Don't Imitate Understand - #9

Hello!

In this ninth issue of the "Don't Imitate Understand" newsletter you'll get:

• Course news then a discussion about the importance of Agent Skills in your AI-assisted workflow.
• A bunch of interesting links!

Course News

• I'm excited to announce new modules on Agent Skills have dropped in my AI-assisted development course.

  I think Agent Skills are important enough that I decided to also release these modules as a standalone course on Udemy, which you can get extremely inexpensively this week only. However, if you've enrolled in the full AI course you get these lectures already for free.

  The AI course will be completed this month, and then in March I will focus on new modules for my Modern JavaScript Frameworks course which is still at an early-access price.

You Should Be Writing Agent Skills

The Agent Skills standard is an open-source standard created by Anthropic that AI agents everywhere are adopting.

Its structure is simple: a folder with a markdown file (SKILL.md) and optional scripts and other content that the agent will load in the LLM's context window if the description of the skill (found in the SKILL.md file) seems to match what the user is trying to do.

Skills let you encapsulate and distribute context and tooling in an easier format than building an MCP server. The Model Context Protocol still has its place, I believe, but Agent Skills are something you should be authoring now.

The standard describes Agent Skills as useful for:

• Domain expertise: Package specialized knowledge into reusable instructions, from legal review processes to data analysis pipelines.
• New capabilities: Give agents new capabilities (e.g. creating presentations, building MCP servers, analyzing datasets).
• Repeatable workflows: Turn multi-step tasks into consistent and auditable workflows.

and once you build a skill, you can use it across agents (Claude Code, Codex, Copilot, etc.) because they all support the same standard.

I have three favorite ways to use Agent Skills:

1. Domain expertise: When starting a new project, package up what you know about the business domain into a skill. If you're building an app for a home construction company, create a skill on home construction.

   That means vocabulary, business processes, common tasks, little-known facts, and more. Everything you learn from discovery.

   This leads to more readable code. Function and variable names match business reality, for example.

2. Development and design guidance: What activities do you want your developers and designers to do in a consistent way, but isn't an activity they are doing constantly? Skills are loaded progressively, meaning they save context window space by only being loaded when the LLM thinks they're needed.

   That means you can put things in a skill that don't need to be in always-loaded rules files like AGENTS.md or CLAUDE.md.

   A good example is Anthropic's frontend design skill.

3. Useful automation: Skills let you bundle scripts for the LLM to choose and the agent to run. Identify areas in your workflow that you want to automate, have an LLM help you write automation scripts, and then bundle them into a skill with instructions on how and when to use it.

You should distribute your Agent Skills to your entire team. These may be skills you create, or skills you discover in skill repositories which are growing fast.

I love the ease-of-distribution, testability, and usefulness of Agent Skills. I think it's a fantastic way to begin to organize your team around consistent and standardized AI-assisted development practices.

Currently too many teams have each dev doing their own approach to AI-assisted development. Your teammates should share rules files and skill folders at the very least. If you aren't doing that yet, I recommend you start doing so soon.

Nothing is certain of course, with non-deterministic LLMs. They may not always follow a skill perfectly. But in practice I find they help a lot, and I think going forward they will be one of the things you first get familiar with when starting a new job or a new project.

You should be writing Agent Skills. Go try some out from skills.sh or hunt for some on GitHub. Read how popular ones are written and see how they work. As I announced above, you can learn more in my AI-assisted dev course or the new Udemy standalone course.

However you choose to learn and practice, I think you can have a fun and productive time organizing and planning your AI-assisted experience around reusable, portable Agent Skills.

A Personal Note

Since you may see me mention this online, I wanted to say here that I'm currently looking for a new remote position to provide value to a company or organization from my home base in New York.

I've always coupled my course development work with design, development, management, and educational work at a company.

If you think your company, or a company you know of, would benefit from my experience in Developer Experience, Developer Education, mentoring, UX, AI strategy, or other areas, feel free to reply to this email and let me know!

Links

• AI fatigue is real and we should be talking about it more.
• Imagine if AI followed "don't imitate, understand" and grabbed the source code for any library it used. There's an experiment to try doing just that!
• Here's a great site that showcases how to replace old hacks with modern CSS alternatives, including removing JavaScript in a lot of cases.
• A good blog post on how to write quality code with AI. Something we'll also discuss in the next modules dropping in my AI course, and I agree with the contents of the post.

That's it for this ninth issue!

You received this because you either a) signed up for the newsletter directly on tonyalicea.dev or b) get emails from dontimitate.dev. If you enjoy my content, please let other people know about the newsletter! They can sign up here: https://tonyalicea.dev/newsletter/.

Happy coding!

Tony Alicea

Read the archive of this newsletter here.