Transcription
You need to understand the six levels of Claude code if you don't want to get stuck in your progress. So, let me break down all six and the skills you need to level up.
Now, level one is the most basic. This is where you're just using Claude code as a straight blunt instrument. This is the prompter. You don't use it as a collaborator. You don't use plan mode. You just show up and you tell it what you want.
Now, to move on to the next level, we need to figure out how to write clear prompts, how do we evaluate Claude's prompts, and also basic terminal literacy, things like bash commands and understanding what Claude's doing under the hood.
And level two is where Claude code has become a partner. Now, the easiest way to do this is by simply using plan mode. Plan mode is going to force Claude code to ask you questions to make it a collaborative process. Also, you need to take an active role. Ask Claude code, like, what am I missing here? Because chances are if you aren't technical, you don't know what you don't know.
Now, the transition to level three is pretty natural. It's when you realize you need to start bringing outside information, outside context, into Claude code and begin managing that as well. Because level three is where you become a context engineer. And the biggest thing is management of the context window. Well, what's the context window? That's the amount of tokens that Claude code can use at any one time. It's 200,000 tokens. And the more tokens we use, the closer we get to that limit, the worse Claude code performs. At about 100,000, 110,000 tokens, its performance drops drastically. You need to manage that. On top of that, you need to figure out what sort of examples we need to feed it in the prompting page, what we need to do with Claude MD, and how to insert the right context at the right time.
And once we get a handle of the context window, that's when we move on to level four and start adding outside tools. Outside tools include things like MCPs and frameworks. And this is where people become sort of a kid in the candy store and just try to add as many things as possible. However, we need to learn to be surgical. And the only way you learn to be surgical and understand what the right tool is for the right job is by actually understanding what's going on under the hood. You don't need to know how to code, but you need to understand what a front end is, what a back end is, authentication, databases from a theoretical overview level. And the kid in the candy store is a very apt analogy because capability, adding all these different things, does not equal performance. Right? We need to add the right things at the right time.
Now, level five is all about skills and workflows and how we can use those two things to sort of supercharge our process. Now, luckily for us, they just released a brand new skill creator a couple days ago, which makes all these skills very easy to master. You just use the brand new skill creator skill, and it can create custom skills for you. It can audit uh current skills. It can do benchmarks. It can do tests. We now have data that can tell us, is the skill working the way it should? And is this the most optimal path forward?
Because level six is when we turn into a manager of Claude code and we start running multiple sessions at once. We start using things like work trees. We start opening multiple sessions. We start using experimental features like agent teams. And the idea behind all these is you are now deciding what gets done by who, not how these things get done. It's at this point of your journey that you are fully actualized with Claude code and you're able to take all the lessons learned from level one to five and scale them.