Windows Copilot Vision Text Edits In Real Time During Sessions

  • Thread Author
Microsoft has started rolling out a targeted Insider preview that lets Copilot edit text in-place during a Copilot Vision session — a real‑time “rewrite, refine, edit” mode that inserts suggested changes directly into whatever document or text field you’re sharing, with a preview step before any changes are applied.

Background​

Copilot on Windows has evolved from a basic chat assistant into a layered, system‑level productivity service that combines Voice, Vision, and Actions. The December 19 announcement expands that vision by enabling Copilot to perform text editing actions while it can “see” a shared app window or document. This builds on earlier Copilot updates that introduced Vision as a permissioned screen‑sharing analysis tool and Copilot Actions as an experimental agent framework that can run contained tasks on local files. These additions follow a pattern Microsoft has used in recent months: ship capabilities to Windows Insiders via Microsoft Store updates, gate features server‑side, collect feedback, and iterate before broader availability. That staged, opt‑in approach has been used for Vision, Copilot Actions, and other multimodal features introduced earlier in the year.

What’s new: Copilot Actions for Text Editing (what Microsoft says)​

  • Copilot can now rewrite, refine, or edit text in real time during a Copilot Vision session. To trigger it you begin a Vision session from the Copilot app (click the glasses icon in the composer), share the window that contains the document or text field, place your cursor in the text field, and ask Copilot to perform an editing task in natural language (for example: “rewrite this to be more formal”, “make this clearer”, “simplify this text”). You’ll see Copilot’s suggested edit as a preview before changes are applied. Copilot Vision remains fully opt‑in and activates only when a user chooses to turn it on.
  • The feature requires the Copilot Actions toggle to be enabled in the Copilot app Settings. Users who turn on that toggle will gain access to both the text editing actions and Copilot Actions for longer running/agentic tasks (the latter is surfaced through Copilot Labs).
  • Package and OS requirements: the update is distributed via the Microsoft Store as Copilot app version 1.25121.60.0 (and higher) and requires Windows version 26200.6899 (and higher). Availability is staged and currently excludes the European Economic Area (EEA).
These are Microsoft’s headline claims as published in the Windows Insider Blog. The post is the primary record for the precise version numbers and gating details.

How the feature works — practical, hands‑on flow​

Starting a text editing Vision session​

  • Open the Copilot app (or Copilot composer via the taskbar/quick view).
  • Click the glasses (Vision) icon in the bottom right of the chat bar to begin a Vision session.
  • Choose the window or region you want to share — Copilot will analyze what’s on that screen.
  • Place the cursor in a text field in the shared window, click into the field, and speak naturally (or type): e.g., “rewrite this to be more formal” or “make this shorter”.
  • Copilot will analyze the visible content, generate an in‑flow editing suggestion, and show a preview before applying changes; accept, refine, or reject the suggestion.
This interaction is explicitly session‑bound and permissioned: Copilot only sees what you select to share, and you stop sharing when you press Stop/X. That UX decision — visible glow around shared windows and clear stop controls — is consistent with Microsoft’s privacy framing for Vision.

Typical supported use cases​

  • Polishing email drafts, messages or form entries without leaving the original app.
  • Rewriting or simplifying text in web editors, Word/Docs, chat windows, and note apps.
  • In‑place localization or tone‑adjustment (formal ↔ casual).
  • Quick clarity passes for copy, commit messages, and short documentation snippets.
These are the scenarios Microsoft highlights; real‑world reliability will depend on the preview’s OCR accuracy, app compatibility, and the underlying model behavior.

Technical and rollout details (verified)​

  • Copilot app: the new text editing actions ship with Copilot app version 1.25121.60.0 and higher distributed via the Microsoft Store. Confirming the package version in the app’s About page or Microsoft Store history is the reliable way to verify whether the package reached a device.
  • Windows build: Insiders must be on Windows version 26200.6899 or later to use this specific text editing capability. This aligns with Microsoft’s practice of tying some Copilot features to specific Windows Insider builds and cumulative updates.
  • Opt‑in and region gating: the feature is opt‑in (Vision must be enabled per session) and is gradually rolling out worldwide, excluding the EEA. That regional exclusion mirrors earlier Copilot Actions rollouts where Microsoft staged availability and excluded the EEA initially.
  • Precedent: Copilot Actions (the agent runtime and Agent Workspace) began rolling out in November as an experimental Copilot Labs feature; the December text editing addition is an incremental, content‑focused extension of that capabilities set. This evolution — from agentic file operations to text editing during Vision sessions — reflects Microsoft’s staged feature layering.

