Learn Openai, Ramda and Solid

John Lindquist

AI Dev Essentials #8: Claude's Costly Lessons, Google's AI Flood & Workshop Alert 🚀

Claude Max Mode in Cursor is powerful but pricey—one vague task can rack up costs fast. Learn how to control it and explore the latest AI dev tools.

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

AI agents are reshaping coding in GitHub and VS Code, while NotebookLM’s mobile app quietly revolutionizes how developers collect and recall research on the go.

Hey Everyone 👋,  
  
John Lindquist here with the **fifth issue of AI Dev Essentials!** This week, I've been relying more on GPT-4.1 in Cursor (which seems vastly improved since I last tried it) when building CLI and node tools. GPT 4.1 seems more aggressive in using tools as a pseudo reasoning step, while Gemini 2.5 Pro often jumps straight to code changes based on its "real" reasoning. I'm finding that GPT 4.1 in Cursor is becoming legit, especially with a "please explain that back to me" workflow. The speed allows for almost instant alignment. This "time to alignment" is so critical. It's that moment you \*know\* the AI is on the same page after a brain dump. On another note, programming is getting weird! I had a moment working on an MCP process manager where all the code was correct, but my \*English descriptions\* of what it was doing were too technical and thus misleading. Wild times!  
  
Big news: **Gemini 2.5 Pro upgrade** launched a couple hours ago. I'm super excited for this to land in Cursor, hopefully even later today! For now, you can try it out in AI Studio. I can't wait to see the improvements in Cursor's agent mode, especially since the tool calling is reportedly vastly improved. Beyond that, the major themes this week include impressive new **generative AI capabilities** for voice, a continued surge in **developer tooling and frameworks** (especially open source), and ongoing discussions around **prompting best practices and AI model quirks.**  
  
Let's dive in!

* * *

🎓 New egghead.io Lessons This Week
-----------------------------------

*   [When to Use Cursor Rules Instead of MCPs](https://egghead.io/when-to-use-cursor-rules-instead-of-mcps~zazdq)
    
    This lesson demonstrates creating a Cursor Rule for a CLI tool (\`ghx\` for scraping GitHub examples) to make it available to the agent. It highlights how agent requested rules can be a simpler alternative to MCPs for tool integration, offering more control over context.
    
*   [Brainstorming and Planning with Mermaid Diagrams in Cursor](https://egghead.io/brainstorm-and-planning-with-mermaid-diagrams-in-cursor~9iu69)
    
    Learn to use Cursor Rules to enable Mermaid diagram generation. The lesson shows how to create and iterate on diagrams (like basic to do app flows and sequence diagrams for optimistic updates) visually within Cursor, providing a way to plan before coding.
    
*   [Create Complex Cursor Shortcuts to Run Commands and Generate AI Commits](https://egghead.io/create-complex-cursor-shortcuts-to-run-commands-and-generate-ai-commits~7arof)
    
    This lesson walks through setting up a custom keyboard shortcut (Command+Shift+S) in Cursor's JSON settings to run a sequence of commands: stage all git changes, use Cursor to generate a commit message, commit, and then push/pull.
    

* * *

💰 AI Coding Assistant Valuations & Acquisitions Skyrocket
----------------------------------------------------------

_The AI coding assistant space is on fire with some massive financial moves happening._

*   **Anysphere (Cursor IDE) Hits $9 Billion Valuation**
    
    Cursor's parent company, Anysphere, reportedly secured a new $900 million funding round, pushing its valuation to an estimated $9 billion. Key investors include Thrive Capital, Andreessen Horowitz (a16z), and Accel. This is a huge jump from their $2.5 billion valuation in January 2025, showing incredible investor confidence. (Source details from PitchBook: "Maker of AI-coding software Cursor is doubling its valuation every 8 weeks", citing The Financial Times; and Crunchbase News: "AI-Powered Coding Tool Anysphere Lands $900M At $9B — Report".)
    
*   **OpenAI Reportedly Acquires Windsurf (formerly Codeium) for ~$3 Billion**
    
    In what would be its largest acquisition, OpenAI is said to be acquiring Windsurf. Windsurf, an AI-assisted coding tool, was previously valued around $1.25 billion. This move clearly signals OpenAI's intent to bolster its position in the AI developer tools market. (Source details from Mashable citing Bloomberg: "OpenAI is buying Windsurf for $3 billion. What does that mean for ChatGPT?"; and Investing.com citing Reuters/Bloomberg: "OpenAI agrees to buy Windsurf for about $3 billion, Bloomberg News reports".)
    
    My take: It seems like OpenAI, above all, needed the team behind Windsurf. The value of people talented in code generation is through the roof right now. But they also needed a product ASAP, so it's essentially like they're paying for same day shipping, even with this incredible amount of money. And it boggles my mind that Cursor's valuation is estimated at $9 billion. I know a lot of us developers look at both Windsurf and Cursor and kind of think to ourselves that we could have forked VS Code and built these things. So, seeing an incomprehensible amount of value being assigned to what looks like something that we could build if we just would have dedicated our time to it is always pretty shocking.
    

* * *

Here are the highlights that caught my eye this past week:

🚀 Model & Platform Updates
---------------------------

_This week saw some impressive new releases and updates in the model space, particularly around generative capabilities for voice, alongside ongoing platform refinements._

*   **Gemini 2.5 Pro Upgrade: Now with Better Coding and Function Calling.**
    
    Google announced a significant upgrade to Gemini 2.5 Pro. It boasts improved performance in coding (especially frontend web dev, editing, and transformation) and more reliable function calling. The new model (\`gemini-2.5-pro-preview-05-06\`) automatically replaces the previous version with the same pricing and rate limits. [Details](https://developers.googleblog.com/en/gemini-2-5-pro-io-improved-coding-performance)
    
    This is the update I've been waiting for! Reliable tool/function calling is a massive unlock for agentic workflows. Eager to put this through its paces in Cursor.
    
