# ChatGPT Work Overview and Use Cases

> Comprehensive guide to ChatGPT Work: GPT-5.6 agentic architecture, differences from standard Chat and Codex, enterprise plugins, Computer Use, ChatGPT Sites, and reusable Skills.

## 1. What is ChatGPT Work

ChatGPT Work is a specialized agentic mode of ChatGPT designed for long-running, multi-step business tasks rather than quick conversational Q&A exchanges. The agent independently retrieves context from connected enterprise apps, analyzes complex datasets, and drives projects through to finished deliverables: structured spreadsheets, analytical reports, interactive slide decks, or internal lightweight web apps.

Under the hood, ChatGPT Work builds upon the autonomous execution foundations originally pioneered for Codex. While Codex was initially created for software engineering, real-world adoption revealed massive demand outside engineering teams — over one million non-developers use it daily for general business tasks. ChatGPT Work extends this agentic delegation paradigm across operational and business workflows.

> [!NOTE]
> **Next-Generation Agentic Foundation:** ChatGPT Work is powered by the GPT-5.6 model, tailored specifically for complex multi-step reasoning. It demonstrates state-of-the-art adherence to source files, strict enterprise templates, and business logic without hallucinating facts.

### 1.1. How the Agentic Model Works in Practice

The core philosophy is delegating entire workflows that you thoroughly understand: for instance, investigating month-end budget variances, compiling a customer discovery brief for a major sales pitch, or synthesizing market feedback into a comprehensive campaign plan.

Throughout execution, you maintain complete supervisory control:
- **Monitor real-time progress:** inspect the transparent chain of thought and external tool invocations as they occur.
- **Answer clarifying questions:** the agent reaches out when it encounters ambiguities or requires missing critical inputs.
- **Steer the direction of work:** refine constraints and modify directives in-flight without restarting from scratch.
- **Approve critical actions:** review and authorize external file edits, outbound emails, or database write operations.

ChatGPT Work executes full chains of dependent actions from a single prompt: for example, analyzing customer interview transcripts, synthesizing an advertising brief, generating localized copy variants, and formatting executive decks while preserving end-to-end context.

### 1.2. Real-World Case Studies Across Industries

- **Marketing & Lead Qualification (Zapier):** The Zapier growth team built an automated qualification pipeline evaluating thousands of inbound leads monthly. The agent tracked customer touchpoints across CRM, email, and trackers, flagged stalled follow-ups, and generated an executive dashboard every Monday morning.
- **Enterprise Sales Enablement:** An enterprise sales agent converted raw notes from a needs-discovery call into a personalized, functional solution prototype within 24 hours — a process that historically took weeks. The agent organized specifications, coordinated with solutions architects, and aligned technical deliverables.
- **Corporate Finance & Accounting:** ChatGPT Work reduced the month-end closing cycle from multiple days to hours. The agent locates bank statements, synchronizes data into Google Sheets / Excel, reconciles discrepancies, and prepares audit-ready slide decks.

## 2. Chat, Work, and Codex — What’s the Difference

The modern ChatGPT ecosystem is structured around three dedicated modes, each calibrated for distinct operational demands.

### 2.1. Chat: Quick Answers and Daily Brainstorming
The familiar conversational mode for rapid questions, ad-hoc research, drafting short messages, and creative brainstorming. Context is confined to the active conversation window, operating on a request-and-reply rhythm.

### 2.2. Work: Autonomous Multi-Step Projects
An agentic workspace designed to produce finished business artifacts: comprehensive research memos, data models, slide presentations, and lightweight sites. It features native access to local folders, corporate connectors, and automated workflows.

### 2.3. Codex: Software Engineering and Repositories
A dedicated environment for software developers: code generation, debugging, Git repository navigation, running terminal commands, executing unit tests, and preparing pull requests. Operates independently from daily chat histories.

