AI agent glossary
Knowledge cutoff
The point in time after which a model's training data ends, so it knows nothing about later events unless they are supplied in context.
A language model learns from a large snapshot of text collected up to some date. Anything that happened afterward, such as new library versions, product launches, price changes, or news, is invisible to it unless you provide it.
Models do not always know where their own knowledge ends, and they may confidently describe an outdated API or assume an old version is the latest. This is a common source of hallucination in coding agents working with fast-moving libraries.
The fix is to put current information in context: give the agent web search, documentation lookup tools, or MCP servers that expose up-to-date docs, and tell it the current date. Model providers publish each model's training cutoff in their documentation.
An agent writes code using a framework option that was renamed in a release after its cutoff. Pointing it at the current migration guide fixes the error on the next attempt.
Learn more in the course
Related terms
Go from definitions to doing
Learn to use and build AI agents in 65 practical lessons.