*   **Minimax's High Ranking Text to Speech Models on Replicate.**
    
    Minimax's speech-02 models (speech-02-hd and speech-02-turbo) and voice cloning capabilities are now available on Replicate. They are touted as top ranking TTS models. [Replicate Blog](https://replicate.com/blog/minimax-text-to-speech)
    
*   **Gemini 2.0 Flash Photo Editing.**
    
    Google's Gemini 2.0 Flash is showcased for its ability to edit photos using natural language, without needing traditional tools like layers or brushes. (Mentioned by Rishabh, [Source](https://x.com/Rixhabh__/status/1919641187696418896))
    
*   **Grok 3.5 Beta Coming Soon.**
    
    Elon Musk announced that Grok 3.5 will have an early beta release next week for SuperGrok subscribers. He claims it can accurately answer technical questions by reasoning from first principles. ([Source](https://x.com/elonmusk/status/1917099777327829386))
    

🛠️ Developer Tooling, Frameworks & Ecosystem
---------------------------------------------

_The developer ecosystem continues to expand with new open source tools for data extraction, updates to agent frameworks, and interesting discussions on platform choices and best practices._

*   **ContextGem for Structured Data Extraction.**
    
    A new open source LLM framework, ContextGem, aims to simplify extracting structured data and insights from documents with minimal Python code. ([GitHub ContextGem](https://github.com/shcherbak-ai/contextgem) via Sumanth)
    
*   **Cloudflare Agents SDK Update: Upcoming Auth and State Handling.**
    
    An upcoming PR for the Cloudflare Agents SDK will enhance Agent classes to handle OAuth, redirect URLs, and reconnections out of the box. This makes every Agent an MCP client with state management. (Mentioned by Sunil Pai, [PR Link](https://github.com/cloudflare/agents/pull/254))
    
    Authorization has been one of the biggest challenges for MCPs, specifically because the MCP sits behind the AI agent. Most things are just configured in a simple JSON file defining your MCP servers. Triggering OAuth workflows inside UIs, which are essentially streaming text and working on other tasks, has been weird. Any sort of standardization and convenience they can add to authentication is very welcome, in my personal opinion.
    
*   **Persistent File Storage for Cloudflare Durable Objects Demo.**
    
    A demonstration of a real filesystem and terminal for Cloudflare Durable Objects using a Rust based chunked SQLite FUSE strategy ported to \`ctx.storage.sql\` and an xterm.js frontend. ([Demo](https://dofs.verifried.workers.dev/), [Tweet](https://x.com/benallfree/status/1919224851409379786))
    
    I love that as more new developers come on board, we can give them paradigms they're familiar with, like file systems and basic terminals to manipulate files. I think one of the biggest holdups for developers with things like images on their websites is they're just so unfamiliar with the various concepts around routing and where things are hosted. Any effort that brings existing paradigms that beginners understand to complex subjects is welcome, in my opinion.
    
*   **Stripe MCP Monetization.**
    
    Stripe now allows MCP developers to monetize their MCPs with subscriptions or usage based billing. It's compatible with various clients including Cloudflare's Agents SDK. ([Stripe MCP Payments](http://stri.pe/go/mcp-payments) via Jeff Weinstein)
    
    Stripe is obviously the biggest player in allowing people to spend money on things, so it's no surprise they're introducing ways to make money with MCPs. I'm definitely curious to see what sort of services will pop into existence now that you could request them from AIs. I'm in no hurry to pay for any MCP service myself, and I don't have any to recommend yet, but I'll definitely keep my eyes out for what may manifest from this.
    
*   **Cloudflare Workers Monorepo Template.**
    
    Jacob Hands shared a template for creating Workers monorepos. It's inspired by the \`mcp-server-cloudflare\` repo which deploys multiple Workers from one repo with shared packages. ([GitHub Template](http://github.com/jahands/workers-monorepo-template) via Jacob Hands)
    
*   **A Story of Migrating from Vercel to Cloudflare.**
    
    A developer shared their experience migrating their entire infrastructure to Cloudflare. They cited reasons related to control, security, performance, and pricing. ([Source](https://x.com/Osintly/status/1917964184198823951))
    
    I've seen a lot of people try to do this, especially with the perception that Next.js kind of locks you into Vercel. So I just think it's valuable for people to realize that options are out there and that a lot of your projects are more portable than you think they are.
    
*   **New MCP Authorization Spec.**
    
    A new MCP authorization spec has been introduced. It defines the MCP server as an OAuth "resource server" and outlines the authentication flow. Den Delimarsky provided a blog post with more details. ([Den Delimarsky's Blog](https://den.dev/blog/new-mcp-authorization-spec) via Yenkel)
    
    It'll be interesting to watch any of the MCP specs evolve over time. I know there's been a lot of work done with streaming. Authorization is obviously going to play a huge part. It'll be interesting to see all these pieces of MCPs fitting together and how all their auth will sync up, where the walls are built, and what they can and cannot do to communicate with each other. I can't imagine a system of roles and identifications for the various MCPs and how they allow each other to speak to one another is going to be critical for MCP moving forward.
    

💡 Prompting, Workflows & Best Practices
----------------------------------------

_This week brought fresh perspectives on interacting with LLMs, managing context, and structuring UI generation prompts._

*   **"AI Rivalry" Prompting Technique.**
    
    Greg Isenberg shared a technique of pitting different AI models (ChatGPT, Claude, Grok) against each other. You feed one's output to another with critical feedback to elicit better results. ([Source](https://x.com/gregisenberg/status/1918661279641723158))
    
    There's so much potential left to unlock when multiple models work with and against each other. It's still a little cost prohibitive to use this for a lot of the tools you're probably building by yourself. But I love any sort of back and forth or ping pong or even parallel agents coming to conclusions and fighting over what the best answers will be. I can only imagine what these techniques unlock. Hopefully someone will send me a credit card to run a bunch of experiments.
    
*   **Theo's Take on Major AI Models' Quirks.**
    
    Theo (t3.gg) shared a list of issues and frustrations with various AI models and providers including Gemini, Claude, ChatGPT, and Mistral. He covers setup, cost, and availability. ([Source](https://x.com/theo/status/1918810601968513249))
    
    I really agree with Theo's breakdown here. I've had a very similar experience trying out each of the models. So, definitely worth checking out his list. It's just a well written, very concise list.
    
*   **Tips for LLM Tools Editing Files with Diffs.**
    
    Matt Pocock initiated a discussion seeking examples and best practices for implementing LLM tools that edit files using diffs in TypeScript. ([Source](https://x.com/mattpocockuk/status/1918758488777408528))
    
    I hopped in on this discussion with Matt about my personal experience working with diffs. It's just so challenging to get LLMs to produce precise diff worthy code based on line numbers and white space, and then trying to match that up with the original file. It was a fascinating thread to read about how some people are solving the problem by using a second LLM trained specifically on diffing two files and essentially rewriting the entire file. It's a problem space I spent a lot of time in last year, but I'm just going to leave it up to all of these IDEs and tooling who are so far beyond where I got to now that it's not worth spending time on for myself.
    