### 2.4. Core Comparison Matrix

| Criteria | ChatGPT Chat | ChatGPT Work | Codex |
| :--- | :--- | :--- | :--- |
| **Primary Purpose** | Quick inquiries, creative ideation, short copy | Multi-step projects, spreadsheets, slide decks, internal sites | Software engineering, debugging, testing, Git workflows |
| **Context Scope** | Uploaded files within the active session | Local project folders, cloud storage, enterprise data rooms | Full project repositories, Git branches, system files |
| **Tool Ecosystem** | Web search, custom GPTs | Enterprise plugins, Computer Use, embedded browser | CLI, Git, terminal commands, debuggers, IDE extensions |
| **Execution Environment** | Cloud: web, mobile app, desktop | Cloud + secured local desktop agent | Dedicated desktop app, CLI terminal, VS Code / JetBrains |
| **Resource Accounting** | Fixed message caps per 3-hour window | Task-based metering determined by complexity | Complexity-based token and compute consumption |

### 2.5. Platform Availability and Synchronization

Cloud-based Work threads synchronize seamlessly across web browsers, smartphones, and desktop clients. You can initiate an analysis on your laptop and monitor execution progress from your phone.

Sessions operating against local desktop files remain securely isolated on that specific machine to ensure maximum data privacy.

## 3. Getting Started with Work Mode

Onboarding is optimized for diverse device setups: the desktop client offers deep filesystem integration, while the web app allows instant cloud-based collaboration.

:::tabs
=== Desktop Application (macOS / Windows)
1. **Launch the App:** Open the ChatGPT desktop application and sign in with your enterprise or personal account.
2. **Select the Environment:** In the upper-left navigation, select `ChatGPT` (or switch to `Codex` when working on codebase repos).
3. **Activate Work Mode:** Toggle the top mode switch from `Chat` to `Work`.
4. **Attach Local Project Directory:** Click to select a local folder. All contained files, templates, spreadsheets, and documentation will instantly serve as grounded agent context.
5. **Define the Objective:** Clearly state the target deliverable, provide templates or tone guidelines, and specify acceptance criteria.
6. **Iterative Review:** Supervise intermediate milestones, answer agent inquiries, and approve system operations.

=== Web Browser and Mobile Apps
1. **Access the Portal:** Navigate to [chatgpt.com](https://chatgpt.com) in any modern browser or open the mobile app on iOS/Android.
2. **Switch to Work:** From the model and mode dropdown menu, activate `Work`.
3. **Create or Join a Project:** Open a dedicated project space or join a shared team workspace.
4. **Load Knowledge Context:** Attach corporate documentation, cloud files, or links to shared Google Docs and SharePoint drives.
5. **Execute and Monitor:** Send your workflow prompt and track execution updates across devices with automatic cloud state synchronization.
:::

> [!IMPORTANT]
> **Local Data Isolation Principle:** The web and mobile apps cannot directly browse your local computer's drive. Filesystem read/write operations execute exclusively through the desktop client, strictly confined to the directory you explicitly authorize.

## 4. Connecting Applications and Managing Context

ChatGPT Work delivers maximum value when integrated into the systems where your team does daily work: documentation hubs, email, CRMs, and issue trackers.

### 4.1. The Plugin Ecosystem and Catalog
Plugins establish bidirectional bridges between ChatGPT and your tech stack: Slack, Microsoft Teams, Google Drive, SharePoint, Notion, Jira, Salesforce, HubSpot, and proprietary internal databases via custom APIs.

The agent intelligently routes tasks to appropriate plugins based on your instructions. You can also explicitly trigger a tool using the `@` mention syntax (e.g. `@Slack` or `@GoogleDrive`).

### 4.2. Workflows with Connected Data
Once connectors are authorized, the agent can:
- **Correlate cross-platform signals:** synthesize customer discussions from Slack with Salesforce records and product roadmaps.
- **Generate grounded artifacts:** formulate research memos and financial models backed by verified source facts.
- **Perform background synchronization:** continuously check for modified files and reflect changes in existing presentations.

### 4.3. Industry-Specific Toolsets
Dedicated bundles exist for specific departments. For example, the Sales Suite connects Salesforce, HubSpot, Slack, Outreach, Clay, Rox, and Actively: the agent pinpoints high-priority accounts, compiles pre-meeting briefing sheets, and logs updates into CRM fields.

