7/16/2026↻ Updated 10/9/2026

ChatGPT Work, Claude Cowork and Gemini: AI takes on ongoing work

Realistický cover sjednocené AI pracovní aplikace

October 9, 2026 update: Gemini takes on a team role

Google announced Gemini agent on October 8: a cloud agent for extended tasks, connected to Workspace. Its coworker version gets an account, email, calendar and storage. Google describes separate agent identity and action auditing. This announcement does not establish availability of every feature in a particular Czech business account.

The operational issue is ownership. An agent with an inbox needs a process owner, rules for incoming requests and a clear handover when work remains unfinished. Giving it a name and inviting it into team chat is only the beginning.

A pilot for sales meeting preparation

I would start with one team and twenty meetings. The agent could read a designated Drive folder and materials for the relevant customers. It would return a brief with source links, unresolved facts and proposed next steps. CRM changes and customer messages would require review.

In n8n or Make, I would maintain a separate task record: meeting identifier, status, brief link, reviewer and confirmation of the completed write. Repeated requests must not create duplicate follow-ups. This is a proposed pilot architecture, not a verified Gemini connector or an integration tested here.

I would measure factual corrections and the salesperson's remaining work after receiving the brief. Tests should include revoked folder access, a cancelled meeting and a missing CRM record. The agent should stop or flag missing information instead of inventing it.

Before enabling the pilot, I would check the actual account's plan, region, permission controls and how running work is terminated. A team identity is useful when its actions can be traced and stopped. The dated sections below preserve the history of Work, Dots and Cowork.

September 30, 2026 update: Dots and ongoing responsibility

Dots arrived with the September 29 DevDay. Coverage by Jirka Herník and Marek Bartoš prompted this review; the official weekly update confirms the capabilities. A dot takes on an ongoing responsibility and can delegate tasks to Work or Codex. The useful change is continuity: people need not reconstruct the same assignment every morning.

Access depends on the plan. Pro 100/200/500 rolls out to adults outside the EEA, UK and Switzerland. Business Premium and Enterprise roll out worldwide; Enterprise starts disabled. The video's shorthand that Dots are unavailable in the EU is therefore too broad. A rollout also does not promise immediate access.

Give it one useful responsibility

My first pilot would be morning sales-meeting preparation: provide sources, flag missing information and queue proposed CRM changes for review. After a week, measure corrections to the briefs and actual time saved by sales staff. This is a proposed pilot, not a test we have completed.

Setup treats app connections, messaging and computer access separately. Contact through Slack does not grant CRM access. Work on a local device requires that computer to stay online with the app open.

Pause is not a master stop button

The controls guide separates pausing the main task, stopping delegated work and canceling a schedule. During an incident, inspect both Activity and Scheduled; pausing alone does not stop everything. I would require approval for customer messages. Custom rules are instructions the agent can get wrong, so connected-system permissions must enforce the intended boundaries too.

For a content pipeline, I would initially authorize research and drafts while keeping publication with an editor. For invoices, separate extraction from payment approval. An ongoing agent needs a named process owner just as an n8n or Make automation does.

Historical context: the following sections retain earlier Voice, Scheduled and Work/Cowork information. Use this update and current documentation for Dots availability.

September 10, 2026 update: Voice is an agent control plane, not just dictation

OpenAI is rolling out ChatGPT Voice across Chat, Work, and Codex in the desktop app. This is more than speech-to-text. A voice session uses the context and selected model of the task you are discussing, so you can ask for progress, interrupt a response, and redirect the work without opening a new conversation.

The coordination layer matters more. Voice can start a separate long-running task, inspect existing tasks, send follow-up instructions, and switch the conversation to another Codex task on a connected host. After pairing a phone with the desktop, the same flow is available through Remote on iOS. Voice is becoming a control surface for work that continues on the computer.

In a company workflow, I would use it for three situations: changing priorities while away from the desk, checking blockers without opening a laptop, and adding context to a report, analysis, or coding task already in progress. For example: “pause deployment and run the integration tests first,” “tell me what is blocking invoice processing,” or “add the latest Slack feedback to the brief and prepare a revised version for review.”

Voice does not change task permissions. OpenAI says it follows the same permission boundaries as the Chat, Work, or Codex task it directs. That is the right model: speech may redirect work, but it should not bypass approval for sending email, writing to a CRM, publishing content, or changing production. On macOS, optional Screen context can also share an appshot containing both the window image and accessible text outside the visible scroll area, so sensitive applications should stay out of shared context.

