AI agentsNews analysis · 2 min read
GitHub adds persistent memory and opt-in Copilot dreaming
GitHub’s new intelligence layer aims to stop coding agents rediscovering the same project knowledge in every session. The update combines persistent memory, cross-repository search and an optional process for refining context from past work.
What changed
The October 28 release makes Copilot Memory generally available on paid plans and introduces dreaming in public preview. Dreaming can curate memories and propose changes to checked-in skills or instruction files through pull requests. It is opt-in and uses AI credits. The post’s preview-session statistics are GitHub’s own observations from a selected set of sessions, not a universal prediction of developer productivity. Source: GitHub intelligence puts code, context and history to work
Why it matters
Remembering a convention can save time, but remembering an incorrect assumption can repeatedly steer future work in the wrong direction. Our assessment is that inspectability and correction are central to useful agent memory. Teams should be able to identify which guidance influenced a change and distinguish personal preferences from repository requirements. A learned habit should not silently become stronger than an explicit current instruction.
What to watch next
Check the view-and-delete controls, repository access settings and review path for proposed guidance changes. Evaluate whether retained context remains accurate after a migration or team-policy change. The next useful evidence is better accepted work over time without stale instructions accumulating. Memory is a maintained information system, and its usefulness depends on curation rather than simply retaining more text.
Sources and editorial note
- GitHub intelligence puts code, context and history to work
Original source · 28 October 2026. Accessed 3 November 2026.
AI-assisted reporting and analysis based on the linked primary source. Company results are attributed; we have not independently tested the systems. The context and outlook sections are our analysis. Our editorial standards.