Developers can now submit apps to ChatGPT

Original Article ↗ Hacker News Discussion ↗

App Store Redux & Strategic Positioning

  • Many see this as the revival of the short‑lived GPT Store, now reframed as “Apps” integrated into ChatGPT conversations.
  • Some argue this signals that GPT alone won’t be an “everything machine” soon; instead OpenAI is moving toward a platform/ecosystem play.
  • Comparisons are made to earlier platform waves (mobile app stores, browser toolbars, boxed software, SMS downloads), with predictions of a similar gold-rush → consolidation → enclosure cycle.

Value Proposition for Companies & Developers

  • Some expect strong B2C adoption driven by FOMO: companies don’t care about intermediaries as long as they reach users where they are.
  • Others think many brands will resist losing UI/control, especially those that tightly manage customer experience.
  • Several commenters question why a developer should build here: unclear monetization, risk of low usage, and fear that successful ideas will be cloned or “absorbed” by the platform.

Monetization & Distribution Concerns

  • Currently, apps can only link out for transactions; digital goods and deeper monetization are “exploratory,” which some label vaporware.
  • A recurring concern: “free labor” for OpenAI, plus the risk of OpenAI or others copying the best apps once product–market fit is demonstrated.
  • Others counter that distribution is always monetized; the real problem is dominant platforms controlling both access and distribution.

Technical Architecture & UI Direction

  • Apps are essentially MCP servers plus custom UI (React or vanilla JS) rendered in frames; porting an existing app is non-trivial.
  • Some dislike pushing full webstack (HTML/CSS/JS) into MCP, preferring native UIs, but acknowledge this is probably a losing battle.
  • There’s speculation (and some references to Google’s work) that new agent-driven UI frameworks will emerge, with reusable primitives (cards, carousels, tables) tailored for chat contexts.

User Auth, Tokens & “Bring Your Own Model”

  • Strong interest in “Log in with OpenAI/Gemini/Anthropic” so user quotas fund usage, avoiding developers eating all token costs.
  • Existing partial analogs (e.g., Google AI Studio sharing, MCP + OAuth) are seen as too clunky or limited; most users won’t manage API keys.

Platform Power, Walled Gardens & Cannibalization

  • Many fear another walled garden: closed discovery, opaque featuring, and eventual cannibalization of successful vertical apps.
  • Some predict that as LLMs subsume more UI and workflow, many SaaS products will be reduced to commoditized tool/API calls behind the AI front-end.

UX, Reliability & Safety Frictions

  • Early experiences (e.g., GitHub app) show confusing permission flows and brittle behavior; screenshots sometimes “unstick” refusals due to internal routing/verification quirks.
  • Questions are raised about execution environments (security, cryptojacking), prompt exfiltration (mostly seen as unavoidable), and stringent identity verification for app publishers.

Broader Skepticism & Societal Impact

  • Several commenters doubt OpenAI’s focus, seeing “MBA/VC playbook” platform moves instead of clear model improvements.
  • Others worry about long-term skill atrophy (coding, writing, critical thinking) as more interaction is offloaded to AI-mediated apps.