*   **Prompting Guide for Lovable.dev for UI Generation.**
    
    Felix Haas shared a step by step guide on how to prompt Lovable.dev for better UI generation. It covers high level descriptions, look and feel, brand guidelines, UI library specification, and component details. ([Source](https://x.com/felixhhaas/status/1918316680108736775))
    
    I have some non developer friends using Lovable, and they seem to be really enjoying this, so I'm sending this over to them. This newsletter is more focused on a bit more experienced developers and what they would expect to read about, but this is definitely something I'm going to shoot over to a few people I know.
    
*   **Survey of AI Agent Protocols.**
    
    Elvis (omarsar0) highlighted key takeaways from a survey paper on AI Agent Protocols. He discussed the agent internet ecosystem, timelines, protocol overviews (MCP, A2A), evaluation considerations, and use case analyses. ([Source](https://x.com/omarsar0/status/1918723145923453022))
    
    I know that MCP gets so much attention, and it's interesting how it's taken over so quickly, but there are many other concepts to pay attention to, such as the A2A protocol. So it's definitely worth reading through how some of these things, when and why you would use different protocols.
    
*   **Vercel AI SDK \`useAssistant\` Hook.**
    
    Sam Denty pointed to the \`useAssistant\` hook in the Vercel AI SDK for building AI assistants. ([AI SDK Docs](https://ai-sdk.dev/docs/reference/ai-sdk-ui/use-assistant))
    
    The \`useAssistant\` hook was something I wasn't aware of, and Sam shared it with me. For an awesome hook for building out web based AI tools, I didn't know it existed, and I thought I'd pass it along because it covers a lot of scenarios that I would be building for web cases.
    

👨‍💻 X Posts from the Cursor Team
----------------------------------

_Folks from the Cursor team continue to share valuable insights and tips for getting the most out of the AI-first IDE._

*   **Effective Context Management in Cursor.**
    
    Eric Zakariasson (from the Cursor team) provided tips on managing context in Cursor. He emphasized the combination of intent and existing codebase state, using surgical @ mentions, rules for patterns, and MCP for external systems. ([Cursor Docs](https://docs.cursor.com/guides/working-with-context) via Eric Zakariasson)
    
*   **New Cursor Guide for Web Development.**
    
    Eric Zakariasson (from the Cursor team) also published a new guide for web development workflows in Cursor. ([Cursor Docs](https://docs.cursor.com/guides/tutorials/web-development))
    
*   **Cursor Rules for Better Results.**
    
    Ian (shaoruu) (from the Cursor team) shared examples of Cursor rules they use. These include specifying libraries, listing UI components, defining commands, and using "Auto Attach" for specific file contexts. ([Cursor Docs](https://docs.cursor.com/context/rules) via Ian)
    

✨ 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 8, 2025, 9:00 AM to 2:00 PM (PDT)

**Where:** Zoom (Live Q&A included)

**Investment:** $249

[Read More](https://egghead.io/workshop/cursor) | [Register Now](https://buy.stripe.com/3csdSh2GTdhucTedQW)

_(Team training also available)_

That’s the scoop for this issue! As a final note, I'm excited for Google I/O in a couple of weeks. Since they're already releasing a huge update to my favorite model, and there are rumors around other models that are even better, I just can't wait to see what they have in store for us.  
  
If you have any feedback or questions, hit reply! Always happy to chat about the latest in AI dev tools.  
  
John Lindquist  
[egghead.io](https://egghead.io/)

AI Dev Essentials #5: 🤯 Cursor Valued at $9B, Gemini 2.5 Pro & More AI Updates

GPT-4.1 and Gemini 2.5 Pro are changing how devs build tools in Cursor. Plus: massive funding for Cursor, OpenAI acquires Windsurf, and new agentic dev tools.


Hey everyone 👋,  
  
John Lindquist here with the **fourth issue of AI Dev Essentials!** After months of teaching Cursor workshops, I've really enjoyed iterating on the AI assisted feedback loops: using AIs for spec work, diagramming, task management, and integrating them every step of the development process. I've even built a local MCP server that can monitor dev servers and test runners, checking on their progress to remove my biggest friction points when letting AI agents work autonomously. Expect an egghead.io lesson on it any day now. 🔧✨

🎓 New egghead.io Lesson
------------------------

[AI Driven Design Workflow: Playwright MCP, Screenshots, Visual Diffs & Cursor Rules](https://egghead.io/ai-driven-design-workflow-playwright-mcp-screenshots-visual-diffs-and-cursor-rules~aulxx)(egghead.io)

Walk through an automated loop where Cursor uses Playwright screenshots + Pixelmatch visual diffs to iteratively nudge your UI toward a target design.

🤖 Model Mania & Platform Shifts
--------------------------------

_The model landscape continues to shift rapidly, with major players releasing new contenders and tweaking existing ones based on user feedback and performance goals._

*   **OpenAI Rolls Back “Sycophantic” GPT 4o Update**
    
    OpenAI reverted last week’s GPT 4o tweak after users complained the model became an over eager yes man. The post explains how they detect and curb flattery bias. [Details](https://openai.com/index/sycophancy-in-gpt-4o/)(openai.com)
    
    It's fascinating when AI models get pushed so far by providers that public backlash forces a rollback to a previous state. This serves as a strong reminder that AI is software; it can be reverted, and users grow to expect predictability not just in information, but also in behavior and tone.
    
*   **Alibaba Unveils Open Weight _Qwen3_ Family**
    
    Eight new dense/MoE models (0.6B 235B) land with strong coding & math scores. The 235B flagship rivals closed source giants while staying fully open weight. [Benchmark blog](https://qwenlm.github.io/blog/qwen3/)(qwenlm.github.io)
    
    It's great to see more open source models being released. Personally, I tend to prefer the smartest models available for planning and coding tasks, and often lack the home hardware to run larger open models effectively. Still, these releases fuel hope for powerful on device models eventually rivaling giants like Gemini 2.5 Pro and O3.
    