### 4.4. Real-Time Dynamic Data Enrichment
ChatGPT Work utilizes real-time browsing to enrich deliverables with external market intelligence (such as regulatory updates or competitor filings), ensuring your deliverables remain living assets rather than static point-in-time snapshots.

## 5. Automation: Scheduled Tasks

The Scheduled Tasks feature automates repetitive operational workflows, freeing up valuable time for strategic analysis.

### 5.1. Typical Background Automation Scenarios
Through native browser and application hooks, the agent can run on scheduled intervals:
- **Weekly Meeting Prep:** summarize new Slack/Teams discussions and update recurring team agendas every Monday.
- **Morning Metric Digest:** scan analytics dashboards and competitor portals, compiling an executive highlight reel with detected variances.
- **Customer Feedback Aggregation:** parse incoming support tickets, cluster recurring pain points, and draft roadmap suggestions.
- **Dynamic Slide Updates:** refresh client-facing presentation decks immediately upon receiving updated metrics via email.

### 5.2. Human-in-the-Loop Governance

> [!TIP]
> **Retaining Human Authority:** Autonomous execution does not mean loss of oversight. You establish permission boundaries: low-risk activities (such as reading public metrics or drafting internal memos) run autonomously, while sensitive operations (such as notifying external partners or committing database updates) require explicit one-click human sign-off.

## 6. Browser and Desktop Agentic Capabilities

The desktop client incorporates a sandboxed embedded browser and desktop execution capabilities to eliminate context-switching friction.

### 6.1. Built-in Browser Capabilities
- **Frictionless Research:** explore market trends, compare external benchmarks, and cross-examine documentation without switching windows.
- **Cloud Workspace Operations:** directly view, edit, and verify assets in Google Workspace and Microsoft 365 environments.
- **Data Validation:** cross-reference public figures against internal company spreadsheets.

### 6.2. Computer Use & Desktop Automation
The Computer Use module allows the agent, when instructed, to interact directly with desktop software: moving files between local directories, interacting with legacy desktop software lacking modern APIs, and processing data across application windows.

### 6.3. Chrome Extension & Atlas Evolution

> [!NOTE]
> **Evolution of Project Atlas:** The autonomous browsing capabilities developed within the experimental Atlas browser are now natively integrated into the ChatGPT Chrome sidebar extension and the desktop app, enabling on-page agentic assistance on any website.

## 7. ChatGPT Sites — Websites and Lightweight Apps

ChatGPT Sites allows teams to assemble and publish responsive internal web portals or interactive micro-apps directly from Codex in minutes.

### 7.1. Comparing Formats: From Documents to Web Apps

| Format | Core Purpose | When to Choose | Example Deliverable |
| :--- | :--- | :--- | :--- |
| **Document** | Static reading and information transmission | Single-instance briefings, policies, strategic memos | A release roadmap PDF or Google Doc |
| **Internal Site** | Collaborative hub revisited continuously | Long-term projects requiring aligned multi-team status | Product launch tracker with milestones and checklists |
| **Lightweight App** | Interactive calculations and scenario modeling | Streamlining recurring calculations without engineering | ROI calculator or financial modeling micro-tool |
| **Full Software** | Scalable enterprise software platform | Heavy backend logic, proprietary databases, high security | Custom in-house ERP or CRM software system |

### 7.2. High-Impact Internal Site Concepts

| Team Need | Site Format | Key Components to Include |
| :--- | :--- | :--- |
| **Launch Coordination** | Product Release Tracker | Milestones, deadlines, task owners, resource links, risk register |
| **Metrics Review** | Team Executive Dashboard | Real-time charts, weekly KPI comparisons, OKR milestone progress |
| **Scenario Modeling** | Interactive Decision Calculator | Input sliders, logic formulas, dynamic projection charts |
| **New Hire Onboarding** | Welcome Portal | Project context, team contacts, essential links, week-one checklist |
| **Knowledge Management** | Central Team Hub | Team conventions, recurring updates, archive of decision logs |

### 7.3. Step-by-Step Construction in Codex

1. **Specify Requirements:** Describe the purpose, target users, and interface elements to Codex.

```text
Build a launch tracker for our cross-functional product team.
Include:
1. Release stages and deadlines with filtering by owner.
2. A prominent risk evaluation section at the top of the page.
3. A weekly update log optimized for fast scanning during status reviews.
Ensure the layout is responsive, clean, and styled according to corporate brand standards.
```

2. **Inspect the Prototype:** Launch the preview build to test UX responsiveness, layout balance, and interactive filters.
3. **Iterate with Agent:** Instruct Codex to rearrange components, adjust typography, or introduce additional interactive widgets.
4. **Deploy Live:** Issue the command "Deploy this site and provide the URL." Codex returns a secure internal web link.
5. **Configure Workspace Permissions:** Specify who in the organization can access the deployed portal.