Availability is still rolling out. Voice is offered in the desktop app for Plus, Pro, Business, Edu, and Enterprise plans, with a two-week early-access period for Edu and Enterprise. Only one voice chat can be active at a time, voice has a separate allowance measured in rolling five-hour windows, and Codex tasks started through Voice continue to consume the Codex usage budget.

The practical takeaway is simple: voice shortens the distance between human intent and a running agent, but audit, budgets, permissions, and approval gates must remain part of the workflow.

Source for this update: OpenAI ChatGPT Voice; practical signal and walkthrough from AI s rozumem.

Update, 29 August 2026: “schedule this” is becoming an event-driven workflow

OpenAI’s latest documentation clarifies what Scheduled in ChatGPT and ChatGPT Work can do today. It is no longer only a reminder or recurring digest. For eligible accounts, event-triggered tasks in Work can react to a new Gmail message, a new Slack-channel message, or supported GitHub pull-request activity. The user reviews the trigger, condition, and prompt before it runs; actions that need approval pause for a human.

That is a practical shift. Rather than adding another isolated bot, a team can describe a small work loop: “when an email from this customer arrives, prepare a summary, check CRM, and draft a reply”; “when an incident appears in the support channel, classify it and prepare an escalation”; or “when a PR is created, check the checklist and prepare review context.” It does not mean an agent should automatically send the email or merge the code. The first deployment should end with a draft, ticket, or approval step.

The limits matter too. Event-triggered tasks require Work and an eligible paid plan; Free and Go do not have them. Gmail can react to a new message and filter sender or subject, Slack monitors new messages in an authorized channel with @ChatGPT, and GitHub supports selected PR events. Enterprise and Edu admins must enable the capability. That is a useful reminder that connector permissions, auditability, and operational ownership are not details to solve after the pilot.

For companies, I would use one simple rule: the trigger reads narrowly, the output is traceable, and an external action needs approval. Keep CRM writes, customer email, production-data changes, and PR merges behind a person at first. Move that boundary only after results are measurable and rollback is clear.

Source for this update: OpenAI Scheduled Tasks in ChatGPT. The new weekly overview from Zaujaloma AI was the signal to verify it.

For the last two years, AI tools have looked like many separate apps. Chat for questions. Codex or Claude Code for code. Canvas or artifacts for writing. Agents in separate windows. Connectors somewhere in settings. Automations somewhere else.

That direction is changing. On July 9, 2026, OpenAI introduced ChatGPT Work and a new desktop app that brings Chat, Work, and Codex together. Anthropic is building Claude Cowork as a work mode next to chat: an agent that works with files, tools, plugins, scheduled tasks, and longer deliverables. The recent YouTube signals from AI s rozumem and Zaujaloma AI captured the practical issue well: people are no longer asking only "which model is new", but where the boundary between chat, app, agent, and work environment is going.

That is more important than the redesign itself. AI apps are starting to become a work operating layer.

Diagram: AI apps as a work operating layer

What is new at OpenAI

OpenAI describes ChatGPT Work as an agent for longer, more involved tasks. It can research and analyze information, work across connected apps and files, and create finished documents, spreadsheets, presentations, reports, and Sites. The user can follow progress, answer questions, change direction, and approve important actions.

The new desktop app for macOS and Windows makes the product shift explicit: Chat is for questions and conversation, Work is for research and finished deliverables, and Codex is for software. The former Codex app updates into the new ChatGPT desktop app. The previous ChatGPT desktop app may remain as ChatGPT Classic, but new agent features live in the new app.

One important detail: Work on web and mobile runs in the cloud. Work in the desktop app can use local files and desktop apps with permission. OpenAI also says that at launch, cloud Work conversations do not appear in desktop Work, and desktop Work threads with local files remain on that computer. So the strategic direction is unification, but technically it is not yet one perfectly connected memory.

ChatGPT Sites points in the same direction. A website or lightweight internal app is no longer a separate product on the side; it can be created from Work on the web or from Work/Codex in the desktop app. OpenAI is also retiring Atlas as a separate browser experiment and moving browser-based agentic capabilities into ChatGPT and Codex.

What Claude Cowork shows

Anthropic describes Claude Cowork simply: when you want to hand off a task, switch from Chat to Cowork. Cowork works in selected folders and tools, carries the task end-to-end, and delivers work for review. It runs on web, desktop, and mobile, supports scheduled tasks, and can split larger projects into chunks that run in parallel.

The official product guide makes the practical point clearer: Claude Cowork is for knowledge work, not just programmers. It works with local files, connected apps such as Slack and Google Drive, subagents, long-running tasks, scheduled tasks, and plugins. Anthropic also published data from 1.2 million anonymized Cowork sessions in May 2026. The largest category was business process and operations: reports from scattered updates, onboarding checklists, and spreadsheet reconciliation. The next large category was content creation and copywriting.