Cross‑checking Microsoft’s claims (independent context and verification)​

The Windows Insider Blog entry is the canonical source for the new text editing claims; independent reporting and documentation confirm Microsoft’s broader Copilot strategy and the staged rollout approach:
  • Earlier Copilot Actions rollouts were documented by Microsoft and reported by independent outlets as an experimental, opt‑in agent workspace feature for Insiders, which required a Copilot app package update and explicit enabling in Settings. That earlier announcement aligns with the requirement here to enable a Copilot Actions toggle before text editing becomes available.
  • Microsoft’s ongoing expansion of editorial features across its ecosystem — for example, interactive rewriting and in‑app suggestions in Word and Microsoft 365 — shows the company has been progressively integrating rewriting workflows into its products. Those existing controls in Word mirror the editing affordances Microsoft is now bringing to Copilot Vision across arbitrary apps.
  • Independent assessments of Copilot’s Vision and Actions capabilities have repeatedly emphasized the usefulness of permissioned, session‑bound sharing and the staged rollout model — along with persistent concerns about accuracy and governance. That context is useful for evaluating both the potential and the risks of in‑flow text editing.
Where independent coverage of the December 19 text editing feature does not yet exist, treat the Windows Insider Blog entry as the authoritative record for version numbers, gating, and the precise UX description; third‑party verification will follow as Insiders test the feature.

Critical analysis — strengths​

  • Productivity and flow: Letting Copilot edit text directly inside the app you’re using removes friction: no copy/paste, no toggling between windows, no manual reconciliation of suggested edits. This is a genuine productivity win for short edits and iterative drafting in context.
  • Multimodal continuity: The ability to run text edits within a Vision session — and to switch between typing, speaking, or accepting in‑flow edits — aligns with Microsoft’s goal of a fluid assistant that adapts to user preference and environment (quiet meetings, shared spaces, accessibility needs). Earlier updates already added text‑in/text‑out to Vision; this extends that usefulness to editing.
  • Explicit UX boundaries: The preview emphasizes session‑bound sharing and visible indicators (glow around shared windows), which are necessary UX/Privacy primitives. These controls reduce accidental exposure compared with an always‑on screen monitoring model.
  • Consistency with Microsoft 365 editing patterns: Microsoft has shipped interactive rewriting and suggestion workflows inside Word and other apps; extending similar capabilities to arbitrary third‑party text fields via Vision is a logical extension that can increase cross‑app productivity.

Critical analysis — risks and limitations​

  • OCR and context limits: Copilot Vision’s ability to edit text depends on accurate OCR and context extraction in arbitrary apps and editors. Complex rich‑text editors, embedded web components, or apps with custom controls may yield degraded results or cause misplaced edits. Users should not rely on this for mission‑critical or compliance‑sensitive edits without auditing outputs.
  • App compatibility and focus: The UX assumes that Copilot can place edits into the active text field you choose. Some apps use custom input controls or block programmatic text insertion; behavior will vary. Expect a compatibility matrix and bugs during preview.
  • Privacy and data flow: While sessions are permissioned and session‑bound, Microsoft’s feature uses cloud reasoning for many Copilot tasks on non‑Copilot+ PCs. That implies text and visual context may traverse Microsoft services. Enterprises concerned about data residency, retention, and auditability will need clear documentation and DLP controls before enabling this widely. Earlier Copilot Actions and Vision previews raised the same governance questions.
  • Regulatory and regional availability: Microsoft explicitly excluded the EEA from this staged rollout. That reflects either compliance/regulatory caution or localization readiness; either way, it limits broader testing and may delay enterprise adoption in jurisdictions with strict privacy/regulatory frameworks.
  • Model hallucination / inaccurate rewrites: Language models can and do make stylistic or factual errors. When Copilot rewrites a technical instruction, contract clause, or legal phrasing, small changes can have outsized consequences. The preview’s acceptance preview step is essential, but organizations must still enforce review workflows. Independent testing of earlier Copilot Vision interactions revealed accuracy issues for complex analysis tasks.

Enterprise considerations and recommended pilot checklist​

Enterprises evaluating Copilot Vision text editing should treat the December preview as a test rather than a production readiness signal. The following checklist is a practical pilot plan:
  • Inventory eligible devices and check prerequisites: ensure test devices have Copilot app 1.25121.60.0+ and Windows 26200.6899+.
  • Establish a pilot group limited to non‑sensitive content and to volunteers who can provide structured feedback.
  • Enable Copilot Actions toggle in the Copilot app Settings only on pilot devices and document the enabling process.
  • Configure Data Loss Prevention (DLP) and Conditional Access policies to restrict Vision and Actions usage on devices handling regulated data. Demand vendor documentation for data flows, retention, and logs.
  • Test app compatibility matrix: popular editors (Word, Outlook Web, Gmail in browser, Slack/Teams message boxes), content management systems, and custom enterprise apps. Note where text insertion fails or is inconsistent.
  • Define a human‑in‑the‑loop acceptance policy: require manual reviewer approval for any non‑trivial edits in regulated content (legal, compliance, finance).
  • Capture telemetry: track false positives/negatives, OCR failures, and time saved vs. manual editing. Feed results back to Microsoft via the Copilot app feedback mechanism.
This cautious, measured approach protects sensitive workflows while allowing IT to assess real productivity gains.