### 7.4. Pre-Deployment Governance Checklist
- Audit code and assets to ensure no sensitive credentials, API keys, or private customer records are exposed.
- Verify that all outbound links and interactive filters function as intended.
- Confirm access boundaries conform to enterprise workspace sharing policies.

### 7.5. Maintaining Up-to-Date Site Content
Because lightweight sites lack direct live database connections, content freshness is maintained via scheduled tasks: an agent collects morning metrics from Jira or Sheets and refreshes the site content automatically.

### 7.6. Frequently Asked Questions

| Question | Answer and Guidance |
| :--- | :--- |
| **Do I need software engineering skills?** | No. You describe requirements in natural language. Codex scaffolds the frontend and refines styling based on your feedback. |
| **Does this replace full-stack development?** | No. Sites are suited for agile team dashboards and calculators. Mission-critical platforms still require full software development cycles. |
| **Can I update the site after publishing?** | Yes. Ask Codex for updates ("adjust button styles", "add chart"), review changes in preview, and republish with one click. |
| **Is there direct live database connectivity?** | Direct live database write-backs are not currently available. Content is updated via automated background scheduled jobs. |

## 8. Skills — Reusable Workflows for Teams

Skills transform bespoke prompting habits into reusable, standardized operating procedures that ChatGPT executes with predictable fidelity.

### 8.1. Anatomy of an Agent Skill
Every skill incorporates three core components:
- **Name and Trigger Description:** semantic definitions enabling the model to know exactly when to invoke the skill.
- **Actionable Playbook (`SKILL.md`):** structured guidelines, input schemas, processing steps, and output specifications.
- **Resource Repository:** reference examples, brand guidelines, document templates, and compliance constraints.

### 8.2. The SKILL.md Configuration Blueprint

The `SKILL.md` file follows the open Agent Skills standard, recognized across modern agent ecosystems:

```markdown
---
name: executive-weekly-brief
description: Compiles an executive summary from connected Slack channels, Jira boards, and analytics sheets
version: 1.0.0
---

# Skill Purpose
Synthesizes team achievements from the preceding week, identifies blockers, and formats an executive brief for leadership.

## Required Inputs
- Recent messages from #announcements and #project-updates channels.
- Current sprint ticket statuses from Jira.
- Conversion metric extracts from Google Sheets.

## Execution Steps
1. Retrieve completed tasks from the active sprint.
2. Categorize items by stream: Engineering, Product, Design.
3. Flag high-priority blockers referencing @leadership.
4. Generate the final briefing memo matching the corporate template:
   - 🎯 Key Wins
   - ⚠️ Risks & Blockers
   - 📈 Weekly Metrics
   - ➡️ Next Week Priorities
```

### 8.3. Advantages of Skills Over Ad-Hoc Prompting
- **Unwavering Consistency:** eliminates accidental omissions and formatting discrepancies across team members.
- **Institutional Memory:** codified standards ensure newcomers produce senior-level outputs from day one.
- **True Knowledge Transfer:** processes reside in transparent version-controlled files, not isolated in personal prompt histories.
- **Cross-Platform Portability:** skills can be exported and utilized across ChatGPT, Codex, and compatible agent environments.

### 8.4. Skills vs Custom GPTs vs Projects
- **Skills:** focused procedural blueprints teaching the agent "how to execute this specific business routine."
- **Custom GPTs:** persona-driven agents tailored with custom system prompts for ongoing conversational assistance.
- **Projects:** shared workspace hubs bringing together context files, chat threads, and deliverables under a single team initiative.

