The chat era is coming to an end.
I’ve basically stopped using ChatGPT and Claude’s default chat for daily work and many AI builders I talk to have done the same.
Instead, we’ve switched to Codex and Claude Code for everything.
The reason is simple. In Codex and Claude Code, I can get AI to edit Google Docs, set up cron jobs, ship features and actually get work done. In ChatGPT and Claude, I can just...chat.
Of course, there's one big problem with all this.
The average person isn't going to want to set up GitHub, understand worktrees, or learn what an API, MCP, or CLI is.
They just want to ask the AI to do things and have it work.
We're getting an early glimpse of this through OpenClaw and Hermes, agents that live in your favorite messaging apps. You can ask them to “schedule a meeting” or “draft an email reply” and they'll just figure it out. You don't have to see the tool calls and plumbing.
But these agents still require real setup and maintenance. You have to run terminal commands, get API keys, and more. It's a deal breaker for the next billion users.
So the roadmap for good ol' ChatGPT, Claude, and Gemini is pretty clear to me:
Do what OpenClaw, Hermes, Codex, and Claude Code can do while abstracting away all the technical stuff.
Evolve into a personal agent that deeply understands you and can basically do anything online instead of just a chatbot.
I'll close with this.
The conversations I have with my closest coworkers, friends, and family would be incomprehensible to any stranger:
“Hey, can you take care of that thing?”
“Can you call the person?”
And yet these people know exactly what I want because we have years of shared context built up.
That's the vision that I think every chatbot should evolve towards.
You want an AI that gets you so deeply that you don’t have to write elaborate prompts, install 40 skills, or learn what an MCP is. You can give it the laziest possible instructions and it just works.