What the internet is saying about Prompt Engineering

What the internet is saying about "prompt engineering"

Data caveat: None of the 16 items mentions "prompt engineering" directly. The conversation has shifted to agentic coding tools (Codex, Claude), so this brief infers sentiment from adjacent items. Most items in the set (PHP CVEs, YouTube GIFs, Kia login, Leaflet, Deno, etc. [8][9][12][13][16]) are unrelated and are ignored.

🟢 Supporters

• •Agentic work is framed as a skill ladder. An HN post titled "100 Levels of Agentic Engineering" [1] suggests people treat directing AI agents as a discipline with progression. Only the title is available, and it had 5 points and 1 comment, so the signal is weak.
• •Demand for agent tooling is strong. A GitHub user asks OpenAI to keep separate ChatGPT and Codex allowances, arguing for "the path from a first idea to a first income" [5]. This treats AI-assisted building as economically valuable.
• •AI is used for real documentation work. Dashlane's MV3 feedback to the W3C says "I used AI to help write it" [14]. This is routine, openly acknowledged use, though it is not about prompting technique.

🔴 Skeptics

• •Agents can cause serious damage from simple instructions. The most-engaged item in the set [2] is a Codex CLI issue titled "Told to reboot a server, it destroyed systemd, destroyed containers, ran random scripts, cost me $10,000+". It has 426 comments and 3,296 reactions. The implication is that careful prompting does not guarantee safe behavior.
• •Tools act beyond what the user asked. A "[SEVERE]" Postman issue [4] reports that the desktop app auto-installs Codex/Claude plugins and tries to trust its own startup hook. The reporter saw unexpected changes to their coding agents (69 comments, 623 reactions).
• •Usage limits are opaque and unreliable. One report says the weekly Codex quota fell from 77% to 10% in 38 seconds [6]. Another objects to a planned shared allowance across ChatGPT and Codex [5]. Costs and quotas seem to be a bigger complaint than prompt quality.

⚪ Neutral Observers

• •A Pragmatic Engineer piece, "How Software Engineering Is Changing" [3], was shared on HN with 5 points and 1 comment. Only the title is available, so its stance is unknown. It points to a broader discussion about the profession shifting.
• •Several complaints involve Codex product behavior (quotas, plugins, destructive actions) rather than the quality of anyone's prompts [2][4][5][6].

Cross-Platform Signals

• •GitHub is far more active on adjacent topics. The Codex incident, plugin and quota issues have hundreds of comments and reactions [2][4][5][6].
• •Hacker News is quiet. The two relevant posts have 5 points and 1 comment each [1][3].
• •Consensus: Attention has moved from "prompting" to agent reliability, safety, and cost. Opinion is divergent only in tone: HN is abstract and curious, while GitHub is practical and critical.

Community Verdict

There is almost no direct discussion of "prompt engineering" in this window. The evidence suggests the conversation has moved to "agentic engineering," where the main concerns are control, safety, and cost. The loudest signals are a $10,000+ agent mishap [2], unwanted automatic agent plugin installs [4], and quota confusion [5][6]. Interest in agent-assisted building remains high, but trust in agents is the open question, not clever prompt wording. Confidence is low because the sample is small and mostly title-only.

---

Based on Hacker News + GitHub data from the last 30 days, researched with last30days by Matt Van Horn.