<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Danny C. FSGen</title>
    <description>The latest articles on DEV Community by Danny C. FSGen (@dannyc_fsgen).</description>
    <link>https://dev.to/dannyc_fsgen</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4021235%2F476f58af-2cd0-446c-9f35-d94c8f4ed0b5.png</url>
      <title>DEV Community: Danny C. FSGen</title>
      <link>https://dev.to/dannyc_fsgen</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/dannyc_fsgen"/>
    <language>en</language>
    <item>
      <title>I asked ChatGPT to write my project brief</title>
      <dc:creator>Danny C. FSGen</dc:creator>
      <pubDate>Wed, 08 Jul 2026 11:43:48 +0000</pubDate>
      <link>https://dev.to/dannyc_fsgen/i-asked-chatgpt-to-write-my-project-brief-22ij</link>
      <guid>https://dev.to/dannyc_fsgen/i-asked-chatgpt-to-write-my-project-brief-22ij</guid>
      <description>&lt;p&gt;Everyone warns that AI over-engineers. I expected a spec bloated with 2FA, microservices, and a Kubernetes cluster for a CRUD app. What I actually got was worse, because it was reasonable.&lt;/p&gt;

&lt;p&gt;I gave ChatGPT one sentence describing a property-inspection app and got back a clean, professional brief. Then I read it like an engineer scoping the work, and found three failure modes worth naming:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;&lt;p&gt;The existential risk, buried in a subordinate clause. The app's core feature depends on a public land registry — and those are famously fragmented, often with no public API. The brief handled this make-or-break unknown with "integration with registry APIs or data sources where available." "Where available" is doing catastrophic amounts of load-bearing work. A generator optimizes for a document that looks complete; "this might be impossible, verify first" doesn't fill a section nicely, so it gets compressed and hidden.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;A silent scope decision. It specced "document capture with automatic image organization" — deciding, without telling me, that documents are photos to file, not text to extract. But comparing a deed against the registry means pulling the parcel number off the image. That's OCR, non-trivial, and it was assumed away in the gap between "photograph" and "organize."&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;A state machine hiding inside two verbs. Inspections get "submitted" and "flagged." Innocent words that smuggle in a question nobody asked: is there an approval workflow? A supervisor role? If so, every inspection now has states, transitions, and per-state permissions — the difference between a form that saves data and a stateful workflow engine.&lt;/p&gt;&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The root cause is the same each time: it never asks, it fills. A brief's job is to force decisions; a generator's job is to look finished. Those are opposite goals.&lt;/p&gt;

&lt;p&gt;Full write-up (with the actual ChatGPT output): &lt;a href="https://fsgen.com/blog/chatgpt-brief-teardown" rel="noopener noreferrer"&gt;https://fsgen.com/blog/chatgpt-brief-teardown&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Originally published on fsgen.com.&lt;/p&gt;

</description>
      <category>programming</category>
      <category>startup</category>
      <category>webdev</category>
      <category>ai</category>
    </item>
  </channel>
</rss>
