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 strategies, all of which I take advantage of every day.
I consider that with the intention to be an environment friendly programmer right now, it’s important to closely make the most of AI instruments. Should 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 excited about
- Brokers can assessment 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 potential. On this article, I’ll cowl, on a excessive degree, 4 extra strategies I make the most of that I consider make me a extra environment friendly programmer.
- Macwhisper for sooner agent prompting
- Claude Code assessment
- Parallel brokers
- Interacting with GitHub utilizing brokers
I’m additionally very excited about listening to in case you have any strategies which are vital in your programming workflows. When you have particular strategies in thoughts, be happy to achieve out, as I’d love to listen to about it.
Why you need 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 occasions via using AI, and it merely permits me to do rather more than I did beforehand.
A typical counterargument to AI brokers is that it’s essential perceive your code earlier than pushing it to manufacturing. I agree with this evaluation to some extent if you happen to’re working with crucial techniques which are onerous to carry out end-to-end checks on.
Nevertheless, most web sites and purposes are usually not 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 typically check behaviour just by actually testing if the characteristic works if 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 subject
- Planning and creating new options
4 Methods for coding effectivity
On this part, I’ll cowl 4 strategies that I take advantage of for my AI-native programming workflows. These are particular strategies that I actually use each day I program.
Macwhisper
MacWhisper is a superb transcription device obtainable on Mac. Merely put, Macwhisper means that you can press a button, discuss to your laptop, and the textual content is robotically transcribed and pasted wherever your mouse cursor is.
That is useful as a result of a whole lot of my programming workflows have moved from pure code to pure language. Utilizing a transcription device for coding would naturally be onerous as a result of coding requires a whole lot of particular characters like colons, parentheses, and tabs, that are sooner to sort on a keyboard.
With AI brokers, increasingly more work is finished in pure langauge, as an alternative 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 wish to immediate my agent. I would, for instance, ask:
Test the logs for this doc id, was it processed accurately On this instance, I paste within the doc id after saying the sentence out loud.
The explanation I take advantage of Macwhisper is solely that I discuss sooner than I can sort. The common speaking velocity is round 150 phrases per minute, whereas most individuals can’t sort 100 phrases per minute at most velocity. Moreover, you’re hardly ever capable of sort at max velocity when it’s important to suppose as effectively.
Claude Code assessment
This step is break up into two elements:
- After implementing a characteristic, I ask Cursor if the code is production-ready, and solely push when Cursor is happy
- At any time when I make a PR, I’ve Claude Code to a code assessment 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 assessment of my modifications and repair any small points that may not work as meant.
Moreover, having a totally separate LLM assessment the code with no context of how the implementation was achieved is tremendous useful. This typically 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 (fireplace and overlook)
One other vital method is to make use of parallel brokers. At any time when I’m blocked by an agent doing a little work, I all the time begin a brand new agent. This may very well be a coding agent implementing one other characteristic, or it may very well be Gemini deep analysis, researching a subject I’m excited about. The purpose is that I by no means merely wait on my agent with out doing the rest.
When operating parallel brokers, you may begin scuffling with context switching. Switching contexts typically could be very taxing in your mind, and is certainly one thing you wish to decrease.
Thus, I all the time make it possible for I work on a job till I’m absolutely blocked. I attempt to decrease the variety of occasions I change context, and solely begin a parallel job as soon as I’m certain I’ve to attend a while for my coding agent to complete its implementation.
One other vital level right here is that you just give your coding brokers sufficient permissions to run for an prolonged time frame. Should 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. Should you’re all the time interrupted with a permission request, it’s onerous to work successfully.
Commit and PR with brokers
Lastly, I wish to spotlight how I all the time work together with GitHub utilizing my coding brokers, as an alternative of writing the instructions myself. The explanation 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 stunning period of time. Particularly if you’re performing fast actions, comparable to including translations or shifting a button within the UI. Due to this fact, I all the time 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 typically onerous 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 an alternative of performing all the GitHub actions myself, comparable to:
- Pulling
- Rebasing
- Commiting
- Amending
- Pushing
- Creating PR’s
I merely immediate Cursor to do it for me. Thus, I can simply fireplace and overlook. My workflow after implementing a brand new characteristic is solely to supply the next immediate to Cursor:
Run all precommit checks (black, mypy, pytest), commit and push.
Then create a PR on this department and provides me
the hyperlink to the PR That is a lot sooner than writing the GitHub instructions your self. Not having to jot down pull requests myself might be an important bit, as this was tremendous time-consuming beforehand once I made my pull requests within the GitHub UI. Now I merely click on the hyperlink my agent supplies me, and the PR is prepared. Then I take a look at the Claude Code assessment supplied to me, and repair any potential points.
Conclusion
On this article, I’ve mentioned 4 particular strategies I take advantage of each single day once I’m coding. I mentioned Macwhisper for transcription, Claude Code critiques, parallel brokers, and interacting with GitHub utilizing my agent. Collectively, I estimate that these strategies save me not less than 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 challenge. I consider that being efficient with AI brokers is an excellent vital ability, and positively a subject you need to spend time turning into good at.
👉 My Free Sources
🚀 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
