Work IQ Updates Memory and Context: Check Permissions and Rollout Before Enabling
Separate remembered preferences from current work context, then validate visibility and rollback with non-sensitive tests
Microsoft described new Work IQ progress on November 18, 2026, separating memory, context, and skills while saying Microsoft 365 work data continues to respect existing permissions. Because the capability rolls out in stages, this guide starts with availability rather than assuming every tenant has the same behavior. Use non-sensitive examples to test remembered preferences, current document context, and skill actions separately. Confirm that results do not cross existing access boundaries, and record how to clear memory, disconnect a source, or return to the previous workflow. A visible product label does not prove that every user, tenant, or region has received the same release. This article did not access a Microsoft account or tenant.
Key takeaways
By ENHE AI · 2026-11-20 · AI-assisted drafting; pre-publication quality checks completed.
Separate the three layers
Memory retains useful preferences, context assembles current work signals such as documents, mail, or meetings, and skills take actions. Mixing them makes failures hard to isolate.
Create one non-sensitive test per layer: change a writing preference, ask about a currently accessible document, and trigger one reversible skill. Score only the intended layer.
Permissions are not an add-on
Microsoft says Work IQ respects existing Microsoft 365 permissions. Use two test accounts with different access scopes and ask the same question rather than using an administrator as a normal user.
If a lower-privilege account receives a title, summary, or citation it should not see, stop expansion, preserve redacted evidence, and have an administrator inspect permissions and connections.
Check availability during rollout
A staged rollout means accounts in one organization can differ. Record tenant, account type, entry point, client version, and check time instead of treating a colleague screenshot as proof.
If unavailable, continue the current search or manual workflow. Do not repeatedly broaden permissions to “unlock” it; test only when the official administrative surface shows availability.
Rollback and clearing records
Document how to inspect or clear memory, disconnect data, cancel a skill, and restore human review. Mark the pilot complete only when all three rollback paths can be demonstrated.
This article did not sign in to a Microsoft 365 tenant or read real mail, files, or meetings, so it makes no account-level availability claim.
FAQ
Does Work IQ memory mean reading every file?
No. Memory and current context are separate layers, and Microsoft says existing permissions still apply.
Does seeing the Work IQ name mean everyone has the update?
No. The update is staged and should be checked per account and entry point.
Fact evidence
- Work IQ distinguishes memory, context, and skills as separate layers. — Microsoft Copilot Blog — How Work IQ helps Copilot understand work
- It uses Microsoft 365 work data while respecting existing permissions. — Microsoft Copilot Blog — How Work IQ helps Copilot understand work
- The update is rolling out rather than appearing for every tenant at once. — Microsoft Copilot Blog — How Work IQ helps Copilot understand work
- Memory retains useful preferences while context assembles current work signals for a task. — Microsoft Copilot Blog — How Work IQ helps Copilot understand work
Related tools and reading
These production pages support the same decision, workflow, or safety boundary.
Conclusion
Work IQ acceptance is not the slogan that Copilot understands more. Memory, context, and skills must each be explainable, permissioned, and reversible before use expands.
What this means for everyday users
Document how to inspect or clear memory, disconnect data, cancel a skill, and restore human review. Mark the pilot complete only when all three rollback paths can be demonstrated.
Related reading
GitHub Copilot Memory Is GA: Check and Delete Context Before Enabling Dreaming
GitHub announced GitHub intelligence on October 28, 2026. Copilot Memory is generally available on paid plans, Dreaming is an opt-in public preview, and code search can reach related repositories and dependencies. Dreaming may add, update, or remove personal and repository memories, while proposed changes to checked-in skills and instructions arrive as pull requests for team review. Dream runs consume AI credits. This article gives developers a practical way to inspect and delete saved preferences, separate personal from repository context, limit repository access, and decide whether Dreaming and cross-repository search belong in a low-risk trial before broader adoption. The trial remains reversible.
Meta Muse Launches Spaces: Pilot One Podcast or Client Workspace Before Connecting More Data
Meta introduced Spaces in the Muse app on November 6, 2026. A user can describe a custom podcast studio, client hub, order tracker, school dashboard, or trip planner, place it in the sidebar or main tabs, and connect apps so information updates in the background. This is a seven-day-window fill-in because stronger creator-facing opportunities were limited in the latest 24 hours. The article provides a minimum-workspace pilot: select one low-sensitivity use case, connect one source, list the fields Muse may read, distinguish viewing from external actions, set a refresh expectation, require human review before messages or schedule changes, and rehearse disconnection and deletion before expanding.
How Does an AI Pilot Become an Acceptable Workflow? Define Baseline, Ownership, and Rollback
Microsoft published several AI-transformation cases on November 12, 2026 and summarized a recurring pattern: start from value, redesign the process rather than one task, build in fast cycles with the customer team, and include security, governance, and cost management from day one. This durable guide converts that pattern into a one-page pilot charter. Record the current workflow and observation window, define inputs and final deliverable, name the human owner, separate assistance from final authority, set acceptance and cost limits, describe failure and rollback, and compare results over the same window. If local data is unavailable, mark it unknown rather than importing Microsoft customer figures as a forecast.
From Chat Boxes to Personal AI Companions: AI Assistants Are Entering the Desktop Execution Era
AI assistants are moving from answering questions toward continuing real tasks. AI agents, MCP tool ecosystems, personal memory, and local workbenches are pushing this shift together. For users, the real value is not another chat box, but less repeated context setup and more continuity from thinking to doing.
ChatGPT Adds an Accounting Plugin: Keep Source, Variance, and Approval Tables for Month-End Work
OpenAI announced the ChatGPT Accounting plugin on November 18, 2026. It can help reconcile transactions, draft invoice entries, and prepare month-end reports from records and templates, while Enterprise workspaces still require administrator approval before installation. This article does not provide personalized accounting or financial advice. Instead, it supplies three control tables: a source table for files, periods, and versions; a variance table for every unmatched amount; and an approval table that names who may confirm, export, or roll back a change. Bank access, payments, credentials, and final posting remain human-reviewed steps. The plugin output is a reviewable draft, not a closed accounting period or evidence that the underlying records are complete.
GitHub Copilot Evals Enters Preview: Regression-Test Skills, Agents, and MCP Changes
GitHub launched Copilot Evals in public preview on November 17, 2026. It creates repeatable tests for skills, custom agents, and MCP tools and lets users compare runs before and after a customization change. This guide defines a minimum regression set: choose three realistic tasks, then specify the input, allowed tools, success criteria, and prohibited behavior for each. Preserve at least one baseline run and one post-change run, including the result, execution path, failed checks, and human review. Merge a customization only when the change preserves the original task, permission boundary, and evidence chain. A single passing answer is not enough because model output can vary, and an automated check cannot decide every domain-specific quality or safety question.
