1
Prompt hacks are obsolete — AI models like GPT image 2 and Claude now respond to plain English descriptions, so the only lasting skill is learning to describe what you want clearly
2
Skills (task-specific instruction files) are the core unit of agent productivity; you can create and auto-improve them by simply asking your agent, and tools like the Hermes agent will soon self-assemble them
3
Codex and Claude Desktop are emerging as all-in-one super apps covering chat, vibe coding, automations, hosted sites, and in-app browsers — learning one deeply is more valuable than sampling many tools
4
Cloud-based agents (e.g. via Chorus) now live inside iMessage, Slack, WhatsApp, and Telegram, so agents are accessible through tools billions already use without keeping your computer open
5
Frontier models (e.g. Grok/Fable 5 costing ~$250 for 9 prompts) are getting more expensive, while open-source models like GLM 5.2 are nearly matching Claude Opus 4.8 at a fraction of the cost — OpenRouter lets you access both with one API key