Issue #1: Cursor is all you need

Hey there!

You will have known us before as Generative Nation, but now we are back as PALINDROM.

We’ve been super busy working hard, learning lots, building our business, and now we are back with a fresh new look.

We built generative AI systems for many companies over the past 12 months, and now we’re restarting our newsletter series to share our learnings about deploying gen AI systems in production.

We are going to share war-stories, some things that worked and, others that didn’t. We hope you’ll get a ton of value from these.

Intro

This is the first issue of the new edition and we're diving headfirst into the world of Cursor - the AI coding tool that's been buzzing around your ears.

Guess what?

It's not just noise; this bad boy has quickly climbed to the top of our 'can't-live-without' list here at PALINDROM.

So, here's the real scoop: we're not just going to sing praises about Cursor; we're going to break down how this gem is revolutionizing our daily grind.

And who knows?

By the time you're done reading this, you might just be ready to let Cursor work its magic on your workflow too. But before we get into it, let’s give you some of the latest developments we’re looking at…

Quick links

Nomic Embed - the first fully open long context text embedder to beat OpenAI.

InstantID - might come handy for your next AI avatar app.

Enterprise Scenarios Leaderboard - the first LLM leaderboard for real world enterprise use cases.

Alright, back to Cursor.

The app that rivals Copilot

We started to use Cursor internally mainly because of their chat feature and we've quickly found several key ways to leverage its capabilities.

We’ll share a few practical use cases in this issue including:

• API Documentation Integration

• Lint Error Management

• The vision model for UI Design

• …and our Task Driven Development workflow.

Before we jump into the deep end with those mind-blowing, advanced workflows, it's crucial we get the fundamentals down pat.

So, let's roll up our sleeves and start from square one.

Ready?

Let’s do this! 👊

The essentials

There are two core ways to interact with AI models in Cursor:

• through ⌘-K

• and via chat.

⌘-K

You can select any piece of code, hit crtl-k or ⌘-k and then write a prompt for an LLM to update your code following your instructions.

A suggestion from the model will be presented to you inline and you can see suggested new lines in green and removals in red.

Nice.

Chat

Another way to use the Cursor is through the chat window, which offers three key advantages:

1. having the chat window in the same app means less context switching, and much more convenience,

2. development specific prompts,

3. and fine-tuned models specifically for development.

Code base relevant chat

Chat in your editor is great, but we can make it even more powerful.

Language models have been trained on the internet, and many have further been fine tuned for specific tasks like code generation or other specific domains.

But here's the kicker – these models, as smart as they are, they're kinda clueless about your specific codebase. It's like, if your code's a secret club, they're not on the guest list, especially if it's a private repo.

Now, enter Cursor. It doesn't just get a peek at your codebase; it dives in headfirst. It's like having a super-smart buddy who knows your code inside out.

What does this mean for you?

Instead of asking those generic, "How do I do X" kind of questions, you're hitting the jackpot with "How do I do X in THIS specific codebase."

It's like going from a one-size-fits-all to a tailor-made suit.

And that is where the real magic happens.

To use this power, all you need to do is command-return or ctrl-return when searching and cursor will add your codebase to the search context.

Real-world applications

Alright, we've got the basics down.

Now, let’s pivot to the juicy stuff – real-world applications.

This is where the rubber meets the road, folks. Let's see how this theory plays out IRL! 🚀

1. API Documentation Integration

Hallucination of non existent functions, or LLMs not knowing about the API you are using because it finishing before the version of an API you are using was developed, can be very frustrating problems.

We can mitigate this by supplying the relevant documentation to your prompt.

Cursor makes it incredibly easy to integrate documentation with quick macros.

Simply type @docs to add api documentation to the context.

A menu will popup, and you can type and select from common APIs. Hit enter and now the LLM will have fresh documentation to build your response from.

What do you do though if the documentation you need isn’t included in the common APIs?

Cursor make it very easy to add extra documents: simply provide the URL and cursor will scrape the URL (it must be a public URL) and then its quickly available whenever you need it.

2. Lint Error Management

Cursor also provides another macro you can quickly use to quickly pipe linter error message into your chat prompt.

Simply type @lint errors and then any error messages from your currently open file into will be passed to your chat prompt.

No more spending half an hour trying to work out which bracket is missing.

3. GPT-4 Vision for UI Design

Cursor’s power doesn’t just end with text based prompts. It has multimodal features with the power of GPT-4V.

For example, you can pass your mockup in for instant code.

Mock up as a prompt:

Generated result:

Pretty awesome.

How about if you want to edit

And then boom: your code is updated. We even spiced this up with some cool Midjourney cupcakes.

These are obviously very basic examples, but they should show you the power at hand.

Some other experiments we ran in the past included errors, network diagrams, and other visual aid - we’ll defo share these examples in a future issue.

4. Task Driven Development

Finally, let’s see Cursor in action.

Here’s a video from Jon showing one of his favourite workflows. He uses a task list to keep context and tasks fresh with Cursor.

Final words

To summarise, here at PALINDROM, we’re using Cursor as our coding wingman. It's our secret sauce in the tech kitchen, blending AI seamlessly into our development workflow.

Remember, AI in coding isn’t about replacing us – it's about augmenting us. It's our Iron Man suit.

And with Cursor, we're not just coding; we're crafting the future, line by line.

That’s it for now folks, see you next time.