Quick answer: in the same few weeks, OpenAI, Microsoft, and Google each shipped a version of the same underlying bet — an AI agent that reaches across multiple apps to actually complete a task, rather than a chat window you copy answers out of. OpenAI brought voice-triggered agentic workflows to ChatGPT mobile, Microsoft rebuilt Copilot Cowork around federated connectors and a model picker that includes both GPT and Claude models, and Google shipped five new agentic capabilities across Gmail, Docs, Sheets, and Slides plus a native Gemini app for Windows. The capability differences are real, but so is the shared direction: the product surface for "AI at work" is moving out of a single chat box and into the tools people already have open.
OpenAI: ChatGPT's Work Tab Comes to Mobile, With Voice
OpenAI announced on September 23 that ChatGPT's mobile app now supports voice-triggered agentic workflows — Pro and Plus subscribers can talk to the phone and have it draft a document, write an email, or summarize a Slack channel through the existing Work tab, which also covers building simple sites, generating presentations, and using a cloud browser. Free and Go-tier users get plugins and connected apps instead of the full agentic Work tab. OpenAI says voice conversations now produce richer text output and that a task started by voice on a phone can be picked up on desktop — with recurring scheduled agents (via the existing Pulse feature) and deeper calendar/reminder integration named as the near-term roadmap.
Microsoft: Copilot Cowork Gets Federated Connectors — Built on MCP
Microsoft's September Copilot wave, detailed by M365 Copilot Connection, centers on Copilot Cowork: General Availability for federated Copilot connectors, which let Copilot securely pull real-time data from third-party sources using the Model Context Protocol — the same open standard we've covered as the emerging way AI assistants connect to external tools and data, now shipping as a first-party Microsoft 365 feature rather than a developer-only integration pattern. Cowork also gained cost visibility, shareable skills, and a rebuilt Notebooks UI. One detail worth flagging for anyone standardizing a company's AI stack: the model picker moved into the prompting box itself and is now split into a GPT section and a Claude section — GPT-6 Astra and Claude Fable 5.1 are both available inside Copilot Cowork and Copilot Studio, the same two models we compared head-to-head earlier this month — meaning the model-choice decision we wrote about now happens inside Microsoft's own product, not just at the raw API level.
Google: Cross-App Agentic Capabilities, Plus a Native Windows App
Google's September Workspace update, per Google's own announcement, lets a user prompt Gemini to complete a task that spans Gmail, Drive, Docs, Slides, and Chat from wherever they're currently working — Gemini acts as an orchestrator, generating a formatted Doc, a structured Sheet, or a styled Slide deck without the user needing to manually switch apps to assemble the result. Google also shipped a native Gemini desktop app for Windows, invoked with an Alt+Space shortcut to bring Gemini over whatever's currently on screen for a quick fact-check or brainstorm, and added the ability to use Gemini Notebook as a grounding context source directly inside Google Docs — pulling from a curated research base into a draft without tab-switching or copy-paste.
What Shipped, Side by Side
| OpenAI (ChatGPT) | Microsoft (365 Copilot Cowork) | Google (Workspace) | |
|---|---|---|---|
| Core new capability | Voice-triggered agentic Work tab on mobile | Federated connectors (MCP-based) + dual GPT/Claude model picker | Cross-app orchestration (Gmail/Drive/Docs/Slides/Chat) |
| New surface | Mobile app | Prompting box (model picker), Notebooks UI | Native Windows desktop app |
| Underlying model choice | OpenAI models only | User-selectable: GPT-6 Astra or Claude Fable 5.1 | Gemini only |
| Announced | Sep 23, 2026 | September 2026 (Cowork wave) | September 2026 |
The Pattern: From Chat Window to Embedded Cross-App Agent
The specific features differ, but all three launches point at the same underlying shift: none of them are primarily about a smarter model this time — Sonnet 5.5, GPT-6 Sol/Luna, and Astra already covered that ground earlier this month. These three releases are about where the AI sits relative to the actual work — inside the document editor, triggered by voice from a phone, orchestrating across five apps instead of waiting in one chat tab for a copy-pasted answer. That's a meaningfully different adoption curve for a business to plan around than a raw model upgrade: it changes onboarding, IT policy, and which team owns the rollout, not just which API key gets called.
What This Means for Choosing (or Standardizing) a Workplace AI Stack
Microsoft's dual-model picker is the most consequential detail here for a business that's already committed to one ecosystem: it means "which frontier model" and "which productivity suite" are no longer the same decision inside Microsoft 365 — a Copilot Cowork user can pick GPT-6 Astra or Claude Fable 5.1 per task, without switching platforms. That's worth knowing whether you're already on Microsoft 365 or evaluating it against a Google Workspace or standalone-ChatGPT deployment, since it changes what "which AI vendor should we standardize on" actually means when the productivity suite is decoupled from the underlying model. As with any new agentic surface, the evaluation discipline is the same one we lay out in our enterprise AI vendor checklist: pilot it against a real task in your actual workflow before rolling it out broadly, rather than assuming a keynote demo generalizes to your document set and your data.
If you're evaluating which of these platforms actually fits how your team works — or whether a custom-built agent would serve you better than any of the three — reach out at info@digit.com.pk.