Lesson 3, Part 1: Picking the Right Task

Where we are

You have Claude Code and VS Code installed. You understand the four building blocks that make up every AI workflow. Now it's time to actually build something. But before we start, we need to answer the most important question: what should you actually build?

Environment Claude Code inside VS Code

Lesson 2

The Task Identifying what to build

This part

Context Library Your files and knowledge base

Part 2

Tools & Instructions context.md, persona, process, best practices

Part 3

Execution First draft, feedback, iteration

Part 4

Proving the Workflow Reusability and the four building blocks

Part 5

Which tasks are worth automating?

Not every task benefits from building a workflow around it. Some tasks are better done manually, and that's fine. As a rough rule of thumb, good candidates tend to share a few common traits.

Repetitive Similar structure, similar format, most of the time

Research-heavy Searching through databases, content libraries, or multiple sources to compile an output

Instruction-driven Follows a clear process with a definable output

Tool-dependent Needs and uses different tools for input and output

The more of these traits a task has, the stronger the case for building a workflow around it. A task with all four is a great candidate. A task with only one might not be worth the setup time.

A few examples

Strong candidates Build it

Weekly client reports, branded communication materials, recurring invoicing, custom outreach.

Probably not worth it Skip it

A one-off email to a colleague, a quick brainstorm, a single ad-hoc question you need answered once.

This is a rough guide, not a rigid framework. You can automate almost any task. But tasks with more of these traits give you the most value for your setup time.

Introducing the task for this demo

A lot of what professionals do involves creating communication collateral, whether it's a document, a presentation, a pitch deck, or a social media post. The format changes, but the core task is the same: take scattered inputs and turn them into something polished and on-brand.

For this demo, the task comes from a real project. I work with a nonprofit called the Assistive Technologies for Disability Trust that is building the entrepreneurial ecosystem around disability inclusion in Kenya and across the continent. We've raised a small pilot fund to invest in assistive technology startups, and to grow that fund, we need to build partnerships with organisations across the region and internationally. That means strong communication materials.

The problem is familiar: an assortment of inputs, from correspondence to brand assets, scattered across different places, that need to become one coherent, brand-consistent PDF.

Input Scattered

Alignment docs, correspondence, brand assets, photos, across multiple files.

Output Polished

One brand-consistent PDF, ready to send to potential partners.

Key Insight

Once you have one well-structured source, you can transform it into other formats too. A web page, a presentation, a LinkedIn carousel. One input, multiple outputs.

What does "done" look like for our task?

Before building anything, it helps to define what the finished output should look like. For the AT4D communication pack, "done" means:

Easy to skim and read through

Clear message on each page

Brand-consistent colours, typography, and layout

Includes brand assets and photos

Simple, straightforward. Having this list before you start means you can check each version against it, give targeted feedback, and know when you're actually done.

What's next

Now we know what to build and what "done" looks like. The next step is planning. In Part 2, you'll build the context library: the folder system that gives Claude everything it needs to produce quality output. This is where most of the magic happens, and where you should spend the majority of your time.