Glaze by Raycast

Original Article ↗ Hacker News Discussion ↗

Security, Trust, and Permissions

  • Many are uneasy installing AI-generated desktop apps with broad system access, unlike web apps in a browser sandbox.
  • Glaze will reportedly have a permission model (per-file/directory, network domains, etc.), but several argue this doesn’t remove the core risk of arbitrary code execution.
  • Concerns include: no clear security story on the landing page, potential for “trust-me” curl2bash-like behavior, and non-technical users running opaque binaries with sensitive permissions.
  • Some expect this to be a “security nightmare” and question how an in-app store of unsigned apps fits with OS security policies.

Tech Stack and “Native vs Electron” Debate

  • Speculation that Glaze apps are likely Electron or a webview wrapper; others wonder about Tauri, Wails, SwiftUI, or a Raycast-style runtime that maps React components to native widgets.
  • Several wish strongly for truly native apps (SwiftUI, WinUI, GTK) and see more JavaScript-based desktops as “slop.”
  • The actual implementation stack is not clearly disclosed; multiple commenters explicitly call this “unclear.”

Relationship to Claude Code and Other AI Builders

  • Many see Glaze as “Claude Code with extra tooling”: packaging, signing, distribution, and opinionated design prompts.
  • Some argue anyone comfortable with Claude can already build better, fully-owned native apps without vendor lock-in.
  • Others counter that Glaze targets less-technical users and teams who value easy publishing, sharing, and sensible UI defaults over raw flexibility.

Use Cases and Value Proposition

  • Enthusiasts mention “vibe-coding” personal utilities: menu bar tools, system monitors, text utilities, and niche business workflows.
  • A key selling point is integrated distribution: a built-in store for private/public sharing across teams, avoiding App Store friction.
  • Examples mentioned as built with Glaze include internal support tools, a MIDI-capable synthesizer, and a replacement for an existing team SaaS.

Product Direction and AI Fatigue

  • Long-time Raycast users are split: some see this as a natural extension of workflow automation; others fear a distracting pivot driven by VC expectations.
  • Skeptics predict many such AI app builders won’t move beyond polished demos, given iteration cost and complexity.
  • Several note a broader saturation of “AI app builders” and question whether there are more platforms than real apps.

Design Quality and Agentic UI Capabilities

  • Some report that recent AI models now generate surprisingly good, even novel UIs, versus crude results a year ago.
  • Others remain doubtful that agents can reliably create complex, dynamic interfaces without heavy human iteration.
  • There is interest in tooling that would let agents iteratively “see” and refine UIs, but concrete mechanisms remain unclear.