This is my petition to coin a new term - AI Goulash.

Goulash is a human led mix augmented by AI. It keeps the weird carrots. Don’t ask about the ingredients - it’s unclear what the hell is actually in it half the time. It should never quite be the same thing twice (make a joke about non-determinism if you like). You eat goulash and ask questions later. It does not care if you are vegan. You hope for savory chunks and ingest it at risk of gastrointestinal distress.

Goulash is the adventure I want from AI.

Why now?

Volume is winning. There are more words, images, and videos of bunnies jumping on trampolines than hours in your life, and you will not beat paste with more paste. The better move is to assemble from your own inputs and let a model assist with structure and speed.

The only move is to cook. Goulash lets us salvage all the leftovers the day throws at us - voice notes, charts, code snippets, the “is this something”s. Toss them in and ask the model to chop fine or leave chunky.

What counts as AI goulash?

Using inputs that actually belong to you. These can be scribbles, photos of whiteboard, that email you never sent, or a janky bash script before my boy Claude Code has entered the chat. Keep the raw stuff with fingerprints (Is that analogy making this sound like light promotion of cannibalism?). At risk of overusing the term “taste” - Goulash has a taste. Not a brand voice sheet, but an unapologetic bias. The fact that I’m the only Android user in the Bay area and can’t do that bump share contact thing. Whatever it is GenZ means by “aura” (I think?)

What does not count?

Anything that reads like it was sanded smooth by a thousand identical summaries. A product of OpenAI taking the entire internet, computing a lot of dot products, jiggling it around and giving you a UI to generate text in Thinking Mode. A sea of goddamn em dashes. Slop removes you. Goulash reveals you.

How do I make goulash?

I say it how it is in your frantic head. I ignore spelling and grammar. I’m still deciding if Goulash needs to be capitalized. Then, I let the model sort, group, and compress. I want to keep the bits that taste like something.

Slop wants puree. Goulash wants pieces. Examples from my actual day, not from the library of nowhere or a “@grok is this true?”. I iterate and tell the model what it missed.

The model can augment you, but please make your own dish.

GPT-5 helped me make this one in the style of Seurat. (It was on the first try so I’m kind of not proving my point.)

It's pronounced DEE-mun not DAY-mon and this is a hill Ann will die on.

We’re going to write some content about tech things. We hope you like it. Both of us hail from Northeast Ohio and have Sun Microsystems jackets. Views are our own.

Figure 1: a height mogging incident

Jacob

Jacob is a Stanford undergrad with a background in physics, chip fab, and scraping the interwebs. Like all youthful men in tech, Jacob is thinking about selling his plasma to Peter Thiel. In his spare time, Jacob enjoys ocean swimming and poker.

Figure 2: Jacob outside

Ann

Ann is an engineer with a background in security, infrastructure, and magic internet money. Like all self-respecting SF residents of a certain age, she was a B2B SaaS founder in a previous life. In her spare time, Ann enjoys blob storage and Taylor Swift.

Figure 3: Ann and monitors

Subscribe now