A product manager pays $200/month for ChatGPT Pro. She uses it to draft emails. She asks it questions. She generates marketing copy. She considers it “really good AI.” She’s using OpenAI’s most powerful model GPT-6 Astra Pro as a $200/month autocomplete. Her coworker a software engineer who’s been building agentic workflows on GPT-6 Astra since the September 3 launch watched her type a prompt, wait for the response, copy-paste it into an email, and close the tab. He told her something that made her reopen ChatGPT and navigate to a screen she’d never visited. “You’re paying $200/month for GPT-6 Astra Pro. That model can drive your computer fill out forms, update spreadsheets, navigate software, and test websites by watching the screen and clicking. It can spawn multiple sub-agents that work on different parts of a task simultaneously while you watch their progress. It can create finished documents, spreadsheets, and presentations that match your templates not rough drafts, finished work. It can schedule tasks that run while you sleep. It can keep persistent notes across sessions so it remembers why your last debugging approach failed. It can complete Financial Modeling World Cup challenges 4x faster than the winning human. And it has Deep Research that reads hundreds of sources and produces analyst-grade reports in minutes. You’ve been typing ‘rewrite this email to sound more professional’ into the most capable AI model ever released and closing the tab.” He showed her 11 GPT-6 Astra Pro features that transform ChatGPT from “an AI that answers questions” into an autonomous work partner that drives computers, manages projects, researches markets, and builds deliverables while the user focuses on decisions. Here’s every hidden GPT-6 Astra Pro feature most subscribers don’t use 🧵