I'll cover some of the newest techniques I've invented and actively use whenever I write about Claude Code and Codex. Both of these are excellent coding models that I use every day when programming. They are great models out of the box; However, if you do it right, you can get a lot out of them.
This is what I am going to say in this article. I'm going to discuss some strategies you can use right now, which will increase how much you can get out of your coding agents. I strongly urge you to try these methods first, as I believe they can be beneficial for all coders.
I'm going to present some specific techniques that are quick updates, but some of them are also along lines of thought that are thought processes that you want to apply to your coding practices. This can also be seen as inspiration on how to improve your general coding.
Why did you expand Code Claude
First, I always like to cover why you should be interested in the topic. The reason you should expand Code Claude and Codex is that if you use the right techniques, you can get a lot more out of coding agents.
A common saying is:
The biggest beneficiaries of AI are the people who are already at the forefront
This means that AI is, of course, a skill enhancer, but it doesn't work on a flat level. It works mostly as an enhancer of your current skills.
Let's pretend that programming ability is a scale where you score your ability in programming and how effectively you can implement new solutions. For example, if your coding skill level is 10 points, the AI might make you work three times as hard, and you'll end up with a total of 30 points.
However, if your coding skill level is 50 points instead, you'll still get that 3x performance boost, and you'll end up with 150 points. Previously, the difference in skill points was 20 points, and now, with the use of AI, it has become 120 points. This highlights my view on AI: the people who benefit most from using these new forms of AI are the people who are already at the forefront. If you can use some techniques to get more out of the coding agent, you will get a big productivity boost.
Direct ways to increase Code Claude
Now I'm going to get into some of the techniques I use to boost Code Claude and the Codex. I will cover four techniques in this article.
- Heavy use of OpenClaw and cron jobs. In general: use as many tokens as possible
- Effective use of Claude Code hooks
- Ultracode leverages the effort used by coding agents to perform more advanced work
- Have your writing agent present the remaining tasks and a summary at the end of their responses
Heavy use of OpenClaw
The first method or idea I will cover is to actively use OpenClaw and, in general, try to use as many tokens as possible.
OpenClaw is a system where you can find bots that work in your messaging channel, for example, Discord or Slack. These bots can be powered by Claude's Codex API or through your Codex subscription. Bots are essentially AI agents that run periodically (which you can do with cron jobs), or you can make them react to specific messages or events. In general, it's easier if I'm more specific about how to use it, so I'll include some areas of use where you can find OpenClaw agents working.
- Have an OpenClaw agent message every time you're tagged in a GitHub pull request and do an automatic code update for you
- Have an agent check your product every night and report problems to you in the morning
- Have the agent perform bug checking automatically so you don't have to check for bugs yourself
There are, of course, many other possible use cases for OpenClaw. The whole idea is that it's basically a 24/7 coding agent, which you can set up to do certain tasks, and you don't have to sit in the driver's seat of the model. The model will make all the decisions itself.
Effective use of Claude Code hooks
Claude Code Hooks is also a very interesting idea. Hooks are basic code that you can use at specific points in time. Some of the different hooks you have in Claude Code are:
- At the beginning of Claude's Code
- At the end of Claude Code
- Whenever the agent asks the user a question
- Whenever an agent completes a task
Basically, you can ensure that a piece of code always runs whenever any of these events occur, and of course, several other events. To be specific, for example, you can have Code Claude automatically combine information from the music you're working on when you close it. Or you can have it ping your computer with a sound whenever it asks you a question and completes a task.
The ping sound is something I just did and how I'm very happy with it. I make my Claude Code make a sound on my computer whenever it asks me a question or when it finishes a task I need to review.
This is good because I don't need to pay attention to the terminal itself. I just need to wait for the sound, then I know I have to check my agents. This makes it easier for me to focus on other tasks once I have spun my agents.
