Agentic AI coding tools are quietly going mainstream
The thread on ChatGPT Work/Codex is generating real excitement, with users describing how they remote into a browser session from their phone, voice instructions to ChatGPT, and let it work in the background. One commenter notes most people are 'sleeping on' this feature. Separately, the thread on Claude Code re-igniting a 60-year-old's passion for programming, and another on whether local models can replace Claude/GPT for daily coding, all point to the same moment: AI coding assistants have crossed from autocomplete into genuine autonomous execution.
The pattern here is speed of adoption, not capability announcements. Nobody is waiting for a press release. Builders are just using these tools and reporting back in real time. The ChatGPT Work thread specifically notes that Work sessions bill against Codex allowance separately from chat sessions, which matters for anyone trying to manage costs at scale.
The counterpoint shows up in the local model thread, where people running DeepSeek V4 Flash on dual RTX Pro 6000 Blackwells or Qwen 3.6 27B dense are getting results 'about equal to Claude Haiku 4.5.' The gap between cloud and local is closing, which means the agentic layer is becoming a commodity battleground faster than the model providers probably want.
So what?
If you are building developer tools or infra, the agentic layer is not coming, it is here. The question is whether your product fits into a workflow where an AI agent is the primary actor, not a human. Founders should be stress-testing their APIs and CLIs against agent-driven usage patterns right now, because that is already how a meaningful slice of your users are interacting with you.
Read these
Understanding ChatGPT Work
AI-Written Code Is Still *Your* Code. Are You OK with That?
Tell HN: I'm 60 years old. Claude Code has re-ignited a passion
Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?