That matters. Claude Code is for code. Claude Cowork is for the work around work: reports, decks, checklists, documents, CRM prep, folder audits, and meeting prep. These are exactly the things that consume time inside companies because they require copying context between email, spreadsheets, Slack, CRM, and documents.

Not one super app. Three work modes.

I would not read this as "everything will live in one app". A better mental model is three modes.

The first mode is Chat. Quick questions, brainstorming, explanation, search, decisions, and short iteration. The output can stay in the conversation.

The second mode is Work / Cowork. Here I do not want only an answer. I want a finished artifact: document, spreadsheet, presentation, report, brief, internal website, onboarding checklist, meeting prep, or CRM summary. The agent needs sources, a plan, intermediate questions, longer runtime, and clear review.

The third mode is agent runtime. Codex, Claude Code, browser, terminal, local files, repositories, tests, and tool calls. Here the agent is not only writing text. It is touching systems. That requires stricter permissions, isolation, audit, and approvals.

If a company does not separate these modes, it will create confusion. People will try to run heavy work in ordinary chat, agents will receive too much access, outputs will not land where they should, and nobody will know what actually happened.

How to deploy this inside a company

I would start with tasks that are expensive because of context switching, not because they require genius. For example:

  • sales meeting prep: CRM, email, call notes, customer website, recent support tickets,
  • weekly status report: Slack, Linear/Jira, GitHub, metrics spreadsheet, comments from team leads,
  • content pipeline: brief, sources, draft, review, localization, publishing,
  • support classification: ticket, customer history, response draft, escalation path,
  • document extraction: contracts, invoices, audit checklist, risks, human approval.

For every process, I would define five things.

First, sources. Which folders, apps, and data types may the agent read? Does it need CRM, Gmail, Drive, Slack, a local folder, or only uploaded files?

Second, actions. Can it only create a draft, or can it write to CRM, send an email, publish a Site, change a file, or create a pull request?

Third, approval. What can happen automatically, and what requires a person? Sending emails, changing customer data, creating legal commitments, and modifying production code should require approval.

Fourth, audit. Store the prompt, sources, output, changes, approvals, and rollback path. Without this, errors and trust will be hard to manage.

Fifth, measurement. Not "how many people use AI", but whether time to output went down, CRM quality improved, manual copying decreased, support triage became faster, or invoice errors dropped.

Why this is a strong trend

An arXiv study of Codex usage reports that active users of agentic AI grew more than fivefold in the first half of 2026, with the fastest growth outside the original developer audience. More than 10% of users managed three or more concurrent Codex agents in at least one week, and 26.6% used skills to share instructions for more complex workflows.

In other words, agents are not only about programming. Programming was the first good test environment because it has files, tests, git, diffs, and a clear review process. Now the same pattern is moving into knowledge work. Instead of a pull request, the output is a report, deck, spreadsheet, internal website, CRM update, or audit folder.

WorkBench Revisited adds the necessary caution. Workplace agents have improved substantially, but they can still make harmful mistakes, such as emailing the wrong person. So the conclusion is not "let agents into everything". The right conclusion is: start designing work processes where it is clear what the agent may access, what it must cite, and what a human must approve.

Where this is heading

A year from now, the most important question will not be whether you use ChatGPT, Claude, Gemini, or something else. The more important question will be whether you have a work layer for agents.

That layer will manage:

  • unified access to sources,
  • separation between fast chat and longer work tasks,
  • specialized agents for code, data, documents, support, and sales,
  • company plugins and skills,
  • audit trail,
  • costs and limits,
  • and human approval where an action can break something.

OpenAI and Anthropic are approaching this differently, but the target is similar. Chat is no longer the main product. Chat is the input mode. The value is moving into work agents that can turn conversation, files, and apps into finished output.

For companies, this is good news if they do not treat it as just another desktop icon. It is a chance to redesign a few concrete workflows: less copy-paste, fewer lost notes, fewer manual reports, better audit, and faster follow-up. But only if agents get the right boundaries.

Sources: OpenAI ChatGPT release notes, OpenAI ChatGPT Work and Codex, OpenAI Moving to the new ChatGPT desktop app, OpenAI Creating and managing ChatGPT Sites, Claude Claude Cowork product page, Claude How people are using Claude Cowork, Claude The Claude Cowork product guide, Axios Finding your goldilocks GPT-5.6 model, arXiv The Shift to Agentic AI: Evidence from Codex, arXiv WorkBench Revisited, YouTube signals from AI s rozumem and Zaujaloma AI.