AI Dev Essentials #7: Microsoft's AI Blitz, Google's NotebookLM Shines & New Coding Agents

Hey Everyone 👋,

John Lindquist here with the seventh issue of AI Dev Essentials! It's been another week of whirlwind developments, with AI further embedding itself into the very fabric of our developer environments and workflows. Microsoft, in particular, made some significant announcements around VS Code, GitHub, and even Windows itself. We're also seeing a continued flourishing of the MCP ecosystem and new tools designed to enhance our productivity.

While the Microsoft news is big, I think the sleeper hit of all the recent announcements might just be the NotebookLM mobile app from Google. If you've spent any time on your phone trying to gather information from social media, searching, and attempting to consolidate it all into one place, you know how frustratingly difficult and how easy it is to lose track of things. The NotebookLM mobile app added the ability to share anything you're viewing directly to the app, allowing you to build up these focused "notebooks." These can then be easily searched with AI, used to generate reports, and help you rediscover all those bits of information you previously viewed. I'm really hoping as Google progresses, they can announce even deeper integration. I had a brief discussion on X with the NotebookLM team lead about adding the ability to include screenshots. Imagine: anytime you're looking at your phone, whether it's news feeds, dev articles, something AI-generated, a quote from a book, or anything else you want to capture, just taking a quick screenshot and dropping it into NotebookLM. This is essentially the primary use-case for my phone and I couldn’t be happier!

Now on to the updates:

New egghead.io Lessons This Week

Automatically Improve Cursor Rules Using Custom Prompts(egghead.io)
Learn to refine your Cursor rules or AI prompts by referencing a dedicated 'prompt improve' rule with Cmd+K, automatically applying best practices for clarity and precision. You can then use the agent (Cmd+L) to compare changes and understand how the language was improved.

Clean up Legacy Functions for Testability in Cursor (0.50+) with cmd+k(egghead.io)
Discover how to refactor legacy code in Cursor by using Cmd+K with the 'edit full file' option (Cmd+Shift+Enter) to extract logic into new, top-level, testable functions. This lesson also covers using 'fix in chat' (Cmd+Shift+D) for quick error resolution.

🤖 AI Coding Agents: Automated GitHub Workflows

A new wave of AI agents is emerging, designed to autonomously work on tasks within your GitHub repositories, from fixing bugs to proposing pull requests.

• OpenAI Introduces Codex Agent

  OpenAI has launched a research preview of Codex, a cloud-based software engineering agent powered by codex-1 (an o3 version optimized for software engineering). It can write features, answer codebase questions, fix bugs, and propose PRs, operating in its own cloud sandbox. Accessible via ChatGPT, it can be guided by AGENTS.md files in your repo and is rolling out to Pro, Enterprise, and Team users, with Plus and Edu coming soon. (OpenAI Announcement(openai.com))

• Google Unveils Jules, an Experimental Coding Agent

  Jules is an experimental coding agent from Google that integrates with GitHub to help fix bugs, add documentation, and build features asynchronously. Users can assign tasks, review generated plans, and receive notifications, allowing them to focus on other work while Jules handles the coding. (Jules Website(jules.google))

• GitHub Copilot Gets a New Coding Agent

  GitHub is embedding a new coding agent directly into GitHub Copilot. Assignable via GitHub issues or prompted in VS Code, this agent uses GitHub Actions to create a secure development environment, works on tasks, and pushes commits to a draft PR, providing session logs for transparency. It's designed for low-to-medium complexity tasks and is available for Copilot Enterprise and Pro+ customers. (GitHub Blog(github.blog))

  I've tried all three of these. And the main goals and main wins that I would love to see from these are asking it to generate more test coverage, generate more integration test coverage, running tests, creating benchmarks and running benchmarks and optimizing functions and making sure all the tests pass and kind of things of that nature where if all the tests pass, then I would have confidence in automated refactorings that would improve the code base either through improved modularity, improved optimizations, improved code cleanliness, and other things that I'd love for AI is just to be able to do while increasing test coverage and reliability. Unfortunately, none of these services have really been able to pull that off yet. While I have been impressed with them finding small bugs as far as cleaning up some basic spelling mistakes or variable names and sometimes even suggesting some good significant changes, my level of trust with these running just isn't quite there yet. And I fully understand that after it submits the PR, it could run the tests, but then I'd like it to be able to check on the pull request and to see if all the conditions were met for the pull request. And if they weren't, then go back and look at it again. All of these tools are basically a, “ask for a fix and then leave your desk and check on a later” sort of tools. And if I'm going to wait a long time for something, while I definitely appreciate, and I think this is an awesome step in the right direction, I'd also love to see this being pushed, all of these being pushed much, much further than they currently are.

🚀 Microsoft's AI Offensive: VS Code, GitHub & Windows

Microsoft unveiled a suite of announcements this week, signaling a deep commitment to integrating AI across its developer platforms.

• VS Code to Become an Open Source AI Editor

  Microsoft announced plans to make VS Code an open-source AI editor, emphasizing that AI development should align with VS Code's core principles of being open, collaborative, and community-driven. They've also provided an FAQ(code.visualstudio.com) for more details.

