Build AI workflows with simple flowcharts

Hello everybody, welcome back to Moodboard 🌴

I'm always on the lookout for more ways that we can interface with AI beyond tedious, manual typing of long prompts into your standard LLM chat interface. 

For that reason, I was very excited to stumble upon tldraw.computer - a visual AI workflow builder that lets you create complex AI-powered automations using simple flowcharts.

(Note: I'm not getting paid or sponsored by tldraw - I’ve just been playing with it quite a bit and thought you all would find it useful 😁 )

In today’s guide, I'm going to walk you through how to get started, some of my favorite features, and a marketing automation I built within the platform that you can fork and customize.

Let's dive in 👇

What is tldraw.computer?

tldraw.computer is a (free!) tool that gives you the power to build multi-step AI workflows using simple, intuitive flowcharts.

Instead of writing long, complex prompts or chaining multiple AI tools manually, you can create a visual graph where each individual node/step represents an AI operation. 

Data flows from one node to the next (using easy-to-draw/understand arrows), with each step processing and transforming the input to your instructions.

So, instead of telling ChatGPT "do A, then B, then C" in one massive prompt, you create three connected boxes where box A feeds into box B, which feeds into box C. 

This is powerful for a few reasons:

• You can see your entire AI process/workflow laid out visually. Complex automations become easy to understand and modify when they're displayed as flowcharts instead of hidden in long text prompts.
  

• You can combine multiple types of input/output in one workflow. Text, images, audio, and even live camera feeds can all be processed together in the same visual system.
  

• Collaboration across teams and organizations becomes intuitive. Visual workflows can be shared with team members who can immediately understand, fork, and modify the process without deciphering complex prompt chains.

A First Basic Workflow

Okay, with a basic understanding of what tldraw.computer is, let’s build our first simple workflow!

Every new project starts on a blank canvas (if you’ve ever worked with Figma or similar tools, you’ll feel right at home within the tldraw interface).

To begin building, we simply drag, drop, and connect some building blocks.

Let’s start with a very simple text transformation. To make it a little fun, we’ll build a workflow that writes haikus. It should only take three steps:

Step 1: Add your nodes from the sidebar. For a simple example, you'd drag in:

• A text node (this serves as your input - for this exercise, a subject for our haiku like "the ocean")

• An instruction node (this is your prompt to the AI - like "write a haiku about this")

• A text output node (where your result will appear)

Step 2: Connect with arrows. Connecting nodes is as simple as dragging arrows between them. Draw an arrow from your text input to the instruction node, then from the instruction node to the output.

Step 3: Hit play and watch it work. Click the play button on the instruction node and it processes all the inputs flowing into it, then sends the result to the connected output nodes.

Now, where it gets interesting is when you start introducing multi-modal inputs & outputs (different mediums like image, audio, etc).

What happens if we add a “speech” output after the instruction?

It will create a text-to-speech reading of the haiku alongside the text output.

What if we add an image input that feeds into our instruction? Let’s try with an unrelated image of a red sports car.

It will now treat that image as an equally-weighted input, and incorporate its interpretation of that image into the haiku. 

As you add branches and functions to our workflow, you can very quickly (and very intuitively) create complex, useful multi-step workflows.

Some Other Cool Features

Once you grasp the basics, tldraw.computer becomes very powerful. It rewards creativity, and it gives you a pretty comprehensive set of tools to use when crafting your workflows.

There are extensive tutorials/instructional videos on the tldraw.computer homepage (highly recommend you check those out), but I wanted to highlight a few of the more useful/interesting features that I’ve played around with.

“Labeling” the Inputs

If you double click on the center of an arrow that connects two nodes, you can assign it a text label. This isn’t just for looks (though it does help keep things organized). It acts as an actual parameter name for the input that it passes.

So, if you label an input’s arrow as “word count”, the instruction step will know that you intend for whatever number you input to be the desired word count.

Buttons, Switches, and Toggles

You can add buttons and toggles to your workflow to start and stop the processing at specified points.

For example, if I want to process only the first part of a workflow without it continuing/completing the rest, I can add a toggle in between steps and turn it off.

If I want to make it obvious how to start the workflow, I can add a button that (using an arrow) triggers the starting step to begin.

Instant Website

One of the outputs that you can drag onto your canvas is a website builder. It uses the inputs to create a simple website within tldraw.computer.

I don’t know how immediately useful this feature is, but I thought it was a cool demonstration of the power of the platform when used creatively. 

It’s also always fun to see what kind of website a bunch of random inputs will yield. Here’s the website that it made from the haiku from earlier 😂

The Tool I Built

These are all cool features in isolation, but, to demonstrate a business use-case, I want to show you the lil project I built in tldraw.

A lot of the newsletters we’ve built & partnered with have a ‘curated links’ section - a portion of the newsletter where they provide a bullet-pointed round-up of the latest updates in their niche along with a bit of commentary, usually ~75 words or so. 

If you don’t immediately know what I’m talking about, here’s an example of this kind of section from Morning Brew:

Creating the ‘takes’ for this kind of section is a process that would benefit from some degree of repeatability/systemization, and I thought this would be a cool workflow to build in tldraw.computer.

So, I created "Tailwind Takes.” 

It’s a 5-step workflow that:

1. Takes a sample of your writing and analyzes the style/tone/voice (so that the output sounds like you).

2. Allows you to input a piece of news/article that you want to comment on and any unique perspective you have on that piece (feel free to ramble unorganized thoughts).

3. Creates a concise ~75 word "take" in your voice for use within your newsletter or social media.

Here’s a quick Loom walking through the tool, how to use it, and how I built it.

If you want to experiment with the project (and ‘fork’/modify it), here’s the link to the workflow itself. Simply click the ‘Fork’ button in the top right to make your own version.

Why I Think This Is Cool

What I love most about the tldraw.computer interface is how simple it is to iterate and add features/nodes one-at-a-time and end up with a really complex and useful automation.

The visual nature of the whole thing makes it accessible to non-technical folks and experienced builders alike. The fact that it's modular means that you can very easily tweak and tailor specific parts of the automation rather than rewriting or editing a multi-hundred word prompt. 

It'll get you thinking like an engineer, even if you're just building simple content marketing automations.

I only started experimenting with it this week, and I know I'm only scratching the surface. That being said, I'm really interested to see what you all create.

As always, please feel free to send me any projects that you end up building after reading. I'll feature anything cool in next week's edition!

Until next week,

Ryan