*   **Gemini API Adds “Reasoning Effort” Knob**
    
    Google’s OpenAI compat endpoint now lets you pick _low → high_ reasoning, making it trivial to swap between Flash speed and deep thought. [Docs](https://ai.google.dev/gemini-api/docs/openai)(ai.google.dev)
    
    The distinction between questions needing quick answers versus deep thought is interesting. It's notable how AI models sometimes seem to first assess if a question requires deep thought before actually processing it, much like humans do. We instantly recall basic trivia but need to search our experiences for more complex answers.
    
*   **Rumor Mill: o3 pro Imminent?**
    
    Sam Altman mentioned an "o3 pro" tier a couple of weeks ago, fueling anticipation that a paid model between o3 and GPT 4o could drop any day now. Keep an eye on the pricing page. [Tweet](https://x.com/koltregaskes/status/1917143793649828320)(x.com)
    
*   **DeepSeek R2 Leak Hints at 1.2T Param Bargain LLM**
    
    Slides suggest a Huawei trained hybrid MoE vision model that’s 97% cheaper per token than GPT 4o, if the specs hold water, brace for price wars. [Leak thread](https://x.com/OpulentByte/status/1916195106580828344)(x.com)
    
    This lands squarely in the rumor mill and hasn't been confirmed. However, DeepSeek R1 was super impressive. If R2 lives up to these leaked specs, it has the potential to significantly disrupt the larger model landscape.
    

🛠️ Dev Tooling, Education & Ecosystem
--------------------------------------

_Beyond the models themselves, the surrounding ecosystem of tools, educational resources, and key personnel moves continues to evolve at pace._

*   **React Creator Jordan Walke Joins Replit**
    
    Walke says Replit’s AI first IDE feels like the early Facebook days and could onboard “the next billion programmers.” [Announcement](https://x.com/jordwalke/status/1915103101205594211)(x.com)
    
    I personally don't use full app generators like Replit, Lovable, or Bolt much, but I always keep a close eye on their announcements to see how far they push the tech. As an experienced developer, I haven't had much luck getting them to build apps where I don't know the final shape, but I can only imagine the feeling for a non programmer launching something from these tools. Definitely curious what the team will build next.
    
*   **Microsoft Drops “Generative AI for Beginners” Series**
    
    Free 18 episode curriculum covers LLM basics, responsible AI, and prompt craft, perfect onboarding material. [Watch](https://learn.microsoft.com/en-us/shows/generative-ai-for-beginners/)(learn.microsoft.com)
    
*   **OpenAI Launches Free “Academy” Courses**
    
    A new hub of AI courses, from intro to deep learning, built with universities like Georgia Tech, all 100% free. [Explore](http://academy.openai.com/)(academy.openai.com)
    
*   **GitHub Lists for Repo Curation**
    
    Use Lists to group related repos into smaller, AI friendly contexts, handy for prompting or onboarding. [Demo](https://x.com/janwilmake/status/1915338073061167487)(x.com) | [Example](https://github.com/janwilmake/forgithub.lists)(github.com)
    
    I didn't even know GitHub Lists was a feature! I'm definitely exploring ways to use AI to bring multiple repos together, since so many products are composed of different repositories. This looks like something fun to play with.
    
*   **Music AI Sandbox Upgrades (Lyria 2)**
    
    DeepMind’s tool now lets musicians generate, extend, and edit tracks with higher fidelity audio, all in browser. [Announcement](https://x.com/GoogleDeepMind/status/1915421048171856047)(x.com)
    
    Google getting into generative music tools for musicians is a welcome surprise. Their Veo 2 video generation is top tier, so you can only imagine the potential if they combine their music and video generation for ads and other media formats.
    
*   **Cloudflare Workers + Vercel AI SDK Starter**
    
    Kristian Freeman shared a template showing Gemini web search via the Vercel AI SDK in a Cloudflare Worker. [GitHub Repo](https://github.com/kristianfreeman/workers-vercel-ai-starter)(github.com)
    
    Adding web search to AI tooling significantly improves answers for many questions, especially in the fast moving dev space. I use deep research queries constantly in Gemini and OpenAI, so building easily launchable custom tools with web search enabled is something I'll definitely explore more soon.
    

🔌 Advancements in Agents & Protocols (MCP/A2A)
-----------------------------------------------

_The push towards more capable AI agents continues, with significant developments in the protocols and tools designed to help them interact with the world and each other._

*   **MCP Demo Day (Tomorrow!)**
    
    Cloudflare hosts demos from Atlassian, Intercom, Linear, PayPal, Sentry, Stripe, Webflow & more this Thursday (May 1st). [Details](https://demo-day.mcp.cloudflare.com/)(demo-day.mcp.cloudflare.com)
    
*   **Cloudflare Agents SDK Update**
    
    v0.0.66 includes a default MCP client, letting agents (and even MCP servers) call other MCP servers. [Details](https://x.com/threepointone/status/1915735083278880950)(x.com)
    
*   **\`mcp-memory\` Tool Released**
    
    An open source MCP server memory layer built on Cloudflare tech to share context between AI clients. [GitHub Repo](https://github.com/Puliczek/mcp-memory)(github.com)
    
*   **Zapier MCP Session (May 7th)**
    
    Learn best practices for leveraging Zapier's 7,000+ integrations via its MCP server in this free session. [Register](https://bit.ly/3XRrqfp)(bit.ly)
    
*   **Google: A2A + MCP Needed**
    
    Google reiterates that complex agentic apps likely need both A2A (agent to agent chat) and MCP (tool/data access). [Blog Post](https://goo.gle/4jSta0l)(goo.gle)
    
*   **Goose Agent Primer**
    
    A quick explainer on Block's Goose AI agent and how it uses MCP for task automation. [Project Site](https://block.github.io/goose)(block.github.io)
    
*   **Motia AI Agent Framework**
    
    Check out this open source framework with examples for deep research using OpenAI & Firecrawl. [Tweet](https://x.com/donvito/status/1915451174670749869)(x.com)
    

🚀 Level Up Your AI Dev Skills: Join My Cursor Workshop!
--------------------------------------------------------

Ready to master practical AI development workflows in Cursor? Join me, John Lindquist, for this hands on workshop! I've been teaching these sessions for months, refining the content, and I'm excited to share my latest insights on Agents, Ask mode, Custom Modes, multi file analysis, effective prompting, Cursor rules, and handling AI failures. Let's conquer the complexity together!

