of your AI coding agent is critical to its performance. It is probably going one of the crucial significant aspects determining what number of tasks you may perform with a coding agent and your success rate in doing so.
In this text, I’ll discuss specific techniques I take advantage of to enhance the context of my AI agents. I’ll explain specifically how I do it, and why. It’s vital to grasp why I’m using these techniques, so you may start developing your individual techniques in the long run, and really optimize your agentic coding.
Table of Contents
Why optimize agentic context
The context you provide your coding agent is all the knowledge it has to finish a task. Thus, properly managing your context is incredibly vital in the event you want your coding agent to work well.
Improving your context by a number of percent may have a large impact in your efficiency as an engineer in the event you spend many hours every day programming. I thus spend a variety of time, always attempting to optimize my programming with my coding agent.
The 4 techniques I’ll present in the following section are a results of my testing a wide selection of various techniques and approaches. In this text, I’ll only cover 4 of a very powerful techniques and why they work so well. In the long run, I may additionally cover some failed techniques and reflect on why they didn’t work
4 specific techniques
On this section, I’ll cover 4 specific techniques I utilize to optimize the context for my coding agents. I’ve written the techniques in no particular order, and I consider all of them vital to me in my quest to be as efficient an engineer as possible.
All the time update AGENTS.md
Probably a very powerful technique I take advantage of is to always update the AGENTS.md file. Continual learning remains to be an unsolved problem for LLMs, thus we’d like to give you our own solutions to make coding agents remember our preferences.
I’ve written a rules file for my coding agent, which specifies some preferences I even have:
- All the time write Python 3.13 syntax if using Python
- Never use the Any type
- All the time use types and docstrings for functions
These are preferences I even have across all of the repositories I touch, and which I thus at all times want my agent to follow. I like to recommend spending time reflecting on your individual coding rules and specifying them to your agent.
Moreover, each time my coding agent makes an error, I help it correct the error and tell the agent to recollect the fix in AGENTS.md. This makes sure the agent avoids this error in the long run, and easily makes the agent faster and more efficient.
Should you proceed doing this over time, you’ll notice the agent becoming significantly higher and more adept at performing the tasks you ask it to perform. This could possibly be:
- Implementing recent features
- Fixing bugs
- Checking production logs
This works so well since you’re providing your coding agent with the obligatory context that you simply possess, but you never wrote down. By informing the coding agent in AGENTS.md, you provide the model critical context for problem solving.
Note that you may use any Markdown files that you simply prefer. Claude Code uses CLAUDE.md, Warp uses WARP.md, and Cursor uses .cursorrules. Nevertheless, I find that the majority coding agents at all times read AGENTS.md, which makes it an excellent file name to store agentic memory in.
Provide documentation links
One other tip is to offer relevant documentation links to the model, or to explicitly tell the model to seek out documentation online through an online search.
I sometimes find that my coding agent is using outdated syntax, for instance, when interacting with the OpenAI API. In these instances, I provide the model a link to the newest OpenAI documentation and tell it to base its code on this.
The issue of coding agents using outdated code typically occurs because LLMs have a cut-off date, which necessarily should be before the model was done training. The cutoff date for any given model could possibly be over a 12 months ago, by which a variety of API documentation has modified. Thus, it’s very vital to be certain that the model uses the newest available documentation by providing it with links to those docs.
Coding agents often uses outdated code due to the model knowledge cutoff. The fix to this problem is to offer the agents with the newest API documentation
Provide IaC stack as context
One other technique I utilize is to offer details about my infrastructure as code (IaC) stack as context to my coding agent. That is incredibly useful when using an agent to ascertain out production logs (which you need to do).
I began using this system after I noticed my agent was spending a variety of time finding information, equivalent to the names of my database tables. For instance, if the agent wanted to seek out information from a table, it first needed to list all tables, guess which table is relevant, and check out it. If it failed, it might need to try a unique table.
This takes a variety of time and tokens, costing you each efficiency and money, and is thus something you should avoid.
To offer my agent with all of the IaC context, I had an agent undergo the entire relevant IaC repositories and create a single Markdown file containing all relevant context, for instance, the names of all my database tables. I then provide this file as context to my coding agent each time it’s relevant.
Recent threads on a brand new context
One other easy technique I utilize is to start out recent threads each time I’m coping with recent contexts. For instance ,if I just finished implementing a brand new feature, and now wish to fix a bug, I almost at all times start a brand new thread in Cursor.
The explanation is that when implementing the brand new feature, the model stores a variety of context that is totally irrelevant to fixing the bug. This not only fills up the model context, but can even act as noise, distracting the model from more relevant information.
Thus, each time you may, you need to be certain that to start out recent threads each time changing contexts. This could possibly be after you implemented a brand new feature, and need to repair a bug. Or after you fixed a bug, and need to ascertain out production logs along with your agent.
This works well since the vital context that must be stored across threads is stored in AGENTS.md, as I discussed in an earlier section.
Conclusion
In this text, I’ve covered 4 specific techniques I utilize to optimize the context of my coding agents. Utilizing these techniques makes me a significantly more efficient engineer, because my coding agents can work rather more efficiently. I like to recommend trying out these techniques for yourself to seek out out in the event that they work well for you. Moreover, I like to recommend experimenting with recent techniques and approaches yourself, which might make you more practical. Each time you notice your coding agents are unable to do something, you need to immediately start ideating and serious about learn how to make them capable of perform such tasks.
👉 My Free Resources
🚀 10x Your Engineering with LLMs (Free 3-Day Email Course)
📚 Get my free Vision Language Models ebook
💻 My webinar on Vision Language Models
👉 Find me on socials:
🧑💻 Get in contact
✍️ Medium
