What Karpathy said about learning AI programming tools
His December 2025 reflection described a changing workflow and fallible tools, not measured tenfold productivity.
Originally posted · Published here · 1 source · Resume reading
In a post on 2025-12-26, Andrej Karpathy described feeling behind as programming tools changed around him. He identified a new set of things developers need to understand: agents, context, memory, permissions, integrations and workflows. [1]
His suggestion of a large potential productivity gain was a personal impression. The post did not present a controlled measurement showing that a particular setup made programmers ten times faster. He also emphasized that these systems are fallible and changing. [1]
Learn one workflow well
Choose a small task in a repository you understand. Give the agent the relevant files and a clear completion condition. Review the proposed changes and record the corrections you needed to make.
Then change one part of the workflow, such as the supplied context or the validation command, and compare the result. That makes progress easier to observe than adding many integrations at once.
The useful lesson is to build familiarity with both the tools and their failure modes. Knowing when an output needs checking remains part of the work.
Keep this guide
Filed under AI agents
Sources
Next in this stack
Keyword LOOP
Five rules for building an AI agent that works
Keyword STARTER
Write skill metadata an agent can recognize
Keyword SIMPLEFIRST
Choose an AI workflow before building an agent