**When:** Thursday, May 8, 2025, 9:00 AM - 2:00 PM (PDT)

**Where:** Zoom (Live Q&A included)

**Investment:** $249

Act Fast! Get the Early Bird price of $200 if you register by midnight tonight (Wednesday, April 30th)!

[Read More](https://egghead.io/workshop/cursor)(egghead.io) | [Register Now ($200 Early Bird)](https://buy.stripe.com/3csdSh2GTdhucTedQW?prefilled_promo_code=SLV2PKOR)(buy.stripe.com)

_(Team training also available)_

⚡ Quick Links / Community Buzz
------------------------------

*   **Cursor Tips for Large Codebases:** Enable project structure, @ mention files, break down tasks, start new chats often, use Ask mode for planning. [Thread](https://x.com/ericzakariasson/status/1915072216523497628)(x.com)
    
    These tips come directly from the Cursor team themselves. Definitely worth reading through the thread if you work with large projects.
    
*   **Building Internal Tools w/ Cursor:** How Eric Zakariasson uses Cursor + Rules + MCP to build tools at Cursor. [Thread](https://x.com/ericzakariasson/status/1914696699274252405)(x.com)
    
    Another great thread from a Cursor employee detailing how they set up their internal configurations. Check this one out too!
    
*   **Gemma 3 OCR Demo:** A local OCR app powered by Google's Gemma 3. [GitHub Repo](https://github.com/patchy631/ai-engineering-hub/tree/main/gemma3-ocr)(github.com)
    
    I love seeing local models being used! Google is building AI directly into Chrome, and it's fascinating to watch how far they can push this. I often use OCR on screenshots for context in queries; the future of productivity might involve AIs constantly reading our screens. Faster OCR is better for all of us!
    
*   **Color Matcher Model:** Fix white balance or match colors for image model outputs. [Replicate](https://replicate.com/fofr/color-matcher)(replicate.com)
*   **o3 "Write in My Style" Prompt:** A two prompt technique for better voice cloning. [ShumerPrompt](https://shumerprompt.com/prompts/o3-write-in-my-style-prompt-017ae28e-c135-4f48-8bc5-8b78785a51d7)(shumerprompt.com)
*   **Open Source Prompt Library:** Conversational prompt templates for "vibe coding." [GitHub Repo](https://github.com/TechNomadCode/Open-Source-Prompt-Library/)(github.com)
*   **Kaggle 5 Day GenAI Course:** A quick intro to GenAI concepts. [Kaggle Learn](https://www.kaggle.com/learn-guide/5-day-genai)(kaggle.com)
*   **Underrated AI Coding Tools?:** Peter Yang asks the community for hidden gems. [Join the discussion](https://x.com/petergyang/status/1915934134507344095)(x.com)
*   **Value at the Application Layer:** Logan Kilpatrick on why now is a great time for builders (falling costs, rising intelligence & willingness to pay). [Clip](https://x.com/tbpn/status/1915110608149889385)(x.com)

That’s the scoop for this issue! The pace continues with new models, platform refinements, and a growing ecosystem of tools and educational resources aimed squarely at developers building with AI.  
  
If you have any feedback or questions, hit reply! Always happy to chat about the latest in AI dev tools.  
  
John Lindquist  
[egghead.io](https://egghead.io/)(egghead.io)

AI Dev Essentials #4: GPT-4o Rollback, Qwen3 Models, New AI Workflow Lesson & MCP Updates

Playwright-powered visual diffs, smarter agents, and new model controls define this week. Plus: OpenAI rollbacks, Gemini upgrades, and major MCP ecosystem moves

SolidStart is a JavaScript Framework designed to build full-stack SolidJS apps and enhance your developer experience.

In this course, you will learn how to:
- Set up a SolidStart project
- Toggle different rendering modes
- Leverage file system routing to turn your files into routes
- Create API routes to expose some public endpoints
- Fetch, prefetch, update and cache your data
- Turn your functions and actions into server-executing ones
- Get rid of unnecessary network waterfalls by using Single Flight Mutations.

After this course, you should be able to leverage SolidStart to build your Solid applications.

JavaScript® (often shortened to JS) is a lightweight, interpreted, object-oriented language with first-class functions, most known as the scripting language for Web pages, but used in many non-browser environments as well such as node.js or Apache CouchDB. It is a prototype-based, multi-paradigm scripting language that is dynamic, and supports object-oriented, imperative, and functional programming styles.

javascript

SolidStart is a JavaScript Framework designed to build full-stack SolidJS apps and enhance your developer experience.

In this course, you will learn how to:
- Set up a SolidStart project
- Toggle different rendering modes
- Leverage file system routing to turn your files into routes
- Create API routes to expose some public endpoints
- Fetch, prefetch, update and cache your data
- Turn your functions and actions into server-executing ones
- Get rid of unnecessary network waterfalls by using Single Flight Mutations.

Getting Started with SolidStart

OpenAI's technologies offer a range of AI-powered tools and capabilities, making it easier for developers to build intelligent, responsive, and user-friendly applications.

openai

We can teach an AI model to better understand the information we want to generate responses from. To do that, we can use training data in a particular format, JSONL.

Once we have our data generated and uploaded to OpenAI, we can now use it to fine tune or customize a model to provide better results.

We'll use the training data we generated and uploaded in previous lessons to create a new fine-tuned model based off of the OpenAI davinci model.


- [https://platform.openai.com/docs/api-reference/fine-tunes/create](https://platform.openai.com/docs/api-reference/fine-tunes/create)

Fine Tuning OpenAI Models with Training Data Files

Once we have our training data, we need to a way to deliver it to OpenAI.

We can do that by uploading our training data file to OpenAI directly where we can then reference that with our fine tuning.


- [https://platform.openai.com/docs/api-reference/files/upload](https://platform.openai.com/docs/api-reference/files/upload)


Uploading Training Data Files to OpenAI for Fine Tuning AI Models

Using the models that OpenAI provides will get you really far, but their knowledge base can be somewhat limited for particular information or if its past a certain date that the model doesn't support.

To improve the experience, we can train models to have a better understanding of the data we want to work with, whether that's documentation for a new developer tool or guiding the model to respond closer to how how we expect.

