Getting started with the tools and systems of ai.

Two monitors, both off

AI moves fast, and it can be overwhelming if you focus on the hype. I want to give you a practical approach to help you get going with confidence.

set yourself up for success.

Work in folders

Whether Claude Cowork or Code, folders are your friends. I have a master folder for customer-related work and have broken it down into sub-folders by account. This lets me keep context contained to a particular focus area.

Create READMEs

A README is a markdown or txt file that sits at the top level of a folder and is designed to overview or summarize what’s in a directory. It’s a great way to help AI find what it’s looking for more quickly.

Use Git

You don’t need to learn Git, you just need to use it. Git is version control for your files. It lets you keep track of changes and roll back when AI does something wrong or makes a mistake. AI can help you set up your folder as a ‘repo’ and teach you things like ‘committing’ (saving) files. If you want to share work with others, github.a8c.com is where you can ‘push’ your repo so it has a ‘remote’ (not on your machine) place to live.

Leverage scheduled tasks

My biggest pet peeve with any system is that it gets more cumbersome and annoying to maintain the longer you use it. With AI, from the get-go, you can set up scheduled tasks to do the things that make systems onerous. Have the AI process your transcripts, organize your files and update your READMEs so that every day, you show up to a clean working space, ready to go.

How I talk to AI and get things done.

I don’t format my prompts

Prompt engineering was a thing. Honestly, I don’t do that when I start a task. I just tell it what I want, point it at a folder and hope for the best. If I need it to do something specific, I use a skill.

I use skills a lot

If I have a task I do over and over again or something I want done in a very specific way, I’ll ask AI to build a skill for it. This is the way I get a (mostly) predictable output every time.

I rarely use Fable

Opus 5 is my workhorse model for code. Fable burns tokens when most tasks don’t need it. Figuring out which tasks should use which model is a worthwhile endeavour. As token spend gets more scrutinized, leaning into Haiku and Sonnet can help keep things under control.

I’ve set up a local dev environment

Setting up a local env for WordPress VIP used to cause some headaches for us and our customers. With AI, point it at the docs, and it can get you set up in five minutes. Then you have a local way to play with our tools and systems. It is absolutely worth installing WordPress locally and building your own website with AI. Even if no one sees it, knowing how all the pieces fit together will help you with your role.

I ask for hard things

AI will tell you something is hard. That doesn’t mean it’s impossible; it just means it’s hard. I’ve found that sometimes the AI will try to get out of doing something (it will defer or only pretend to finish something). You have to check its work regularly but the payoff is often worth it.

HOW TO KICK OFF

You’ll likely hear the term PRD or Spec when you speak to folks in PED (product, engineering, design). It’s a document that these teams use to ground whatever they’re working on. It’s the source of truth for what is being built and what is not.

Whether you are building a deck, a report, a new plugin or something else, a plan is important. For anything more than a simple bit of output, I ask AI to write a plan or spec first. This artifact survives an AI session, which tends to get forgetful the longer it goes.

The more intense a project is, the more time I invest in the plan. I’ll have AI review the plan, and in some cases, I’ll even have ChatGPT review a Claude plan to see if it can find any issues or gaps (it’s called an adversarial review).

A few of my go-to skills

I use a mixture of my own skills and collections from others. I like to cherry-pick the ones I like and have built my own workflows to help me get things done.

/plan

Turns a request into a written packet: what is being built, what it has to do, and what is deliberately left out. It researches what already exists before it writes, so the plan cites rather than reinvents.

compound-engineering

/report & /deck

Uses a VIP-inspired design system to create HTML reports and decks.

atelier

/impeccable

Takes anything I have had AI-designed and critiques it for slop and obvious AI tells in design.

Impeccable

/edit

Reviews any report, deck or p2 I co-create with AI and tries to “de-slopify” it. Removes em dashes, checks voice and tone, shortens it, removes obvious AI phrases, etc.

Atelier

/wwjd

“What Would James Do” is a skill I use when trying to solve hard problems. It pulls in the best management consulting frameworks from MBB to help me create strategy.

Atelier

Marquee, from below

Still unsure where to start? No worries. I am happy to hang out with anyone to help unpack how I use AI, the systems and the tools I use. Ping me on Slack or add a time to my calendar.