Smart Zone
The window size where the model is still sharp.
The portion of the context budget within which a model reasons reliably. Operating inside it gives consistent quality; pushing past it trades accuracy for volume.
Treat context like a budget with a quality curve, not a container with a fill line. Staying at thirty percent utilisation frequently outperforms filling to ninety.
The useful operational move is to find your own smart zone empirically. Run the same task at several context sizes and watch where accuracy falls off. That threshold becomes a budget your context manager enforces, which is more useful than any model’s advertised window.
Measuring where quality drops
Deciding how much history to keep in a long-running agent.
Fill the context with progressively more realistic work and score accuracy at each level. Most tasks show a clear knee. Compact when you approach it, rather than waiting for the window to fill and the model to degrade.
- blogEffective context engineering for AI agents ↗
Defines context engineering as curating everything supplied at inference, names context rot, and covers just-in-time retrieval, compaction and keeping the window high-signal.
- paperLost in the middle: how language models use long contexts ↗
Measured U-shaped performance: relevant information in the middle of a long context is used less reliably than at the beginning or end. Order your context deliberately.
- docsContext windows ↗
What actually consumes the window — input, tools, thinking and output — plus context rot and compaction. Authoritative for this product, not for models generally.
Attention Budget
The finite share of the model’s focus each token gets.
Context Rot
A long session slowly getting worse.
Context Window
The hard ceiling on how much the model can see at once.
Compaction · Summarisation
Replacing old context with a shorter summary of it.