The problem—is you need a lot of data in a specific format to do this and its not very practical to write it all our by hand. So we can instead use GPT, where by providing our data, we can ask it to generate JSONL training data for us and format it in a way that we can use to fine tune our model.


- [https://platform.openai.com/docs/guides/fine-tuning/prepare-training-data](https://platform.openai.com/docs/guides/fine-tuning/prepare-training-data)

Generating & Preparing JSONL Training Data with GPT for Fine Tuning an OpenAI Model

Node.js is a platform built on Chrome's JavaScript runtime for easily building fast, scalable network applications. Node.js uses an event-driven, non-blocking I/O model that makes it lightweight and efficient, perfect for data-intensive real-time applications that run across distributed devices.

node

With a reliable data API, we can start using our generative output in creative ways, including feeding that directly into a new generation like creating an image based off of a blog post.

We'll look at taking a GPT-generated blog post and feeding it into DALL-E along with some stylistic prompt engineering to generate an image for our post.

Use Chat Completions for Image Generation Prompts with DALL-E

An important part in creating a data API whether based on generative systems or not is being able to reliably generate responses in the format you need. Variable data structures and irrelevant information could break your application or frustrate the users of your API, resulting in downtime or lost customers.

Using the messaging API, we can train GPT to become more aware of our needs by using System roles to instruct GPT on behavior and User to Assistant examples to give GPT precedent to base its next response on.

[https://platform.openai.com/docs/guides/chat/introduction](https://platform.openai.com/docs/guides/chat/introduction)

Training GPT Using Roles & Message Examples for Better Chat Completions

Getting a long string as a response makes a lot of sense if you're building an app or tool that appears conversational or if you just want to display a little bit of information as a response, but if you want to do something programmatic with the information, such as using different datapoints from the information in different parts of the UI, you'll have a hard time parsing out that information from a single request.

To get around this, we can simply ask GPT to return our data formatted as JSON! Further, we can ask it in a particular shape. You'll learn how to format a chat completion prompt to request JSON data and gain confidence in that data by defining the shape in the request.

Requesting JSON Chat Completions from OpenAI for Programmatic Use


Requests that we typically make in an application come all at the same time. We make the request, we get a response. But using a streaming mechanism, we can get our information as it comes resulting in a quicker initial response, as well as set ourselves up for different user experiences including conversational-like messaging.

Using Edge Functions in Next.js, we can stream our response back to the application and create an experience similar to the official ChatGPT interface. You'll learn how to set up a new Edge function, make a request directly to the OpenAI API, and listen for the stream events in Next.js with [eventsource-parser](https://www.npmjs.com/package/eventsource-parser).

- [https://vercel.com/docs/concepts/functions/edge-functions/edge-functions-api](https://vercel.com/docs/concepts/functions/edge-functions/edge-functions-api)

Stream OpenAI Chat Completions to a Next.js App Using Vercel Edge Functions

Get Started with AI-Driven App Development Using the OpenAI Node.js SDK

ChatGPT and similar conversational AI tools have quickly made impressions in the tech industry and beyond by giving the people using those tools a more natural and a potentially more helpful means to get answers to the questions they have. Bringing GPT into applications can unlock new ways to solve problems and quickly help customers to solve theirs.

We can create ChatGPT-like conversational messages with the OpenAI SDK in a Node.js project. You'll learn how to use the [createChatCompletion](https://platform.openai.com/docs/api-reference/chat/create) method and configure the gpt-3.5-turbo model to generate a response based on a message prompt.

[https://platform.openai.com/docs/api-reference/chat/create](https://platform.openai.com/docs/api-reference/chat/create)


Generate Chat Completions with AI in Node.js Using the OpenAI SDK & GPT

The more descriptive an image prompt, the better the results will be. That includes being able to take user input to help articulate what the image should look like.

Dynamic prompts in Next.js serverless functions can help us generate images based off of application input with the OpenAI SDK. You'll learn how to configure an API Endpoint to parse a body of a POST request to grab a dynamic prompt value and use it to configure the prompt for the createImage method.

[https://nextjs.org/docs/api-routes/introduction](https://nextjs.org/docs/api-routes/introduction)

Create Dynamic Image Generation Prompts for the OpenAI SDK in Next.js API Endpoints

Image generation provides us a way to bring life to natural language through a visual medium. Through a simple or detailed description, we can generate a completely unique image that tells the story we want whether that's for dynamically creating blog images or exploring new forms of art.

We can use AI to generate images with DALL-E and the OpenAI SDK in Node.js. In this lesson you'll learn how to use the [createImage](https://platform.openai.com/docs/api-reference/images/create) method generate an image using a descriptive prompt, the number of images you want to generate, and the size of the image. 

[https://platform.openai.com/docs/api-reference/images/create](https://platform.openai.com/docs/api-reference/images/create)


Generate Images with AI in Node.js Using the OpenAI SDK & DALL-E

OpenAI is an AI research lab that has developed a number of tools for researchers and developers including natural language processing, robotics, and reinforcement learning. By leveraging the OpenAI APIs, we can build faster and smarter, while bringing new and interesting capabilities to help enhance existing solutions and solve new challenges.

Let's get started building with AI using the [OpenAI SDK](https://platform.openai.com/) by installing and configuring the SDK in a Node.js project. Here you'll learn how to install the [openai npm package](https://www.npmjs.com/package/openai) into an existing project, import the openai dependency, and configure the SDK with your OpenAI API Key to make authenticated requests to the OpenAI API.

[https://platform.openai.com/overview](https://platform.openai.com/overview)
https://platform.openai.com/overview

Install and Configure the OpenAI SDK in a Node.js Project

A library designed specifically for a functional programming style, one that makes it easy to create functional pipelines, one that never mutates user data.

ramda

In this lesson we'll see how to use the `without` function to remove items in a known list from an array. Then we'll generate our exclusion list and apply it to `without` using `reduce`, `min`, `max` and `converge` to detect the highest and lowest values in an array and creating a new array that excludes those values.

Discard the High and Low Values From an Array of Numbers with Ramda

Learn how you can use ramda.js to bring functional programming concepts into your JavaScript code. Ramda offers composability and immutability right out of the box, so you can leave your imperative code behind and build cleaner, more maintainable code.