Limitations you will likely encounter in preview​

  • Not all apps will accept the in‑place edits; web editors with rich DOM structures or sandboxed inputs may not behave consistently.
  • Visual “Highlights” and on‑screen pointers — features demonstrated earlier for voice Vision — may not be fully integrated into the text editing path at first. That can make certain step‑by‑step guidance tasks less intuitive.
  • Performance will vary by hardware and network; unless you’re on a Copilot+ PC with local model acceleration, most heavy inference will run in the cloud, incurring latency. Expect slower response times on older devices.

How this fits in Microsoft’s broader Copilot roadmap​

The December text editing addition is an incremental but strategically consistent move: Microsoft is converting Copilot from a conversational helper into an assistant that can operate in context across apps. Prior milestones show this progression:
  • Text‑in/text‑out for Vision moved Vision away from being voice‑only and made typing a first‑class interaction mode.
  • Copilot Actions introduced a contained Agent Workspace for file and desktop tasks, with explicit opt‑in and low‑privilege agent accounts.
  • In‑app editing features in Microsoft 365 demonstrate Microsoft’s investment in contextual rewriting and editorial tools.
Taken together, these threads show Microsoft is pursuing a multi‑modal assistant that can see a screen, understand the content, suggest or perform edits, and act on behalf of the user under observable controls. The company’s focus on staged Insider testing suggests they intend to harden accuracy, app compatibility, and governance before a broad public rollout.

What to watch next​

  • Broader Insider feedback and early independent reviews. These will surface real compatibility and accuracy issues and will be the most reliable gauge of practicality for day‑to‑day work. Early reviewers have already highlighted accuracy and UX problems with Vision in other previews, so pay attention to detailed testing reports.
  • Enterprise controls and DLP integration. Microsoft must provide explicit admin controls, audit logs, and enterprise‑grade retention policies for organizations to enable Vision/Actions safely across large fleets. Recent Copilot previews have signaled Microsoft is aware of these needs but has not yet delivered full enterprise assurances.
  • EEA availability and legal guarantees. The current EEA exclusion suggests Microsoft is addressing regulatory considerations; public confirmation of changes to that stance will be critical for organizations operating in the EU.
  • Feature parity between Copilot+ (on‑device capability) and cloud‑first devices. On‑device inference via NPUs reduces latency and improves privacy guarantees; Microsoft’s continued investment in Copilot+ hardware tiers will shape the long‑term viability of real‑time editing in latency‑sensitive workflows.

Final assessment​

This Copilot Vision text editing preview is a logical, overdue extension of Microsoft’s multimodal Copilot strategy. The promise — convenient, in‑flow editing across apps without copy/paste — is compelling and could save meaningful time for frequent editors. Microsoft’s decision to make Vision session‑scoped and opt‑in, and to require an explicit Copilot Actions toggle, shows attention to consent and containment.
At the same time, the preview raises predictable and real concerns: OCR accuracy, app compatibility, model hallucination risks, and enterprise governance. Organizations and careful Insiders should test this feature on non‑sensitive content, demand strong telemetry and DLP documentation, and avoid enabling it broadly until Microsoft publishes enterprise controls that meet compliance needs. Independent reviews and community pilots will determine how quickly and widely this technology can be trusted in production workflows.
To try it today as an Insider: update to the Copilot app package noted in Microsoft’s announcement, confirm your Windows build meets the requirement, enable the Copilot Actions toggle inside the Copilot app Settings, start a Vision session via the glasses icon, place the cursor in a text field, and ask Copilot to rewrite or simplify the text — then review the preview before accepting changes. Provide feedback inside the Copilot app so Microsoft can iterate the feature based on real‑world usage.
Community commentary and prior Insiders’ writeups provide helpful context about how Microsoft stages and iterates Copilot features; those threaded reports emphasize that the staged rollout is expected and intentional, and they repeatedly urge pilots and careful telemetry collection during testing. For practical planning and risk management, follow that guidance closely while testing the new text editing mode.
This preview is an important milestone for Copilot’s practical utility on the desktop — useful and promising, but not yet a drop‑in replacement for careful human editing in regulated or high‑stakes content.

Source: Microsoft - Windows Insiders Blog Copilot on Windows: New text editing feature begins rolling out to Windows Insiders