can produce an unimaginable quantity of content material in a brief span. This might be creating new options, reviewing manufacturing logs, or fixing a bug report.
The bottleneck in software program engineering and knowledge science has moved from growing code to reviewing what the coding brokers are constructing. On this article, I talk about how I successfully evaluate Claude output to be an much more environment friendly engineer utilizing Claude Code.
Why optimize output reviewing
You would possibly marvel why it is advisable to optimize reviewing code and output. Just some years in the past, the most important bottleneck (by far) was writing code to supply outcomes. Now, nevertheless, we are able to produce code by merely prompting a coding agent like Claude Code.
Producing code is solely not the bottleneck anymore
Thus, since engineers are all the time striving to determine and decrease bottlenecks, we transfer on to the subsequent bottleneck, which is reviewing the output of Claude Code.
In fact, we have to evaluate the code it produces via pull requests. Nonetheless, there may be a lot extra output to evaluate in case you’re utilizing Claude Code to unravel all doable duties (which you undoubtedly needs to be doing). You want to evaluate:
- The report Claude Code generated
- The errors Claude Code present in your manufacturing logs
- The emails Claude Code made on your outreach
You have to be making an attempt to make use of coding brokers for completely each activity you’re doing, not solely programming duties, however your entire industrial work, making shows, reviewing logs, and every part in between. Thus, we have to apply particular methods to evaluate this content material quicker.
In my subsequent part, I’ll cowl among the methods I exploit to evaluate the output of Claude Code.
Strategies to evaluate output
The evaluate method I exploit varies by activity, however I’ll cowl particular examples within the following subsections. I’ll maintain it as particular as doable to my precise use instances, after which you may try to generalize this to your personal duties.
Reviewing code
Clearly, reviewing code is among the most typical duties you do as an engineer, particularly now that coding brokers have turn into so fast and environment friendly at producing code.
To extra successfully carry out code critiques, I’ve executed two foremost issues:
- Arrange a customized code evaluate talent that has a full overview of effectively carry out a code evaluate, what to search for and so forth.
- Have an OpenClaw agent robotically run this talent every time I’m tagged in a pull request.
Thus, every time somebody tags me in a ballot request, my agent robotically sends me a message with the code evaluate that I did and proposes to ship that code evaluate message to GitHub. All I must do then is to easily have a look at the abstract of the ballot request, and if I wish to, merely press ship on the proposed ballot request evaluate. This uncovers a whole lot of points that would have gotten to manufacturing if not detected.
That is in all probability probably the most beneficial or time-critical reviewing method that I’m utilizing, and I might argue environment friendly code critiques are in all probability one of the necessary issues corporations can give attention to now to extend pace, contemplating the elevated output of code with coding brokers.
Reviewing generated emails
One other widespread activity that I do is producing emails that I ship out via a chilly outreach instrument or emails to reply to folks. Oftentimes I wish to evaluate these emails additionally with formatting. For instance, if they’ve hyperlinks in them or some daring lettering and so forth.
Reviewing this in a text-only interface resembling Slack will not be an excellent situation. To start with, it creates a whole lot of mess within the Slack channel and Slack additionally isn’t in a position to format it appropriately all the time.
Thus, one of the environment friendly methods of reviewing generated emails and usually formatted textual content I’ve discovered is to ask Claude Code to generate an HTML file and open it in your browser.
This permits Claude Code to extremely shortly generate formatted content material, making it tremendous simple so that you can evaluate. Claude can’t solely present the formatted emails but in addition present it in a really good method, which individual is receiving which electronic mail, and likewise in case you’re sending electronic mail sequences, it’s tremendous simple to format.
Utilizing HTML to evaluate outputs is among the secret hacks which have saved me, that saves me hours of time each week.
Reviewing manufacturing log experiences
One other quite common activity I exploit Cloud Code for is to evaluate manufacturing log experiences. I sometimes run a each day question the place I analyze manufacturing logs, on the lookout for errors and issues I ought to concentrate on, and even simply log warnings within the code.
That is extremely helpful as a result of reporting providers that ship alerts on errors are sometimes very noisy, and you find yourself getting a whole lot of false alerts.
Thus, as an alternative, I choose to have a each day report despatched to me, which I can then analyze. This report is distributed with an OpenClaw agent, however the best way I preview the outcomes is extremely necessary, and that is the place HTML file formatting is available in once more.
When reviewing these manufacturing logs, there may be a whole lot of info. To start with, you’ve gotten the totally different error messages which you can see. Secondly, you’ve gotten the variety of occasions every error message has occurred. You might need totally different IDs that refer to every error message that you simply additionally wish to show in a easy method. All of this info is tremendous troublesome to supply in a pleasant method in txt formatting, resembling slack for instance, but it surely’s extremely good to preview in an HTML file
Thus, after my agent has reviewed manufacturing logs, I ask it to supply a report and current it in an HTML file, which makes it tremendous simple for me to evaluate all of the output and shortly acquire an summary of what’s necessary, what they’ll skip, and so forth.
One other professional tip right here will not be solely to generate the HTML file but in addition to ask Claude Code to open it up in your particular browser, which it does robotically, and also you shortly get an summary. And mainly get notified every time the agent is completed as a result of the browser pops up in your pc with a brand new tab holding the HTML file that was generated.
Conclusion
On this article, I’ve lined among the particular methods I exploit to evaluate Claude Code output. I mentioned why it’s so necessary to optimize reviewing outputs, highlighting how the bottleneck in software program engineering has shifted from producing code to analyzing the outcomes of code. Thus, because the bottleneck is now the reviewing half, we wish to make that as environment friendly as doable, which is the subject I’ve been discussing right here immediately. I talked concerning the totally different use instances I used Cloud Code for and the way I effectively analyze outcomes. Additional bettering the best way you analyze the output of your coding brokers shall be extremely necessary going ahead, and I urge you to spend time optimizing this course of and fascinated with how one can make reviewing coding agent output extra environment friendly. I’ve lined some methods that I exploit on a day-to-day foundation, however in fact, there are a whole lot of different methods you should use, in addition to the truth that you’ll have your personal set of exams that can require their very own set of methods to make use of which might be totally different from mine.
👉 My free eBook and Webinar:
🚀 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:
💌 Substack