Functional Programming in JavaScript with Ramda.js

Most.js is a toolkit for reactive programming. It helps you compose asynchronous operations on streams of values and events, e.g. WebSocket messages, DOM events, etc, and on time-varying values, e.g. the "current value" of an <input>, without many of the hazards of side effects and mutable shared state.

most

Unlike RxJS, Most does not include an observable-based AJAX helper utility. However, it can be very convenient to have one, because it allows you to use streams to do things like declaratively handling errors & setting up cancellation. In this lesson, we use currying & functional composition to create an Observable-based wrapper around the fetch API with Ramda & Most.

Create an Observable-Based Fetch Wrapper Using Ramda & Most

React is one of the web’s most popular frameworks for building JavaScript applications. 

If you know what you’re doing, React can drastically simplify how you build, use, and maintain code.

Whether you’re a React newbie or you’re ready for advanced techniques, you can level-up with egghead. Explore our in-depth courses, lessons, and community resources to build more powerful applications and crack open your career.

react

In this lesson we'll take a stateful React component and look at how we can refactor our `setState` calls to use an updater function and then leverage Ramda's `evolve` function to make our updater function a reusable utility that isn't tied to the React API.

Update State in React with Ramda's Evolve

Hardcoding your script to search for only one type of file (like `.md`) is a major limitation. This lesson shows you how to make your script smarter by teaching it to infer the desired file type directly from the user's natural language prompt.

You'll implement a two-stage AI process. The first stage analyzes the user's request to determine the file extension they're interested in (e.g., `md`, `txt`). The second stage then uses that file type to dynamically search the file system, find the relevant files, and execute the command.

**Workflow demonstrated in this lesson:**

* Add a preliminary `generateObject` call to extract a `fileType` from the user's initial prompt.
* Define a list of supported `fileTypes` (e.g., `['md', 'txt']`) and use a detailed prompt to guide the AI, instructing it to choose from the list or use a default.
* Dynamically construct the `globby` pattern using the inferred `fileType` (e.g., `*.$[fileType]`).
* Implement error handling to inform the user if they request an unsupported file type or if no files of the specified type are found.
* Demonstrate how the script defaults to searching for Markdown files but can correctly search for other types when specified (e.g., "summarize my profile from my pdf").

This technique adds a significant layer of intelligence to your script, allowing it to adapt its behavior based on the user's intent and making it a far more flexible and intuitive command-line tool.

Dynamically Inferring File Types from User Prompts in Local AI Scripts

Hardcoding command logic and prompts directly into your script makes it rigid and difficult to extend. This lesson demonstrates a more flexible and powerful pattern: defining your script's commands as individual Markdown files in a dedicated directory.

This approach allows you to easily add, remove, or modify commands without touching the core script logic. You'll learn how to dynamically load these command files, present them as options to the AI, and inject the corresponding instructions into your prompt, creating a truly modular and extensible tool.

**Workflow demonstrated in this lesson:**

* Create a `commands` directory to house your command definitions (e.g., `summarize.md`, `translate.md`).
* In each file, write the specific prompt or instructions for that command.
* Use `globby` to discover all available command files at runtime.
* Create a `commandMap` that maps the filename (e.g., "summarize") to its file content.
* Update the Zod schema to use an `enum` of the dynamically discovered command names.
* Dynamically construct the final text generation prompt by injecting the instructions from the user-selected command file.
* Add a `profile.md` file to demonstrate how the script can act on user-created content with the new file-based commands.

By the end of this lesson, your script will be architected for easy expansion, allowing you to build a robust library of custom commands simply by adding new text files.

Creating Dynamic, File-Based Commands for Local AI Scripts

AI models won't always give you the answer you expect; sometimes they'll return invalid data or "hallucinate" a response that doesn't fit your schema. This lesson teaches you how to make your local AI script more robust by implementing essential error handling for user commands.

You'll learn how to validate the AI's output against a predefined list of valid commands and gracefully catch errors when the model generates an unexpected response, ensuring your script fails predictably and provides helpful feedback to the user.

**Workflow demonstrated in this lesson:**

* Define a list of `validCommands` (e.g., `summarize`, `translate`, `review`) to constrain the AI's choices.
* Wrap the `generateObject` call in a `try...catch` block to handle cases where the AI's output doesn't match the Zod schema.
* Implement a type guard function (`isValidCommand`) to check if the command returned by the AI is in your list of valid commands.
* Provide clear error messages to the user, indicating what went wrong and what the valid commands are.
* Test the script with invalid commands (like "taco" or "delete") to confirm that the error handling works as expected.

By adding these checks, you make your script significantly more reliable and user-friendly, preventing it from crashing on bad input and guiding the user toward correct usage.

Handling Invalid Commands and AI Errors in Local Scripts

Once your script can understand a user's command and identify a target file, the next step is to perform an action. This lesson shows you how to use the Vercel AI SDK's `generateText` function to process a file's content and produce a relevant, AI-generated response.

You'll learn how to read the content from the file path you've already identified, combine it with the inferred command into a new, more detailed prompt, and then pass it back to your local AI model to generate a final text output, such as a summary or a review.

**Workflow demonstrated in this lesson:**

* Import the `generateText` function from the `ai` SDK.
* Read the file content from the `filePath` returned by `generateObject`.
* Construct a new, structured prompt that includes the original command (e.g., "summarize") and the full content of the file.
* Pass this new prompt to the `generateText` function.
* Log the resulting AI-generated text to the console.
* Run the script with commands like "Summarize our TypeScript config" or "Review my package file" to see the full, context-aware output.

This lesson completes the core cycle of your AI script: understanding a command, reading the relevant context, and generating a useful, text-based result, all on your local machine.

Generating Text with a Local AI Model Based on File Content

A command is useless without knowing what to act upon. This lesson elevates your local AI script by giving it context from your project's file system. You'll learn how to find all relevant files and empower the AI to intelligently select the correct one based on your prompt.

The core technique involves using the `globby` package to scan for files and then feeding that list into a Zod `enum`. This constrains the AI's choices to a valid set of files, allowing it to accurately determine which file you're referring to in a natural language command.

**Workflow demonstrated in this lesson:**

* Install and import the `globby` library to find files based on a pattern (e.g., all `.json` files).
* Add a `filePath` field to your Zod schema.
* Define the `filePath` as `z.enum(files)`, where `files` is the array of file paths returned by `globby`.
* Pass a natural language prompt like "Summarize our typescript config" to the script.
* Observe how the AI correctly identifies `tsconfig.json` from the list of available JSON files and returns it as structured data.

