You haven't read a book since high school, but now you're expecting others to decipher your 100 file AI-generated PR?

I know how good it feels. The buzz of the LLM feels like awaiting cash at an ATM. The constant stacking of green new files onto your diff reminds you of stacking dollars. Bountiful new lines in green that you hardly understand, but you love the shape of them, and CodeRabbit begrudgingly agrees (after the third re-review) that they meet your coding standards.

It's just as easy to keep going as it is to stop. Why would you stop at 20 files when you can open 100 and impress your non-technical engineering manager with your technical prowess? Besides, you tested locally (by yourself, the QAs were fired because of AI, unfortunately) and it LGT-You. The happy paths look joyous. Your code is behind a switch and isn't going to make it to prod before you leave for Punta Cana.

Here are some reasons to reconsider your approach to opening large PRs in the glory age of compute:

Increased Review Cycles, Exponential Increase in Procrastination

Every file you add to your PR will not only increase the time it takes to review it, but also the desire of the reviewing dev to procrastinate reviewing the PR. I genuinely believe your reviewer wants to do the best job that they can to review your PR. So, instead of reviewing your 100 file PR right now, they are going to go to Starbucks, work on their novel, call their mother, get married to their significant other at a notary, and perhaps then return to your PR sometime later. You have no leverage in this conversation either. You sent them a 100 file PR. Any excuse in the world will afford the reviewer sufficient time to buy time before reviewing your PR.

You Are Holding Your Reviewer Captive

Did you know the Bible and Torah are quite large, and consequently they're read in clearly defined daily and weekly segments?

Here is the origin of Torah portions from Wikipedia:

The custom of dividing the Torah readings dates to the time of the Babylonian captivity (6th century BCE). The origin of the first public Torah readings is found in the Book of Nehemiah, where Ezra the scribe writes about wanting to find a way to ensure the Israelites would not go astray again. This led to the creation of a weekly system to read the portions of the Torah at synagogues.

You are holding your reviewer in the modern equivalent of Babylonian captivity.

Cognitive Complexity - Human Biology, Not a Skill Issue

As the PR size increases, so does the cognitive complexity. The reviewer not only has to remember how your Abstract Factory pattern was defined in file 34, but in theory they also have to see if "you" wrote sufficient test cases for the code that Opus wrote (but that you GPG signed off on as yours). Now they're reading file 52 and they realized you defined a façade in file 12 that you no longer need to use.

Supposedly humans can remember around 7 items in their short-term memory, so your hundreds of files have saturated your reviewer's brain to the point where they have long forgotten your PR, and perhaps their own identity and will to continue living in the city (they dream of the countryside). This is the most important part:

Now that the cognitive complexity of your PR has reached a critical mass, devs will not be reading line by line anymore. They will rely on their arsenal of weapons-grade AI review tools (CodeRabbit, Convex, Cursor, Claude, Codex) to offload the work from their saturated brain. These tools are very good, and they will find issues in your PR, and when they do you will spend another day fixing those issues, and in microeconomic terms, these are inflationary levers that will balloon your cycle time. Ask yourself how that will look on the Jira burndown chart.

Your Teammates Will Associate Your Name With Negative Experiences

In Pavlovian fashion, your teammates will see your name and associate it with effort. You will notice teammates responding slower, code reviews lagging, and you will no longer get invited to the unofficial team dinners (they haven't had time to review your PR yet and they will be embarrassed that they have enough free time to eat).

Other Types of Self-Sabotage to Try Instead

Keep paying for the AI subscription you haven't been maxing out. You need Codex in your back pocket. Even if you don't need it for coding, you may never know when you might need an emergency couscous recipe for dinner on the fly, and you wouldn't risk using GPT-5.old-auto for that would you? All the cooking websites have too many ads anyway. Your insulation from those corporate nudges is well worth the cost.

Never ship. Keep building. You now have more compute than time. You can continue to build features for your nonexistent user base. As days turn to nights you are achieving glory every day every time you GPG sign a new commit. A grey box turns green and you have struggled for today, like Sisyphus. Besides, Vercel just updated Next.js with a new patch. It implements an LRU cache with invocation ID scoping for minimal mode response cache. Your customers could benefit from this. Tomorrow you will do this same round of updates. Claude can take it, don't worry. Hopefully the end of this hedonistic commit cycle will be quick and not afford you too much time for self-reflection.

"Your worst sin is that you have destroyed and betrayed yourself for nothing"
- Fyodor Dostoevsky