The 3 Ways AI is Changing How We Code

You likely know the term “vibe coding,” especially if you read this newsletter. It can mean different things depending on how you approach coding with AI-powered tools.

LLMs are getting pretty good at turning prompts into code. Now we’re seeing platforms innovate at the software level, building the “harness” that guides LLMs as they generate code.

This week, we’ll look at the main use cases for AI coding tools and share links for deeper reading.

Code it for me 
Replit, Lovable, Vercel v0, Claude Artifacts, Perplexity Labs
These tools take a text-based prompt and return a functional application. You’ll also see them used to ‘clone’ a website in minutes. Of course, that often ignores the backend and infrastructure. Where these shine is in creating single-purpose, customized apps, like a tailored allowance tracker for your kids and their chores.

Try it: go to v0.dev and use the prompt below—or write your own.

👀 Check out my results.

Code it with me 
GitHub Copilot, Claude Code, OpenAI Codex, Cursor
These tools can help developers where they edit code. Whether that's in their IDE or terminal. These will provide the most benefit to developers and can be used to help explain codebases, offer code suggestions, or pair program inside the code editor.

🤿 Dive Deeper - Check out GitHub's Spec Kit tool for developing detailed specs before jumping into vibe coding with AI assistants. I like the concept, particularly when using for adding a feature to an existing application, but I did find it a bit over-engineered when I used it to plan out a brand new application. I had to scale back to stick to an MVP plan.

Code it in the background 
OpenAI Codex Web, Claude Code, GitHub Coding Agent
These are the set of tools that aim to let developers run automated tasks in the background. Is it a good idea to give these the keys to your full file system? No, no it is not. But it is powerful to be able to submit an issue in GitHub and let an agent take the first pass at the fix. This frees up developers to focus on architecture and how the application works.

👀 OpenAI is looking to tackle automated coding with its latest Codex updates including Codex Web for running agents on its infrastructure. They also put out a GPT-5 coding cheatsheet.

🎥 Video
We Tier Ranked Every AI Coding Assistant
Syntax.fm 
If you want to see the sheer volume of AI coding assistants already out there, take a look at Syntax.fm's tier rankings. There are more than any one person could hope to try, and they give a good developer-focused view of the tools.

📰 Article
When to use AI to Generate Code
Carl's Blog
The titular Carl, from Carl's Blog shares his experience on using AI to generate code and gives some useful examples and frameworks for determining your AI coding use case. I tend to agree with his review of when AI coding works well and for handling the different levels of prompt and planning before letting AI tackle a coding task.

📰 Article
Tea encouraged its users to spill. Then the app's data got leaked
npr.org 
I don't believe the Tea data breach has been confirmed to be due to vibe coding malfeasance, but it has become a rallying cry for those against the practice. It does illustrate the dangers of relying on AI coding tools without necessarily understanding or planning for the entire security infrastructure of an application. The male-focused counter to Tea's success, TeaOnHer, was almost assuredly vibe coded and rushed out with its own plethora of security vulnerabilities.

🎥 Video
GitHub Spec Kit 💖 GitHub Copilot CLI 
Den Delimarsky
GitHub Copilot threw its hat into the CLI coding assistant ring recently with [GitHub Copilot CLI](https://github.com/github/copilot-cli. This video showcases the newest way to use Copilot with an added bonus of giving a look at GitHub's Spec Kit in action. I've like the CLI based coding assistants more than I would have expected. Copilot could be interesting to be able to switch between models. GitHub seems to be in a release cycle where they're seeing what the likes of Claude and Cursor release and then implement them into their system with their edge being direct integration with GitHub.

📰 Article
GitHub MCP Registry 
GitHub
One area GitHub does have an advantage over other companies building AI agents is that it is the established platform for hosting code. Leaning into that, GitHub has a MCP (Model Context Protocol) registry. They're looking to allow developers to standardize where developers can publish and find MCP servers for platforms like Figma, Zapier, and more.

Thanks for reading, Jason