Context debt
The growing difficulty of knowing what a persistent AI agent currently believes, where that belief came from, and how to correct it, as the agent accumulates memory over time.
Context debt appears when an AI agent remembers more than anyone can easily inspect. The better a persistent agent gets at carrying memory across days and applications, the more important it becomes to know what it currently believes, whether that belief is an observation or an inference, how long it should stay valid and how simply a person can correct it.
Traditional systems keep consequential state in legible places: a database field, a ticket, a document with an owner. Persistent AI moves that state into conversation history, summaries, memories and tool traces. These are much harder to audit. Useful context compounds, and so do stale assumptions, and the agent keeps acting on both with equal confidence.
Why it matters
A user who has to re-read everything the agent "knows" before trusting it has not been relieved of any work. The promise of persistent delegation is less coordination in the user's head, not a new system to manage. Context debt is what quietly cancels that promise.
What reduces it
- Show the agent's current beliefs and their sources.
- Separate observed facts from inferred ones.
- Give beliefs an expiry and make correction cheap.
Read more in OpenAI Dots is a test of whether AI can carry a goal, not just complete a task.
Related terms
Persistent delegation
Handing an AI agent a goal it keeps carrying over time — remembering context, noticing relevant events and acting across applications — rather than completing a single prompted task.
Capability is not authority
A design principle for AI agents: being technically able to perform an action does not mean the agent should be permitted to perform it. Delegation needs gradients of authority.