  I think this is a bold move in the face of Cursor and Windsurf, especially with Windsurf recently being acquired for $4 billion from OpenAI. The fact that Microsoft can counter by open sourcing the significant process they've made around integrating Copilot into VS Code is pretty awesome. I can't think of the number of times I wish that Cursor was open source so that I could tweak some of the conversation menus, make little fixes like resume this conversation or increase the limit of tool calls it makes and things of that nature. So I'm really excited to see where people can take or where the community can take an open source AI editor. I know there are some other efforts for open source editors, but having the original VS Code release kind of the official version is just straight up awesome.

• GitHub Copilot's New Coding Agent

  (Covered in the new "AI Coding Agents" section above)

  I'm excited for being able to try and fix issues directly from GitHub inside of by assigning issues to an AI. I think this will make good first passes at issues. And even some of the workflows I think we'll see streamlined are pull request workflows inside of VS Code and related tools where you can automatically check out the potential fix and then tweak it locally to test it and make sure it's correct. My only concern is that because of how low the effort will be to create these pull requests, we might see a huge influx of issues in pull requests. And it'll be interesting to see how the moderation comes into play from both GitHub side and from open source maintainer side.

• MCP on Windows Introduced

  Microsoft is bringing the Model Context Protocol (MCP) to Windows. This could open up new avenues for AI agents to interact with the Windows environment and applications. See the announcement(msft.it).

  I'm not a Windows user. I use Mac for everything. And it'll be fascinating to see deeper integration into OS level tools. Microsoft Copilot has seen Microsoft's integration into their office suite and the way they're rolling everything out there has always felt a little disjointed. And I'm not sure if this is just kind of adding more to the pile or how this is all going to be smoothly integrated together. I have been much more impressed with Google's approach with Gemini. Because if you weren't aware, you can open up Gemini and use the at symbol to reference your Gmail and docs and everything from Gemini and run queries against it. And it's already very useful. So I guess only time will tell. It'll be interesting to see the next major version of Windows and how AI driven it will be.

• New 'Edit' Command-Line Text Editor for Windows

  A new command-line text editor called "Edit" has been introduced on Windows. Check it out on GitHub(github.com).

  I just added this in here because I think it's a really interesting project. It's really cool that it's open source. And it's one of those types of tools that you never really think much about in the effort that goes into it. So if you have some extra time, poke around the repo and take a look at something that's probably pretty foreign to you. Something you'll be familiar with from a user interface perspective and how you would use it. But something that's probably pretty foreign from how you would architect an application.

📢 Google's Big Week: I/O Kicks Off & Key Updates

Google I/O kicks off today (May 20th!), and there's already been a flurry of news from them over the past week.

• NotebookLM Mobile App Rolling Out

  As mentioned in the intro, the NotebookLM mobile app is now available, allowing users to add sources from websites, PDFs, or YouTube videos via the share icon. It also features offline audio overviews and interactive audio overviews when connected. (Source(x.com))

• AlphaEvolve Unveiled

  Google DeepMind introduced AlphaEvolve, a Gemini-powered coding agent designed to help discover new algorithms and optimizations. Logan Kilpatrick noted this as great progress in core algorithm improvements using Gemini. (Source(x.com), DeepMind Blog(deepmind.google))

  This is the very beginning of AIs improving their own algorithms, at least publicly announced. This could signify the singularity event. So only time will tell what this actually means and the impact it'll have for the future.

• AI Studio Gets Generative Media Upgrades

  A new generative media experience in Google AI Studio brings together Veo 2, Gemini 2.0 native image generation/editing, and Imagen 3. All of these models are free to try in AI Studio and ready for developers to build with via the API. (Source(x.com))

  As always, I love AI Studio. I love that it's free and it's just a wonderful resource that I use all the time. So any improvements and allowing people to generate images and videos is just awesome. I love how Google is supporting the community with this tool. It's a great way to do it.

• Pre-I/O News:

  • Android 16 redesign with Gemini integration. (Android Blog(blog.google))
  • AlphaEvolve coding agent (mentioned above).
  • YouTube updates: Brandcast event, Shorts revenue parity in the US (Variety(variety.com)), a new top 100 US podcasts chart (YouTube Blog(googleusercontent.com)), and an exclusive NFL game broadcast (Hollywood Reporter(hollywoodreporter.com)).
  • Google One reaching 150 million subscribers. (Reuters(reuters.com))

  (Full Recap Thread(x.com))

🛠️ Dev Tooling, Frameworks & Ecosystem

The ecosystem continues to churn out new tools and resources for AI-assisted development, with a strong focus on MCP and agent capabilities.

• Free Open Source MCP Server Boilerplate

  Ian Nuttall has open-sourced his MCP server boilerplate, refunding all previous orders. It includes features for user authentication (Google/GitHub), hosting on Cloudflare, and Stripe integration for various billing models. (Source(x.com))

• Supermemory Infinite Chat API

  supermemory launched an "Infinite Chat API" that aims to extend the context length of any model while reducing token usage and cost. It acts as a proxy, maintaining context and sending only necessary information to the LLM. (Source(x.com), Docs(docs.supermemory.ai))

  It's interesting to see services unlocking unlimited memories and hosting them as services. I would kind of have a difficult time signing up for one of these and getting locked in, thinking of the migration process from one of these services over to one of the base providers. Who knows? Maybe one of them will be acquired by one of the big dogs.