This lesson is a crucial step toward building powerful, context-aware tools that can intelligently interact with your project's files based on simple, human-readable commands.

Providing File System Context to Local AI Scripts with Globby and Zod Enums

Hardcoded prompts are great for testing, but a truly useful script needs to respond to dynamic user input. This lesson shows you how to make your local AI script interactive by capturing command-line arguments.

You'll learn how to use Node.js's built-in `parseArgs` utility to take a user's natural language command directly from the terminal and pass it to your local language model for processing.

**Workflow demonstrated in this lesson:**

* Import the `parseArgs` function from the `node:util` module.
* Configure `parseArgs` to accept positional arguments from the command line.
* Replace the static, hardcoded prompt with the dynamic input provided by the user.
* Run the script with different sentences (e.g., `"Download a file"` or `"I want to download a file"`) to demonstrate that the AI can correctly infer the core command regardless of the phrasing.

By the end of this short lesson, your script will be able to accept and process commands directly from the user, making it a flexible and reusable command-line tool.

Processing Dynamic User Input in Local AI Scripts

While command-line chats with local AI models are useful, the real power comes from scripting. This lesson walks you through the initial setup for creating powerful, local AI-driven scripts by combining Ollama, the Vercel AI SDK, and Zod.

You'll learn how to take a natural language command, like "Concatenate all of the markdown files in my Downloads folder into a single file," and use the `generateObject` function to intelligently extract structured, actionable data from it.

**Workflow demonstrated in this lesson:**

* Install and set up the `ollama-ai-provider`, `ai`, and `zod` packages.
* Create a local AI model instance using Ollama and the Gemma 3 model.
* Define a Zod schema to specify the exact data structure you want to extract (e.g., `{ command: string }`).
* Use `generateObject` to pass a natural language prompt to the local model and receive a structured JSON object that matches your schema.
* Run the script using `bun` to see the local model successfully infer the "concatenate" command from a full sentence.

This foundational lesson unlocks the ability to build sophisticated, automated workflows by translating human language into machine-readable commands, all running entirely on your local machine.

Scripting Local Language Models with Ollama and the Vercel AI SDK

### **Build Powerful, Local AI-Driven Command-Line Scripts**

Go beyond simple AI chats and unlock the true potential of local language models by learning to script them. This course will guide you step-by-step through the process of building a flexible and intelligent command-line tool that can understand natural language, interact with your file system, and automate complex tasks—all running entirely on your local machine.

Starting with a blank TypeScript file, you'll learn how to combine the power of **Ollama**, the **Vercel AI SDK**, and **Zod** to create a script that's not just functional, but smart. We'll begin by teaching the AI to parse a simple command from a sentence. From there, we'll progressively add layers of sophistication: processing dynamic user input, making the script aware of your project's files, and generating meaningful text-based output.

You'll also learn critical patterns for building robust and extensible tools. We'll cover how to handle invalid commands and unexpected AI responses gracefully. Then, you'll refactor the script to use a modular, file-based system for its commands, allowing you to easily add new functionality without touching the core logic. Finally, we'll add a layer of intelligence that allows your script to infer exactly what you mean, right down to the type of file you want it to process.

By the end of this course, you will have built a powerful, personalized AI assistant and gained the foundational skills to create sophisticated local automations for any task you can imagine.

**What You'll Learn:**

* How to script local language models using Ollama and the Vercel AI SDK.
* To parse and process dynamic user input from the command line.
* To use Zod schemas, including `enums`, to validate and structure AI model output.
* Techniques for providing file-system context to your scripts with `globby`.
* How to generate context-aware text based on file content.
* Robust error handling for invalid user commands and unexpected AI responses.
* To build a modular and extensible tool by defining commands in separate files.
* To intelligently infer user intent, such as dynamically determining file types from a prompt.

Build a local AI-powered command-line assistant from scratch. Learn to use Ollama, Vercel AI SDK & Zod to parse input, automate tasks, and run scripts offline.

This lesson demonstrates how to give Cursor direct access to your terminal output using the [Nx MCP extension](https://nx.dev/getting-started/ai-integration?utm_source=youtube&utm_medium=videos&utm_campaign=Nx-AI&utm_id=Nx-AI), enabling automatic error detection and fixing in monorepo projects.

**Workflow demonstrated in this lesson:**

- Encounter a runtime error when serving a Vite application in an Nx monorepo
- Install and configure the Nx Console extension with MCP integration
- Ask Cursor to check your terminal for errors and fix the underlying issues
- Watch as the AI accesses running task details and terminal output through Nx MCP tools
- Observe how it analyzes workspace relationships to identify incorrect imports and missing dependencies

**Key benefits:**

- Seamless Error Handling: AI gains direct access to terminal output and running task information
- Automatic Diagnosis: Accurately identify the root cause of errors without manual copying
- Context-Aware Fixes: Leverage workspace knowledge to implement proper solutions
- Reduced Context Switching: Eliminate the need to manually relay information between terminal and AI



Juri Strumpflohner

Enable Cursor to directly access the terminal to read the output and fix issues

See how Nx MCP gives AI coding assistants direct access to terminal output and running tasks, enabling automatic error detection and fixing.

This lesson demonstrates how to solve the predictability problem when using AI to generate code in monorepos by combining Cursor with Nx generators.

**Workflow demonstrated in this lesson:**

- Explore how Nx generators provide predictable scaffolding for new libraries
- Set up [Nx Console with MCP](https://nx.dev/features/enhance-AI?utm_source=egghead&utm_medium=videos&utm_campaign=Nx-AI&utm_id=Nx-AI) integration to expose generator capabilities to Cursor
- Use the AI assistant to brainstorm a new feature with the help of the Nx MCP tools
- Have Cursor leverage the Nx MCP tools to trigger the appropriate Nx generator for creating a React-based feature library
- Observe how the AI assistant takes the generated library and further integrates it into the codebase

**Key benefits:**

- **Predictable Code Structure**: Ensure generated libraries follow your project's and organization's conventions
- **Consistent Setup**: Maintain uniform configuration across TypeScript projects and package.json files
- **Guided Implementation**: Combine AI instructions with structured, predictable code generation

Enable Cursor to generate predictable project setups

Learn how to combine Cursor AI with Nx generators for predictable library creation in monorepos, ensuring consistent structure across your codebase.

Search

Search Results