Riley Goodside Says Prompt Engineering Is Dead: Ordinary Users Can Just Ask Directly

The person who sentenced it to death is none other than Riley Goodside—the very man who personally turned "prompt engineer" into a formal job back in the day.

The founding father personally wrote an obituary for his own livelihood.

Here's what happened.

On October 22, security researcher Justin Elze posted on X, and the body of the post was just one line: You're all prompting wrong.

The images attached were two LinkedIn screenshots.

The next day, Riley Goodside reposted it and casually dropped three sentences:

Prompt engineering is dead. Prompt however you want. Have fun.

Three sentences, ten English words in total. But those ten words set off a massive wave in the AI world.

This isn't a statement that carries the same weight no matter who says it.

Goodside worked as Scale AI's chief prompt engineer, and later joined Google DeepMind to continue in the same field. Scale AI CEO Alexandr Wang personally endorsed him back then, saying he'd bet Goodside was the world's first formally employed prompt engineer.

He made his name in his early years by publicly running experiments, focusing specifically on how to interact with large models—even the security concept of "prompt injection attacks" was one of the key directions of his early research.

One netizen's comment hit the nail on the head: coming from a man who almost single-handedly created this position, this is truly an interesting eulogy.

You've been through the days of chanting "spells" too, haven't you?

Let's turn back the clock a bit.

When ChatGPT first blew up, whose bookmarks didn't have a few "universal prompt templates" sitting in them?

First assign the AI an identity—"You are a senior XX expert"—then specify the output format, then add the reminder "please think step by step," and finally tack on a couple of polite phrases.

Back then, people really believed in this.

Knowing how to write prompts was a skill you could put on your resume, and "prompt engineer" was once the most coveted new profession on the entire internet.

Three years later, the man who invented this set of "spells" has stepped forward himself to say: stop chanting, it's useless.

One plain sentence did the work of an expert

The two screenshots Justin attached are living proof.

The screenshots came from Chainguard CEO Dan Lorenc.

This company is renowned in the field of software supply chain security.

In mid-October, Chainguard officially launched its AI vulnerability discovery system—a self-built execution framework paired with frontier large models for security research.

The prompt given to the model was simple to the point of being: find high-severity vulnerabilities in PHP that can be remotely exploited without logging in.

No role-playing, no step-by-step templates, and no "You are a senior security expert, please follow these steps..."

Just one plain sentence.

And that system really did independently reproduce known high-severity vulnerabilities, and even helped verify the official fix.

The incident caused quite a stir. The FBI issued a warning, the PHP Foundation urgently released a fix, and multiple government websites were affected for a time.

In the past, this was work that senior security experts would only dare to touch after repeatedly polishing prompts and carefully designing multi-turn conversations.

Now one plain human sentence gets it done.

What died is that sentence; what lives is that system

Not long after the buzz, Goodside himself added the second half in the comments.

He said he had just been making a witty remark.

The people writing underlying prompts for systems, the people fine-tuning small models, and the engineers embedding prompts into applications haven't seen their workload decrease at all. But for ordinary users sitting in front of a screen, the upper limit of what "just asking directly" can achieve basically no longer has a ceiling.

Someone followed up asking: which new model made you suddenly change your mind?

He said there was no sudden turning point—ever since ChatGPT came out, prompt engineering as a necessary skill for ordinary users has been slowly dying, and it's been a slow, continuous process.

One netizen put it bluntly: "Feels like the work has just shifted from prompting to the harness." Goodside replied with two words: Yes.

Exactly.

The reason Dan Lorenc's "find vulnerabilities" line worked wasn't because the sentence itself was so ingenious, but because of the complete system behind it—including the sandbox environment, the target source code, and the execution engine.

It started with Prompt Engineering, figuring out how to phrase a sentence beautifully; then Karpathy and others pushed Context Engineering, studying how to feed the model the right information; then came Harness Engineering, building a proper runtime environment for the model and setting rules and constraints; and now the community is chanting Loop Engineering, letting the whole system run in loops and iterate on itself.

Layer upon layer, the craft gets further and further from the prompt.

Ordinary users no longer need to memorize spells or save templates—the models understand plain human language.

The most interesting part is Goodside himself. Someone asked him what skills people should focus on now. This founding father answered very honestly: Honestly, I have no idea what skills are worth cultivating right now—everything is changing too fast.

Techniques get caught up to by models every few months.

Editor: Solomon

This article comes from the WeChat public account "Xinzhiyuan" (ID: AI_era), author: ASI Revelation, published by 36Kr with authorization.