| Website | https://www.stellman-greene.com |
| https://twitter.com/AndrewStellman | |
| https://www.linkedin.com/in/andrewstellman | |
| GitHub | https://github.com/andrewstellman/ |
Turns out it was calling a method that generates a new ID and timestamp right before encoding the message.
Vibe coding is great, until you actually need to turn your headlights on.
Classic โCopilot got me 80% of the way and 100% confusedโ fail.
Working on a part of an app that pushes a message to a queue. I wrote a unit test to pull a message off and check the ID and timestamp. They were always wrong.
Spent about ten minutes spinning my wheelsโkept asking Copilot to figure out if it was encoding something wrong, or decoding it weird, or maybe setting a value somewhere it shouldnโt.
Derp, ok, let me actually. Read. The. F*ckn. Code. Jeez.
๐ But if you give it a prompt like this:
๐๐๐ฒ ๐๐ญ๐ญ๐๐ง๐ญ๐ข๐จ๐ง ๐ญ๐จ ๐ฌ๐๐ฉ๐๐ซ๐๐ญ๐ข๐จ๐ง ๐จ๐ ๐๐จ๐ง๐๐๐ซ๐ง๐ฌ. ๐๐ข๐ฏ๐ ๐ฆ๐ ๐ ๐ฉ๐ฅ๐๐ง ๐ญ๐จ ๐ฆ๐๐ค๐ ๐ฌ๐ฎ๐ซ๐ ๐ฒ๐จ๐ฎ ๐๐จ ๐ข๐ญ ๐ข๐ง ๐ญ๐ก๐ ๐๐จ๐๐ ๐ ๐ฃ๐ฎ๐ฌ๐ญ ๐๐ฌ๐ค๐๐ ๐ฒ๐จ๐ฎ ๐ญ๐จ ๐ ๐๐ง๐๐ซ๐๐ญ๐.
Itโll often slow down and come back with something pretty reasonable.
Still not perfect, but way better than just asking it to "write the code."
Did you know Copilot can be ๐ โ๐๐๐๐๐๐๐๐ฆ bad at separation of concerns? I meanโsuper stinky terrible bad.
Like, itโll give you a class that connects to the database, builds the HTML, sends the email, and logs the user inโall in one class.
I wish someone had told me that code is written once, but read over and over againโby other people, and by future you. I eventually read it in Software Engineering at Google, and it clicked instantly. It wouldโve saved me years of digging through undecipherable code I wrote myself.
Iโve been making readability a priority for a long time now. Itโs even something I teach new developers in Head First C#, because it changes how you work.
โฆand I donโt think it should be something we pass on to junior developers like itโs inevitable.
I donโt think going through a โsh*tty code phaseโ should be a rite of passage. You donโt need years of experience to write code that makes sense laterโyou just have to think about how itโll feel to come back to it.
I want more junior developers to hear that early, because I didnโt.
Every week or two I see that meme where a developer looks at old code, gets mad, then realizes they wrote it.
Itโs funny, right? Weโve all been there!
Butโฆ
But, honestly, I also feel conflicted about it, because it reinforces a myth thatโs actually kind of damaging: the idea that youโll always hate your old code. Itโs just part of the job, right?
No!! I donโt think thatโs true at allโwriting bad code is NOT part of the jobโฆ
๐น A Java service that uses OpenNLP to extract names and places from incoming data and log them for follow-up.
๐น A Python script that loads a CSV of transaction data, uses scikit-learn to flag potential anomalies, and writes the results to a simple dashboard or log file.
Projects like that are small enough to build in a weekend, but teach you a lot. Youโll see what the tooling gives you, where it falls short, and how much you still need to think about the rest of the system to make it work.
That kind of thinking gets more valuable, not less.
But you can absolutely start exploring AI without walking away from backend work. One of my favorite ways to do that is by adding a small machine learning feature to something youโve already built.
For example, a few small learning projects Iโve seen:
๐น A C# app that pulls customer support messages from a database and uses https://ML.NET to classify them into categoriesโthen flags anything the model isnโt confident about.