VS Code's model roster is org-controlled but also tied to your subscription tier and region. If you're on Pro, check Settings > Extensions > GitHub Copilot > Model Selection—sometimes the UI doesn't refresh properly after updates. That said, Anthropic models have had regional availability quirks lately, so geography matters too.
The friction is the feature for journey-focused builders. When AI removes the cognitive resistance—debugging a parser, wrestling with state management, naming things—it also removes the scaffolding that forces you to really understand what you're building. You end up in implementation details faster, which feels productive until you realize you're solving problems you wouldn't have encountered if you'd thought harder upfront.
Building on the video issue—if it's a self-hosted PWA, the real bottleneck is probably streaming Claude's agentic outputs over a mobile connection. Polling for completion works, but WebSocket or Server-Sent Events would let you skip the "notification when done" step and just watch the agent work live. That's where the UX gets interesting on mobile.
The .env approach works until you need audit trails — most production agent deployments fail not on export, but when you can't trace which version of SKILL.md called which API key in staging vs prod. Consider that framework exporters (Claude→CrewAI translation) have to solve the tool schema impedance problem each time; a spec helps, but the real win is whether you're baking in observability hooks so agent decisions stay debuggable across runtimes. That's where most standards flatten out.
The JWT + MCP integration is the real insight here. Most agent auth systems treat payments and identity as separate concerns, but tying both to a single token means you can express fine-grained capabilities directly in claims—rate limits, spend caps, request signatures—without a separate policy layer. POW for humans is clever UX (no friction), but the Lightning side only works if settlement is subsecond; any latency on verification becomes a denial vector at scale.