Can A.I. Already Do Your Job?

What It Covers

Kevin Roose demonstrates agentic coding tools like Anthropic's Claude Code that allow non-programmers to build functional software through AI agents. These systems represent a major advancement from ChatGPT-era AI, with Stanford data showing 20% decline in entry-level software engineering employment since 2022. Anthropic CEO predicts potential elimination of half of entry-level white collar jobs within five years.

Key Questions Answered

• •Agentic Coding Evolution: Claude Code and OpenAI's Codex enable autonomous software development where AI creates implementation plans, selects programming languages, and deploys specialized sub-agents for research, building, and testing. Users provide project concepts while agents execute multi-hour tasks independently, writing hundreds of code lines in under two minutes without human programming knowledge required.
• •Self-Improving AI Systems: OpenAI's GPT 5.3 Codex uses earlier model versions to train subsequent iterations, creating recursive self-improvement loops across major AI companies. This acceleration pattern moves from clunky vibe coding tools one year ago to autonomous agents capable of maintaining production software today, with AI engineers reporting they no longer write code manually but orchestrate agent teams.
• •Entry-Level Job Displacement: Stanford payroll data reveals 20% employment drop for early-career software engineers from 2022 peak levels. Companies previously hiring five to ten developers now operate with one or two humans managing AI coding tools. Anthropic CEO Dario Amodei warns this pattern could extend to 50% of entry-level white collar positions across industries within five years.
• •Practical Deployment Speed: Anthropic employees adopted Claude Code organically, starting with 20% of engineers, expanding to 40%, then achieving full technical staff adoption before spreading to marketing, sales, and finance departments. Non-technical workers now automate email management, create data dashboards, and reorganize computer files through terminal-based AI agents previously accessible only to programmers.
• •Verification Advantage in Coding: Software development provides ideal testing ground for AI capabilities because code functionality is binary—programs either execute correctly or fail. This verifiability enables rapid improvement cycles as models train on expanding coding datasets, with systems now producing deployable business software that required human debugging just months earlier, though enterprise-scale deployment still requires oversight.