AI ERA2023

Cheating Is All You Need

Steve Yegge · March 23, 2023 · Sourcegraph blog

Read the originalCheating Is All You Need ↗

TL;DR — Nine days after GPT-4 shipped, Yegge planted a flag: LLMs are the biggest thing to hit software engineering since the World Wide Web. Writing code with one feels like cheating — the same illicit thrill IDEs gave you — and that feeling is the tell that something real just changed. The catch: the model lies constantly, so the job becomes verifying its output. Don't. Trust. Anything! The title riffs on the transformer paper "Attention Is All You Need."

Most takes that week were cautious. Yegge's was not. He wrote from the seat of a veteran engineer — ex-Amazon, ex-Google — telling other veterans to stop hedging: this is historic, you will use it this year, and it will feel like getting away with something.

The thesis: "cheating" means using the LLM

The word "cheating" isn't an accusation — it's the reward. Yegge's claim is that letting a model do the typing produces the same guilty, too-good-to-be-legal rush that IDEs did when they replaced hand-rolled tooling. That feeling, he argues, is the surest signal that the leverage is real rather than hyped.

"There is something legendary and historic happening in software engineering, right now as we speak, and yet most of you don't realize at all how big it is."— Cheating Is All You Need, opening line

"They will feel gloriously like cheating, just like when IDEs came out, back in the days of yore."— Cheating Is All You Need

Why it's a step-change, not a gimmick

Yegge's frame of reference is deliberately enormous. This isn't the next incremental platform wave alongside social, mobile, and cloud — he reaches past all of them for the comparison that actually fits.

"LLMs aren't just the biggest change since social, mobile, or cloud–they're the biggest thing since the World Wide Web."— Cheating Is All You Need

And he doesn't hedge on timing. The prediction is immediate and personal — not "someday," but this calendar year, aimed at the reader.

"You will use them this year. They will absolutely blow your mind. And they'll continue to improve at an astonishing rate."— Cheating Is All You Need

The caveats: don't trust anything

For all the hype, Yegge is blunt that the model is not an oracle — it fabricates, and the engineer's job shifts from writing to verifying. The excitement and the alarm are meant to arrive together, not as a pick-one.

"Don't. Trust. Anything!"— Cheating Is All You Need, on verifying LLM output

Working with an approximate, unreliable helper you have to supervise is the discipline — which is his whole punchline about what the job actually is.

"Welcome to engineering motherfucker."— Cheating Is All You Need

"If you're not pant-peeingly excited and worried about this yet, well… you should be."— Cheating Is All You Need

Why it landed

Timing and voice did the work. GPT-4 had shipped nine days earlier, the industry was still deciding how seriously to take it, and here was a famously credible, famously profane engineer-blogger declaring the argument over. The essay ran on the Sourcegraph blog and doubled as the case for adopting an LLM-backed coding assistant — but it read as conviction, not a pitch.

It gave working engineers permission to be excited in plain, unhedged language, when most coverage was still qualifying every sentence.
The "cheating" frame stuck — it named the feeling everyone was quietly having and made it something to lean into rather than apologize for.
The title's riff on "Attention Is All You Need" planted it firmly in transformer-era lore before that lore had even settled.

The net effect: this short, loud post put Yegge at the front of the AI-coding conversation and set the tone — enthusiasm on top, verification underneath — that his later essays (agents, the "80% problem," the flat curve) would keep refining.

The title is a play on "Attention Is All You Need" (Vaswani et al., 2017), the paper that introduced the transformer architecture underneath every LLM Yegge is cheating with.