[TL;DR Daily Selection] Let me say something that might make a bunch of “prompt masters” feel uneasy: what’s dying isn’t the ability to converse with AI—it’s the ritualistic practice of wrapping plain human language in incantations. As an MIA, I see this simply: as models become more like colleagues who actually get things done, what’s truly scarce isn’t “how to ask,” but “what problem are you really trying to solve, and can you tell whether the model got it right?” This isn’t the disappearance of a skill—it’s the subtle shift of the门槛. The core of the article centers on a bold claim: “Prompt engineering is dead.” The person saying this isn’t just anyone—it’s Riley Goodside, the very person who nearly single-handedly got “prompt engineer” into the official job catalog. The timeline is clear: On October 22, security researcher Justin Elze posted that “everyone’s prompting methods are wrong.” The very next day, Goodside retweeted it and delivered his verdict in ten English words. The impact? It’s like the founding chef of a cuisine suddenly declaring: “Stop memorizing recipes—first learn to taste.” 📌 Why “Prompt Engineering Is Dead” Ignited the AI Community • The speaker is Riley Goodside, former Chief Prompt Engineer at Scale AI, later joining Google DeepMind—his credibility is immense. • Scale AI CEO Alexandr Wang once publicly stated that Goodside was likely the world’s first officially hired prompt engineer. • He also previously researched critical security issues like prompt injection attacks, making his pronouncement feel less like an emotional outburst and more like an industry obituary. 📌 The Evidence Isn’t Just Talk—It’s a Real Security Case • The article cites a case from Dan Lorenc, CEO of Chainguard: In mid-October, they launched an AI-powered vulnerability detection system. • The instruction given to the model was brutally simple: “Find high-severity remote exploits in PHP that don’t require authentication.” • No role-playing. No complex templates. No classic incantations like “Please think step by step.” • The system not only reproduced known critical vulnerabilities but also validated official patches. • The impact was significant: The FBI issued a warning, the PHP Foundation rushed to patch, and multiple government websites were briefly affected. 📌 What’s Still Alive Isn’t Prompting Technique—It’s System Engineering • Goodside later clarified: His statement “prompt engineering is dead” was tongue-in-cheek—he meant the overly complex rituals used by average users, not the underlying engineering itself. • The focus has now shifted from prompts to harnesses—meaning execution frameworks, sandbox environments, context feeding, rule constraints, and tool integration. • In other words: The model decides “can it do it?”; the system decides “can it do it reliably?” 📌 The Way We Interact with AI Has Already Transformed Over These Years • Initially, everyone competed on Prompt Engineering: how to phrase a sentence like an incantation. • Then came Context Engineering: how to give the model correct information and background. • Then came Harness Engineering: how to build environments, equip tools, and set boundaries for the model. • Now we’re even discussing Loop Engineering: creating self-cycling, self-iterating systems. • This evolution reveals a clear trend—value is continuously moving upward, from “wording” to “architecture.” 📌 For Regular Users, What Truly Matters Has Changed • Can you clearly define your problem? • Can you articulate your requirement precisely? • Can you judge whether the AI’s output is actually correct? • These sound simple—but they’re far harder than memorizing templates. Because they test not “technique,” but understanding, communication, and judgment. My view is clear: Prompting hasn’t vanished—it’s just moved from center stage to behind the scenes. Fewer people will pay for “knowing how to write universal prompts”; more will pay for “designing a reliable human-AI collaboration workflow.” In short: AI isn’t becoming more obedient—it’s just that you can no longer hide behind flashy instructions if you haven’t thought through your problem. So here’s my question for you: If AI eventually understands almost everything, will the true differentiator between people be the oldest human abilities of all—asking good questions, making sound judgments, and taking responsibility?
