article

AI Dev Essentials #28: Claude Haiku 4.5 & OpenAI's Big Move

Anthropic's Claude Haiku 4.5 delivers major speed and cost savings, while OpenAI and Broadcom announce a massive custom AI chip partnership.

Hey Everyone πŸ‘‹,

John Lindquist here with the 28th issue of AI Dev Essentials!

Claude Haiku 4.5 dropped with impressive cost savings and speed improvements, OpenAI continues their infrastructure expansion with a major chip partnership, and we're seeing remarkable progress in AI's ability to tackle complex scientific problems.

Personally, I've been improving MCP/Hooks workflows in Cursor (and Claude Code) by authoring custom libraries for mcp servers and advanced hooks. These enable a lot of creativity and power around cost-savings and speed by offloading to different models, improving prompts by fetching context in the background, and much more. I'm very excited to share more about this in my Cursor workshop tomorrow.

🚨 THE CURSOR WORKSHOP IS TOMORROW! TICKET SALES END TONIGHT! 🚨 https://egghead.io/workshop/cursor


πŸŽ“ New egghead.io Lessons This Week?

I recently released mcpez, a new library that simplifies MCP (Model Context Protocol) local development and could reshape how you approach agentic development. The library makes it dramatically easier to build custom tools and integrations for AI agents.

πŸ”— Check it out: https://www.npmjs.com/package/mcpez

I've also been talking to the Cursor team about they're hooks setup so that I can push them a little further. I've got a few more lessons lined up that should be released early next week.


πŸš€ Major Announcements

Claude Haiku 4.5: Speed and Cost Efficiency Revolution

Anthropic released Claude Haiku 4.5 on October 15, 2025, delivering near-frontier model performance at dramatically reduced costs and significantly faster speeds.

Performance highlights:

The model excels at computer use tasks, making Claude for Chrome significantly more responsive while maintaining high coding performance standards.

(Anthropic Official, TechCrunch, AWS)

Definitely awesome at taking a plan and executing it. There's a certain feeling with a faster model and quicker feedback loops, where as a human in the loop, you can stay more engaged and more active in the project. So even if it does something wrong, you can give it feedback and quickly turn it around. It's definitely something you miss when waiting for larger, smarter models to work through problems. So I'm a huge fan of faster, smaller models.

OpenAI and Broadcom Announce 10 Gigawatt Chip Partnership

OpenAI and Broadcom announced a strategic collaboration on October 13, 2025, to deploy 10 gigawatts of custom AI chips designed specifically for OpenAI's infrastructure needs.

Partnership details:

This represents OpenAI's most significant move toward vertical integration and infrastructure control.

(OpenAI Press Release, Broadcom Investor Relations)

10 gigawatts is like 5 Hoover Dams. Or 10x the power Doc needed to send the Delorean to the future. Or 2x the power of Los Angeles at its peak. All of that power going into AI. It's so easy to forget the sheer scale of the AI era when I'm sitting behind my computer complaining about AI-generated em dashes.


πŸ› οΈ Developer Tooling Updates

Claude Code Introduces Plugin System in Public Beta

Anthropic launched Claude Code Plugins in public beta on October 9, 2025, enabling developers to install and share curated collections of tools, agents, and automations.

Plugin capabilities:

The plugin system addresses the complexity of managing multiple AI development tools and configurations.

(Anthropic Claude Code)

Finally! Managing all the different MCPs, hooks, and custom commands was becoming a nightmare. Having a plugin system means I can package up my entire development workflow and share it with the team without everyone having to manually configure dozens of different tools.

Vercel v0 Integrates with Slack at Dreamforce 2025

Vercel demonstrated v0's native Slack integration at Dreamforce 2025 on October 15, 2025, enabling AI-powered application development directly within Slack channels.

Integration features:

The integration represents a significant shift toward conversational development workflows.

(Slack Blog, Salesforce Press Release)

I'm really curious what designs might look like when the barrier to feedback is lowered to posting a message to Slack on your phone. It's a little concerning that execs might start stepping on the designers toes, but the good execs should now be able to give much better actionable feedback. I love enabling people to share their ideas and anything we can do to move those ideas from text to visuals is a huge win IMO.


πŸ€– AI Ecosystem Updates

AI Models Achieve Gold Medal Performance in Astronomy Olympiad

GPT-5 and Gemini 2.5 Pro achieved gold medal performance in the International Olympiad on Astronomy and Astrophysics (IOAA), marking a significant milestone in AI's scientific reasoning capabilities.

Performance metrics:

This represents AI's growing capability in advanced scientific domains requiring complex mathematical reasoning.

(arXiv Research Paper, Google DeepMind Blog)

AI is smarter than me. I get it. But I think it's incredible that the leading labs struggle to find people smart enough to check on the work from the AI. I know they keep promising to "cure cancer" and I take those statements with a healthy dose of skepticism, but what if they do...

Google's C2S-Scale 27B Model Generates Novel Cancer Research Hypothesis

Google and Yale University's C2S-Scale 27B parameter model generated and experimentally validated a novel hypothesis about cancer cellular behavior, demonstrating AI's potential for scientific discovery.

Research breakthrough:

The work represents a significant step toward AI-driven scientific hypothesis generation and testing.

(Interesting Engineering, Google Blog)

This is exactly the kind of AI application that gets me excited. Rather than just optimizing existing processes, the AI actually generated new scientific insights that researchers could test and validate. It's a preview of how AI might accelerate scientific discovery by exploring hypothesis spaces that humans might not think to investigate. We don't know what we don't know!

NotebookLM Introduces Visual Styles for Video Overviews

Google launched six new visual styles for NotebookLM's Video Overviews on October 13, 2025, powered by the Nano Banana image generation model.

New features:

The update significantly enhances NotebookLM's content creation capabilities for educational and professional use.

(9to5Google, Google Labs Blog)

My main struggle with NotebookLM (which I love dearly) is the fact that I'm learning so many new things all at the same time. I never make time to follow-up or study to help solidify the knowledge. I just expect the AI to be able to recall it for me. I'm loving the video overviews to help the knowledge stick more during the first pass.


⚑ Quick Updates

OpenAI Announces Relaxed ChatGPT Content Restrictions

(Official Statement)

Amp Code Becomes Free with Ad-Supported Model

(Amp Code Blog)

Andrej Karpathy Releases NanoChat Training Pipeline

(GitHub Repository)

Microsoft Hires Jared Palmer as VP of CoreAI

n8n Raises $180M Series C at $2.5B Valuation

(n8n Press Release)

Google Launches Gemini Enterprise for Workplace AI

(CNBC, Google Cloud Blog)


✨ Workshop Spotlight

🚨 TODAY IS THE LAST DAY TO BUY TICKETS! 🚨 Cursor Workshop - Friday, October 17th, 2025

Become a Cursor Power User

In just a single day you'll set up shared rules, optimized workflows, and custom agents to help you ship AI-generated code with confidence.

Date: October 17, 2025 Time: 9:00 AM - 2:00 PM (PDT) Platform: Zoom

What You'll Learn:

The AI development landscape continues evolving rapidly with new models, tools, and best practices. This workshop teaches fundamental principles and repeatable workflows that work regardless of which tool you useβ€”Cursor, Claude Code, WindSurf, or the next breakthrough.

Essential Skills:

Register: https://egghead.io/workshop/cursor


Read this far? Share "AI Dev Essentials" with a friend! - https://egghead.io/newsletters/ai-dev-essentials

https://egghead.io