Claude Code Ultracode – super effort coding agents
Claude Code recently released something called Ultracode with more high-level thinking. This is a method where you create a group of agents that perform many different tasks, and it is a way to spend more tokens while completing a task.
In general, using more tokens is a good thing because performance tends to scale with the number of tokens used. Of course, this reaches a limit at some point, so spending a billion tokens regularly would not make AGI, of course. In general, spending more tokens makes agents perform better, and Ultracode with Claude Code makes you spend more tokens.
Now you may be wondering:
Spending more tokens means the agent spends longer to complete the task, right?
The argument here is that it doesn't matter if the agent takes a long time to complete the task; the only thing that really matters is that the agent is able to complete the task successfully, or if you need to correct mistakes made by the agent.
To be specific, you can basically choose between the following:
- Use a fast and cheap model that spends few tokens, completes the implementation in ten minutes, but you have to spend an hour and a half afterwards to fix its mistakes and iterate on it until you reach the desired solution.
- Spend 10 minutes talking to Claude Ultracode, then Claude Ultracode spends 30 minutes actually booting, 40 minutes total. Yes, you spent a lot of time initially talking to the coding agent, and the coding agent spent a lot of time during implementation, but overall, you spent a lot less time because you didn't have to spend time debugging the agent and iterating with it afterwards.
If you look at it from this point of view, it is a very easy decision. Basically you should always choose a model that spends a long time but provides high quality.
Present the remaining tasks and review at the end of the answers
Another great thing I've started doing recently is to have my agents deliver the tasks I need to do, and a recap at the end of each response. I did this by simply notifying my agents at the user level, the CLAUDE.md file. I said at the end of your answer, always, when you ask me to do something, use the following syntax.
- []
- []
- []
... This way it is very easy for me to see if the agent is asking me to do something or if I need to check something. I started doing this because I realized that I can't really read everything that Claude Code gives me, because it writes too much text. In many cases, Claude Code would ask me to do something, but I wouldn't actually read the response it was giving me, and I wouldn't recognize it when it asked me to do something.
This, of course, is a problem, as it can have a negative impact on quality. However, now that I'm doing it using the checkboxes, I quickly realize that Claude Code has jobs for me, and I can easily get rid of them.
This is especially important if you have multiple agents running in parallel and you tend to be away from a particular thread for more than 10 minutes. In these cases, it is often difficult to remember exactly what you were doing in a particular thread when you return to it. You need an easy way to get back to the thread quickly and know what to do.
Another point in this area is that I make my coding agents do a recap as well, under the tasks they ask me to do. This is already revealed in the Claude Code. You can do it to create a recap; however, I've noticed that the repetition is often delayed, and I often want to repeat quickly. Instead I have Claude Code create the recap itself, which makes it very easy for me to return to a thread I've been away from for 10 to 30 minutes while working on other coding agents.
The conclusion
In this article, I've discussed the newest techniques I'm using now to get the most out of my Claude Code. I've covered why it's so important to consistently improve the way you program with coding agents, highlighting that the best people are the people who benefit the most from coding agents. If you want to get the most out of new AI tools, you need to spend time getting them right. Then I put together four specific strategies that I think you can try right away:
- Using OpenClaw
- Using Claude Code hooks
- Using as many tokens as possible with Claude Code Ultracode
- Ensuring that Claude's code presents the remaining functions and iterations at the end of the responses
I believe that if you use these four techniques, you will notice an immediate improvement in your efficiency with Claude Code. I also encourage you to always think of other strategies you can use that can improve your work with coding agents like Claude Code and Codex.
Also check out my article on How to Run Multiple Encoding Agents in Parallel.
👋 Touch
👉 My Free eBook and Webinar:
🚀 10x Your Engineering with LLMs (Free 3-Day Email Course)
📚 Get my free ebook Vision Language Models
💻 My webinar on Vision Language Models
👉 Find me on social media:
💌 Stack
🐦 X / Twitter
