Movie you've watched more than six times using GIFs.
("Hard mode" - no Star Wars, Star Trek, or LOTR)

The oral arguments whole purpose is for the justices to ask questions and lawyers to push arguments the lawyers see as important.

Half the time a really nasty question is being asked by a justice that agrees with that side in order to get the argument on record.

"Easy" questions are a bad sign.

It keeps showing up on best board games of 2025 (so far) lists and I'm now curious to try.

But you still need strong devs. You still need the human in the loop.

This isn't me arguing from "human superiority" or "anti-AI". This is me begging from a position of reality.

Sadly, I know that will be ignored for about five years, when some young hot upstart does artisanal human made code.

If you know coding basics and you know the language and you know AI basics AND you know the blindspots and how to get around them?

AI gen code can be a time saver.

Hell, even with all the complaints I just gave, I bought it for myself. It feels like pairing without another person. A+.

Add to this the AI blind spots that are just common throughout the whole industry: ezyang.github.io/ai-blindspots/

So where does that leave us?

And THEN we get to writing automation tests in that same IDE. Cyrpress, Playwright, Selenium. It has no freaking clue. It doesn't know these languages. It has never seen them before in its life. God bless it, it TRIES. But trying in this case is worse than nothing.

Beyond that? The whole thing is awful on greenfielding. Best way I can describe it is a kid at the candy store panic where everything gets tossed in the shopping cart but it all breaks open so individual candies are falling to the floor and most of the cart is empty wrappers.

And then there's the difficult thing where I wanted to turn a CLI into a GUI.

One of the outputs the AI gave me as it attempted this was "The user is hallucinating. The output and input should be displayed on the screen."

It was a full two days of trying to succeed multiple ways and just failure.

Wait, why that asterisk?

Well, it's complicated. For most things I've attempted I got working if not prod ready code.

Getting the code prod ready? That took an actual human and time. The IDE couldn't help on some of these edge cases.

I'm currently using the newest shiniest hotness. It's an IDE devoted to AI.

This thing is awesome. I bought it for myself personally and I'm a QA.

I feed it in problems, do rubber ducky talks with it, and POOF code that works!*

2024 ones weren't in the training data yet. I did a LOT of breaking the problem down to logical bits.

Feel free to complain it's not a fair data set it ain't. That similar puzzles are in the training data, they are.

Now, ask me about the same AI writing cypress or playwright code. Day and night.

So yeah, the coding AIs are great at getting someone who understands logic, but maybe not the syntax of code over the hump of writing. Do they make good choices? Hahahaha. no.

There should be no god forsaken reason you are putting in 15 hours days. With or without AI.

I just can't.

I'm just so confused. "We are a small team!" "ONLY Put in 15 hours a day!" "AI writes over half the code!" Then why the fuck are you having people put in 15 hours? I fucking write code with AI. It took me less than an hour to do advent of code with AI. ALL OF IT. I don't code well. 15 hours. Bull.

Had to Google the corrupted blood incident. 20 years this oct.

What hyped release of a re-found film left you cold?

I know I'm a lurker without even a profile picture. I'm putting on my QA hat here. For the end user though? They are errors.

They aren't errors. They are exactly what the system was trained to do. Generate. The system worked as intended.
I want more people to understand that so less rely on AI.

My job is 90% being a devil's advocate for good. It's a useful tool.

In most online discourse it's bringing a sextant to build a computer. Wrong tool. Wrong time. Also, the person bringing it can't use it.