Tested @Cloudflare Clef as a judge on string translation disagreements: it flipped its answer when I swapped the A/B order 45%; and where it was consistent, it matched only 54% the actual best answer. Fast and cheap, but not good translation quality.
I understand the follow-up to the failed live demo but I don't understand when you're sat in front of a laptop and an iPhone, you STILL choose to *telephone call* an agent. Even if I was to trust it book the correct flight (it didn't in the demo)!
First day with the Note4C from @itopinion and already have syncing, sleeping, OTA updates, one-click pairing. Promising start! 🤩🚀
This is where x402 payment per call are entering the chat. The market cannot bear “one more subscription”, it’s race to the bottom era of collecting $0.001 for a data lookup
I *love* an all black high density design. I believe two contradictions. AI can enable us to make really nice UIs 'out of nothing' with minimal effort. BUT, I also think we're ~3 years away from UI being obsolete, simply interacting with APIs in chats with UI rendered on demand.
Opus 5.5 built this CRM from one prompt (delegating to 6 subagents).
how:
- the main agent wrote the brief: product, fictional data, design system
- 1 subagent built the foundation, 5 built the rest in two parallel waves
- Mobbin MCP: 57 searches for real app patterns
- 11 design skills (frontend-design, emil-design-eng, impeccable, apple-design…)
- screenshot review loops on every page, dark and light
- one review from me; the agents fixed what I flagged
- ~4 hours in Claude Code
find links below 👇
