In a earlier article, I of the most important techniques I make the most of to code successfully with AI brokers. On this article, I’m persevering with with 4 extra methods, all of which I take advantage of each day.
I imagine that with a view to be an environment friendly programmer as we speak, you need to closely make the most of AI instruments. When you’re not coding utilizing AI brokers, you’re falling behind. Moreover, brokers can be utilized for a lot greater than coding as effectively:
- Brokers can learn and create Linear points
- Brokers can carry out deep analysis on a subject you’re fascinated about
- Brokers can evaluation log messages from manufacturing code
All of that are vital duties, programmers must carry out frequently.
Thus, I’m advocating for heavy use of AI brokers to be as environment friendly as attainable. On this article, I’ll cowl, on a excessive stage, 4 extra methods I make the most of that I imagine make me a extra environment friendly programmer.
- Macwhisper for sooner agent prompting
- Claude Code evaluation
- Parallel brokers
- Interacting with GitHub utilizing brokers
I’m additionally very fascinated about listening to you probably have any methods which might be vital in your programming workflows. When you’ve got particular methods in thoughts, be at liberty to achieve out, as I’d love to listen to about it.
Why it’s best to code with AI brokers
I’ve beforehand described how coding with AI brokers make me much more efficient as a programmer. I’ve multiplied my programming output many instances via using AI, and it merely permits me to do rather more than I did beforehand.
A typical counterargument to AI brokers is that you should perceive your code earlier than pushing it to manufacturing. I agree with this evaluation to some extent for those who’re working with crucial methods which might be arduous to carry out end-to-end exams on.
Nonetheless, most web sites and purposes aren’t like this. Initially, they’re not as crucial, and secondly, most duties you’re employed on as a programmer are verifiable. This implies you’ll be able to usually take a look at behaviour just by actually testing if the function works once you work together with it.
Thus, I’m advocating for extra use of AI brokers and for utilizing them for all programming-related duties. For instance:
- Create Linear points
- Repair bugs by merely linking to the Linear difficulty
- Planning and growing new options
4 Methods for coding effectivity
On this part, I’ll cowl 4 methods that I take advantage of for my AI-native programming workflows. These are particular methods that I actually use day by day I program.
Macwhisper
MacWhisper is a superb transcription software accessible on Mac. Merely put, Macwhisper means that you can press a button, speak to your laptop, and the textual content is routinely transcribed and pasted wherever your mouse cursor is.
That is useful as a result of quite a lot of my programming workflows have moved from pure code to pure language. Utilizing a transcription software for coding would naturally be arduous as a result of coding requires quite a lot of particular characters like colons, parentheses, and tabs, that are sooner to sort on a keyboard.
With AI brokers, increasingly more work is completed in pure langauge, as a substitute of coding language.
Thus, each time I immediate my Cursor agent, I often simply maintain down the button and say out loud no matter I need to immediate my agent. I’d, for instance, ask:
Test the logs for this doc id, was it processed appropriately <doc id>On this instance, I paste within the doc id after saying the sentence out loud.
The rationale I take advantage of Macwhisper is solely that I speak sooner than I can sort. The typical speaking velocity is round 150 phrases per minute, whereas most individuals can’t sort 100 phrases per minute at most velocity. Moreover, you’re not often capable of sort at max velocity when you need to suppose as effectively.
Claude Code evaluation
This step is cut up into two components:
- After implementing a function, I ask Cursor if the code is production-ready, and solely push when Cursor is glad
- Each time I make a PR, I’ve Claude Code to a code evaluation as effectively, with no different context than the PR description, and the Git diff file to the department I’m merging to.
This works very effectively. Asking Cursor if the code is production-ready makes Cursor do a evaluation of my modifications and repair any small points that may not work as meant.
Moreover, having a very separate LLM evaluation the code with no context of how the implementation was achieved is tremendous useful. This usually discovers different errors that I (or Cursor) didn’t take into consideration when implementing the code within the PR. This additionally considerably lowers the quantity of bugs skilled in manufacturing, and is a comparatively low cost addition you can also make to your CICD pipeline.
Parallel brokers (hearth and overlook)
One other vital method is to make use of parallel brokers. Each time I’m blocked by an agent doing a little work, I at all times begin a brand new agent. This may very well be a coding agent implementing one other function, or it may very well be Gemini deep analysis, researching a subject I’m fascinated about. The purpose is that I by no means merely wait on my agent with out doing anything.
When working parallel brokers, you would possibly begin scuffling with context switching. Switching contexts usually may be very taxing to your mind, and is certainly one thing you need to reduce.
Thus, I at all times make it possible for I work on a process till I’m absolutely blocked. I attempt to reduce the variety of instances I swap context, and solely begin a parallel process as soon as I’m positive I’ve to attend a while for my coding agent to complete its implementation.
One other vital level right here is that you simply give your coding brokers sufficient permissions to run for an prolonged time period. When you’re interrupted on a regular basis with the cursor asking for permissions, the parallel workflow doesn’t work effectively.
You must give your coding brokers sufficient permissions. When you’re at all times interrupted with a permission request, it’s arduous to work successfully.
Commit and PR with brokers
Lastly, I need to spotlight how I at all times work together with GitHub utilizing my coding brokers, as a substitute of writing the instructions myself. The rationale I do that is that it’s merely sooner, and I can do one thing else whereas my agent runs precommit hooks, commits, pushes, and makes pull requests.
Writing commit messages, pull request titles, and descriptions takes a shocking period of time. Particularly once you’re performing fast actions, similar to including translations or transferring a button within the UI. Due to this fact, I at all times make the most of Claude to jot down my commit messages, PR titles, and descriptions.
Not solely does this save me time, however I additionally suppose Claude does a greater job at writing these messages for me. With pull requests, for instance, it’s usually arduous for a human to recollect all the modifications made and to summarize them in a pleasant method. It’s a lot simpler for Claude to take a look at the Git diff and supply a abstract of all of the modifications made.
Thus, I’ve given Cursor permission to work together with GitHub for me. As a substitute of performing all the GitHub actions myself, similar to:
- Pulling
- Rebasing
- Commiting
- Amending
- Pushing
- Creating PR’s
I merely immediate Cursor to do it for me. Thus, I can simply hearth and overlook. My workflow after implementing a brand new function is solely to offer the next immediate to Cursor:
Run all precommit checks (black, mypy, pytest), commit and push.
Then create a PR on this department <linear department identify> and provides me
the hyperlink to the PRThat is a lot sooner than writing the GitHub instructions your self. Not having to jot down pull requests myself might be crucial bit, as this was tremendous time-consuming beforehand after I made my pull requests within the GitHub UI. Now I merely click on the hyperlink my agent gives me, and the PR is prepared. Then I take a look at the Claude Code evaluation offered to me, and repair any potential points.
Conclusion
On this article, I’ve mentioned 4 particular methods I take advantage of each single day after I’m coding. I mentioned Macwhisper for transcription, Claude Code evaluations, parallel brokers, and interacting with GitHub utilizing my agent. Collectively, I estimate that these methods save me at the least 1 hour every day, which is a major period of time. Liberating up this time permits me to finish so many extra duties over the course of a mission. I imagine that being efficient with AI brokers is an excellent vital ability, and undoubtedly a subject it’s best to spend time changing into good at.
👉 My Free Assets
🚀 10x Your Engineering with LLMs (Free 3-Day Email Course)
📚 Get my free Vision Language Models ebook
💻 My webinar on Vision Language Models
👉 Discover me on socials:
🧑💻 Get in touch
✍️ Medium