• Amp AI Now Open to Everyone

  The AI-powered code search and understanding tool, Amp, has removed its waitlist and is now available for everyone to sign up. (Source(x.com), Website(ampcode.com))

  This one is getting decent reviews, but make sure you have your wallet open because it doesn't have any restrictions on context or any restrictions on spending. So once you sign up for it, it's going to cost you a ton of money, but I've heard nothing but good things. So if you have extra cash to burn, then it's definitely an AI editor/agent to check out, with the power of Sourcegraph behind it.

• OpenMemory MCP for Private, Persistent Memory

  Taranjeet announced OpenMemory MCP, a locally-running, private memory layer for MCP-compatible clients, powered by mem0.ai. It allows AI tools like Cursor, Claude Desktop, and Windsurf to share persistent context securely on the user's machine. (Source(x.com), Learn More(mem0.ai))

  Same thoughts as Supermemory above 😅

• Coderabbit Extension for Cursor

  There’s a coderabbit extension for Cursor that offers free, context-aware code review for all languages, including line-by-line reviews and one-click fixes. (Source(x.com), Get Started(coderabbit.ai))

  CodeRabbit is an extremely popular PR review tool that you can use for free on open source projects. So it's definitely something worth checking out as more PRs start coming in from other people or from even AIs. This is one of those scenarios where you'll see an AI PR come in and then CodeRabbit can be your AI PR reviewer that reviews it. So it's just AI from start to finish. But definitely worth checking out. And something I'm using more and more. Right now, the teams I'm working on are small enough that we still kind of hop on calls together when going over PRs. But it's always nice to have kind of a second set of AI eyeballs on pull requests.

• Personal Logseq MCP Server

  Joel Hooks shared that he's running a personal MCP server to query his Logseq notes, creating a unique way to interact with his knowledge base. (Source(x.com), GitHub Repo(github.com))

  Joel shared one of his MCP servers based on someone's request for what's a useful MCP server. And I do love the idea of taking notes, especially using MCP, especially from conversations in Cursor and other AI tools where there's so much valuable information that I feel like we lose in a lot of the AI conversations with our editors. Like once we get something working, we need to find better ways of extracting why that worked and what we fixed to build up like these robust systems around keeping our software improving and avoiding regressions and such. And so any step towards using AI to capture that information, whether in notes or whether in tests or in any way is a welcome addition to me.

💰 AI Model Usage & Cloud Updates

• Cursor o3 Max Usage

  Josh Pigford mentioned that his team at Maybe spent over $250 in credits on o3 max in Cursor within 24 hours, calling it "basically witchcraft". (Source(x.com))

  I've seen more and more people talk about how great Cursor Unlimited or Cursor o3 Max is. You can easily end up spending hundreds of dollars a day, but you'll start knocking out tasks like crazy. Unfortunately, it's sad to see that right now we are in a state where you can literally, “throw money at the problem” to fix a lot of difficult issues. I'm really hoping the prices come down on these services because it's just not reasonable for the vast majority of the population to be spending hundreds of dollars on AI tokens per day.

• OpenAI Codex CLI Improvements

  OpenAI updated its Codex CLI, allowing sign-in with ChatGPT to connect an API org and introducing a new codex-mini model optimized for low-latency code Q&A and editing. (This is also related to the new OpenAI Codex Agent mentioned earlier). (Source(x.com), Blog Post(openai.com))

  If you have a pro/team account, go and grab some free credits and give it a spin. Codex is really powerful, but it’s just really, really slow for my tastes and I prefer a much tighter feedback loop.

💡 Perspectives

• Untapped Model Capabilities

  Matt Pocock quoted Microsoft CTO Kevin Scott: "Models are more capable than what they're being used for," agreeing that app developers have significant untapped power. (Source(x.com))

  I do think a lot of the AI models that we have out there are just being used incorrectly. And it takes a long time to learn how to use them. If you've done any sort of prompt engineering or writing evaluations for the AIs, you know that sometimes awesome things, unexpected awesome things happen. And it'll take a while to learn how to properly use the models, even the ones that we have today.

✨ Workshop Spotlight: Conquer the Complexity of Cursor ✨

Master practical AI development workflows in Cursor. This hands-on workshop covers Agents, multi-file analysis, effective prompting, Cursor rules, and strategies for handling AI failures.

When: Thursday, May 22, 2025, 9:00 AM to 2:00 PM (PDT)

Where: Zoom (Live Q&A included)

Investment: $249

Read More(egghead.io) | Register Now(buy.stripe.com)

(Team training also available)

That’s the scoop for this issue! The integration of powerful AI directly into our core development tools is accelerating, and it’s exciting to think about the new workflows and efficiencies this will unlock.

If you have any feedback or questions, hit reply! Always happy to chat about the latest in AI dev tools.

John Lindquist
egghead.io(egghead.io)