl or Node's fetch (make sure you respect redirects if you do that). And it's helped me brainstorm up some really interesting scenarios of websites as prompts that I'm looking forward to play more with.
Nano Banana Prompt Templates: Mastering Gemini 2.5 Flash Image
A comprehensive guide to using Nano Banana (Gemini 2.5 Flash Image) has emerged from the community, providing six essential prompt templates for maximizing image generation quality.
Template categories:
- Photorealistic Scenes: Detailed photographer-like descriptions with camera angles, lighting, and mood specifications
- Stylized Illustrations: Creating specific design styles like kawaii icons and custom artwork
- Accurate Text Rendering: Generating logos and designs with precise text integration
- Product Mockups: Professional commercial photography setups
- Minimalist Design: Negative space compositions for websites and presentations
- Sequential Art: Comic panel development and narrative visual scenes
The core principle emphasized throughout: "Describe the scene, don't just list keywords." Each template provides structured approaches with Python code examples for Gemini API integration.
I shamefully ripped this post off of x.com and saved it as a gist for much easier copy/paste into my prompts 🫣. Here's the original post: https://x.com/GoogleAIStudio/status/1962957615262224511
(Nano Banana Prompt Templates Gist)
Jules Agent Introduces Sample Prompts for New Users
Google's Jules coding agent added sample prompts to the homepage on September 2, 2025, to help new users get started with AI-assisted development.
New user experience improvements:
- One-click examples: Pre-built prompts demonstrate common use cases
- Homepage integration: Sample prompts displayed on Jules home page for all users
- Single-click workflow: Add prompts to input box with one click
- Getting started guide: Reduces barrier to entry for new users
- Best practices: Shows optimal ways to interact with the agent
This addresses one of the main barriers to AI agent adoption: not knowing how to get started effectively.
(Jules Official Changelog, Jules Documentation)
Jules is one of those tools where I feel like most people are aware of its existence. But all the various experiments and tools in the Google ecosystem seem so separate that it's hard to remember what does what. I use Google's AI Studio extensively and I use Gemini extensively. But then opening up another tab to use Jules or Stitch or Flow or many of the other AI experiments, which are all amazing, just seems like a weird context switching that I hope they integrate into either Gemini or AI Studio. Again, this isn't a knock on their team at all, it's just the story of switching or trying to follow all of Google's various projects is turning into a task in and of itself.
💼 AI Ecosystem Updates
Anthropic Addresses Claude Opus Performance Issues
Anthropic experienced significant Claude Opus performance degradation from August 25-28, 2025, affecting Claude Opus 4.1 and 4.0 models for approximately 56.5 hours.
Technical details:
- Inference stack rollout: Problematic infrastructure update caused performance issues
- Performance degradation: Users reported "lower intelligence, malformed responses or issues with tool calling in Claude Code"
- Root cause identified: Infrastructure optimization attempts backfired
- Resolution implemented: Rollback to stable inference architecture
- Duration: 56.5-hour incident affecting multiple model versions
The transparency around infrastructure issues demonstrates Anthropic's commitment to user experience.
(Anthropic Status, Simon Willison's analysis)
I've seen many people commenting on issues with Claude Code over the past week. Whereas I was actually out of town for most of it, so I guess my timing was perfect. But it's always good to see providers own up to their mistakes and be more transparent with why people might be having bad experiences. I know how frustrating it can be to be working at a large company and trying to tell your manager that you're not making progress because your tools aren't working as expected.
⚡ Quick Updates
Claude Artifacts Enable Interactive App Development
- Interactive artifacts: Build and share functional web applications directly in Claude
- Cost-effective sharing: Users pay for their own API usage, not creators
- Real-time iteration: Modify and improve apps through conversation with dedicated artifacts panel
- Shareable links: One-click sharing capability for published artifacts
- General availability: Available to all Claude.ai users across Free, Pro, and Team plans
(Anthropic Official Announcement, Anthropic Help Center)
Backlog.md: Open-Source Task Management for Developers
- Markdown-native: Task manager and Kanban visualizer for Git repositories
- Cross-platform: Available on macOS, Linux, Windows with MIT licensing
- AI-friendly: Designed for collaboration between humans and AI agents
- Developer workflow: CLI and web interface with drag-and-drop functionality
(GitHub Repository, Hacker News Discussion)
✨ Workshop Spotlight
Claude Code Power User Workshop
Ready to master advanced Claude Code workflows? Join the next power user workshop to learn professional automation patterns and advanced techniques.
Almost sold out! Grab your ticket before they're gone.
Next session: Friday, September 12, 2025 Time: 5:00 AM - 10:00 AM (PDT) Platform: Zoom 🇪🇸 Europe-Friendly Time! That's 1:00 PM in London, 2:00 PM in Paris & Berlin
Register: https://egghead.io/workshop/claude-code
Read this far? Share "AI Dev Essentials" with a friend! - https://egghead.io/newsletters/ai-dev-essentials
- John Lindquist