Overview
- Karpathy says he has never felt this far behind as a programmer and warns the profession is being dramatically refactored by AI.
- He argues developers can become roughly 10x more effective by mastering a new layer that includes agents, prompts, tools, plugins, workflows, and IDE integrations.
- He describes modern models as stochastic, fallible, unintelligible, and changing, urging stronger mental models and oversight of their outputs.
- Coverage highlights a workflow shift from writing code to instructing systems, reviewing results, and correcting errors rather than relying on deterministic software.
- Context from industry and research remains mixed, with Google citing over 30% of new code written by AI and Anthropic claiming far higher internal use, while studies report uneven productivity gains.