Karpathy admits he's never felt so behind as a coder
Andrej Karpathy just confessed that he has never felt this far behind as a programmer, warning that AI is driving a "magnitude 9 earthquake" across the craft. Chaining together the latest agents, contexts, and MCP workflows could easily make a developer 10x more powerful, but it feels like operating an alien tool handed down without a manual.
If even Andrej Karpathy feels like he is falling behind in programming, the rest of us might not need to beat ourselves up so much.
The former Tesla AI director and OpenAI co-founder shared a candid confession on X: as a programmer, he has never felt this far behind.
Karpathy noted that the profession is undergoing a dramatic refactoring. The actual bits written by human coders are becoming increasingly sparse. He feels he could easily be 10 times more productive if he simply chained together everything that came out over the past year, adding that failing to unlock that boost feels decidedly like a personal skill issue.
An alien tool without a manual
According to Karpathy, mastering programming now requires tackling an entirely new programmable layer of abstraction stacked on top of traditional engineering.
That list is dizzying: agents, subagents, prompts, contexts, memory, tools, MCP, LSP, slash commands, workflows, and deep IDE integrations. Engineers now have to build mental models for entities that are fundamentally stochastic, fallible, and unintelligible, all while weaving them into deterministic code.
He compared the situation to an alien tool dropped into human hands without an instruction manual, leaving everyone scrambling to figure out how to operate it while a magnitude 9 earthquake shakes the industry.
Why it matters
AI is not eliminating the need for programming skills; it is shifting the job toward orchestrating chaotic systems.
For developers, the edge is no longer about writing boilerplate code line by line. It is about learning to wrangle agents, configure protocol bridges like MCP, and manage non-deterministic systems. Since industry leaders are figuring this out in real time, there has never been a better moment to dive in and experiment.
That said, building mission-critical software on top of stochastic components comes with serious architectural risks. We think chasing that theoretical 10X boost without maintaining rigorous traditional testing and guardrails could easily trade upfront dev speed for a maintenance nightmare down the road.
Sources (1)
- @karpathy(opens in a new tab)Originalx.com