Research Archive

Stack Comparison Matrix

Decision matrices for stack choices

Updated: December 2025 Source: research/STACK-COMPARISON.md

Stack Comparison Matrix

Quick reference for framework decisions

Current Stack vs Proposed AI-Native Stack

Layer Current (Your Projects) Proposed AI-Native Verdict
Runtime Node.js / Bun Go or Elixir Wait - Elixir is better but rewrite cost too high
Framework Next.js 15/16 SvelteKit / Phoenix Partial - Use SvelteKit for simple apps only
UI React 19 Svelte / Web Components Wait - Web Components SSR not ready
Data Various (no standard) EdgeDB / Turso Wait - Both still beta for writes
AI Protocol Direct APIs / Vercel AI SDK MCP + AG-UI Partial - AG-UI yes, MCP for dev tools only

Framework Decision Matrix

By Project Complexity

Complexity Best Choice Why
Landing page Astro Static-first, component islands
Marketing site SvelteKit Simple, fast, good DX
Dashboard SvelteKit or Next.js Depends on data complexity
E-commerce Next.js 16 Ecosystem, Vercel, proven
SaaS app Next.js 16 Full-stack, auth, db integrations
Real-time app Phoenix LiveView WebSockets native, no client state
CLI tool Bun + TypeScript Fast, simple, native TS
Agent-native Phoenix LiveView Supervision trees, fault tolerance

By Team Skill

Team Background Recommendation
React experts Stay Next.js, add AI-friendly patterns
Full-stack generalists Consider SvelteKit for new projects
Backend-heavy Go + HTMX + Templ
Functional programming Elixir + Phoenix LiveView

By AI Coding Friendliness

Framework AI Score Reasoning
HTMX + Server Templates A Minimal client JS, HTML over wire
Phoenix LiveView A No client state, typed, deterministic
SvelteKit B+ Compiler-based, simpler than React
Astro B+ Static-first, minimal JS
Next.js 16 B RSC helps, but still complex
React SPA C Hooks, effects, hydration - AI struggles
Angular C- Verbose, many patterns to learn

Component Registry Comparison

System Status AI-Friendly? Notes
shadcn/ui Production B+ Copy-paste, but React-specific
A2UI (Google) v0.8 Preview A Declarative, cross-platform, but immature
AG-UI Components Production B Event-based, React-focused
Radix Primitives Production B Unstyled, but React-only
Web Components Standard C SSR issues, interop pain

Data Layer Comparison

Solution Status Best For Avoid For
Postgres + Drizzle Production Everything Nothing - this is the safe choice
Supabase Production Auth + realtime + storage bundle Cost-sensitive at scale
Turso Preview Edge read replicas Write-heavy workloads
EdgeDB/Gel Beta Graph-relational data Brownfield, memory-constrained
SQLite (local) Production CLI tools, single-user apps Multi-user, writes
PlanetScale Production MySQL fans, branching Postgres ecosystem

AI Protocol Comparison

Protocol Status Use For Avoid For
Vercel AI SDK Production Streaming, React integration Non-Vercel deploys
AG-UI Production Agent-UI sync, human-in-loop Non-React apps
MCP Draft Auth Dev tools, internal tooling Production customer data
A2UI Preview Cross-platform native UI Production apps (wait)
Direct API Production Everything else When protocols add value

Migration Effort Matrix

From To Effort Worth It?
Next.js SvelteKit High (full rewrite) Only for simple apps
Next.js Phoenix Very High Only for agent-native products
React SPA Next.js Medium Yes, if not done already
Express API Next.js API Routes Low Yes
Prisma Drizzle Medium Maybe (Drizzle is lighter)
Direct APIs MCP Low Dev tools only
Direct APIs AG-UI Low Yes for agentic features

Quick Decision Flowchart

Is this a new project?
├─ No → Stay with current stack, add AI-friendly patterns
└─ Yes →
    ├─ Is it simple (marketing, landing, blog)?
    │   └─ Yes → SvelteKit or Astro
    ├─ Is it an agent-native product?
    │   └─ Yes → Phoenix LiveView (commit to Elixir)
    ├─ Is it a CLI tool?
    │   └─ Yes → Bun + TypeScript
    ├─ Does it need React ecosystem (libraries, hiring)?
    │   └─ Yes → Next.js 16 + component registry
    └─ Default → Next.js 16 + component registry

Your Projects Mapped

Project Current Recommended Action
rally-hq Next.js (existing) Rebuild in SvelteKit Forge proof-of-concept rebuild
six Next.js 16 Stay Add component registry
aix Next.js Stay Add component registry
bc-migration TypeScript CLI Stay Already AI-friendly
signal-forge Next.js Stay Document AI patterns
nino-chavez-website Next.js Could migrate to Astro Low priority
home-dashboard Unknown Evaluate SvelteKit If rebuilding

Note: rally-hq exists as a working Next.js application. For Forge, we are rebuilding it from scratch in SvelteKit to test the hypothesis that simpler patterns produce better AI-generated code. The existing app provides the functional requirements; the rebuild tests the development experience.

Timeline for Stack Evolution

When What Changes Action
Now Nothing mature enough Build component registries within Next.js
Q2 2025 A2UI v1.0 likely Evaluate for new mobile/Flutter projects
Q3 2025 Turso concurrent writes GA Consider for edge-heavy apps
Q4 2025 MCP auth stabilizes Expand MCP usage beyond dev tools
2026 Framework consolidation Re-evaluate full stack migration