### 8.5. How to Build and Distribute a Skill

1. **Identify Repetitive Routines:** look for weekly reports, client discovery synthesis, or compliance audits.
2. **Draft Instructions with the Agent:**

```text
Draft an Agent Skill configuration named "deal-risk-review".
Objective: review open CRM opportunities exceeding $20k every Friday.
Inputs: deal value, expected close date, and latest sales rep notes.
Output: a 3-column table (Deal Name, Identified Risk, Recommended Rep Action).
Include strict quality checks preventing fabricated facts.
```

3. **Install in Workspace:** review the generated configuration and validate against 2–3 sample tasks.
4. **Share with Team:** publish to your department catalog through workspace administrative settings.

### 8.6. Enterprise Departmental Skills Matrix

| Department | Multi-Step Process | Tool-Based Skill | Conventions & Standards |
| :--- | :--- | :--- | :--- |
| **Marketing** | Converts raw campaign concepts into structured briefs with target channels and timeline | Ingests analytics metrics and compiles weekly multi-channel ROI summaries | Rewrites campaign drafts to align strictly with corporate brand tone of voice |
| **Sales** | Synthesizes discovery notes into customer pain-point maps and follow-up agendas | Inspects CRM deal hygiene to flag timeline risks and slippage indicators | Formats outbound outreach drafts to team standards for structure and length |
| **Finance** | Generates month-end variance reports explaining underlying cost drivers | Extracts actuals vs. budget from connected models and flags variance deltas | Formats financial analyses into standardized executive memos with sensitivity tables |
| **Engineering** | Transforms architectural design docs into sprint execution milestones | Converts backlog notes into Jira epics and stories with defined acceptance criteria | Standardizes pull request descriptions and release notes to repo conventions |
| **Operations** | Translates informal workflows into standard operating procedures with SLAs | Pulls operational KPIs from connected sheets and generates weekly efficiency reports | Formats team status updates into a structured framework: Changes, Impact, Action |
| **Management** | Aggregates disparate team updates into executive briefings and blocker lists | Parses signals from Jira, Sheets, and Slack to monitor team velocity and focus | Builds 1:1 meeting agendas incorporating goal alignment and growth discussions |
| **Leadership** | Distills complex cross-department reports into single-page decision memos | Synthesizes enterprise signals for executive morning briefings | Authors board-level communications emphasizing verified performance data |

### 8.7. Administrative Security and Governance FAQ

| Administrator Question | Security Policy & Guidance |
| :--- | :--- |
| **Are Skills enabled by default?** | In the initial enterprise beta, Skills are disabled by default; workspace administrators must activate them in settings. |
| **Are skills portable between ChatGPT and Codex?** | Yes. OpenAI adheres to the open Agent Skills standard, enabling seamless `.md` export and import across tooling. |
| **How are access permissions governed?** | Fully governed via Role-Based Access Control (RBAC): administrators specify permissions for authoring, publishing, and installing. |
| **Do skills impact enterprise data privacy?** | Skills do not alter ChatGPT Enterprise privacy terms: organizational data and skills remain strictly excluded from model training. |
| **How do skills affect credit consumption?** | Running complex multi-step skills frequently triggers deep GPT-5.6 reasoning passes, consuming credits proportional to task complexity. |
| **What policies apply during external API calls?** | A skill can only invoke external tools and connectors that the executing user is already authorized to access within the organization. |

## 9. Departmental Use Cases in Practice

### 9.1. Sales and Revenue Operations
Customer signals are often fragmented across CRM logs, call transcripts, email threads, and Slack discussions. ChatGPT Work consolidates these disparate streams into actionable sales collateral: account briefings, stalled opportunity diagnostics, or personalized proposals.

```text
Create an automated workflow that runs every weekday at 08:30 AM to inspect all new CRM leads,
identify accounts without follow-up activity for over 48 hours, and post the alert list to my notifications channel.
```

### 9.2. Corporate Financial Planning and Analysis
Finance teams can leverage historical expenditure workbooks, close reports, and department lead notes. The agent handles tedious consolidation, enabling analysts to focus on interpreting variance narratives and financial forecasting.

```text
Reconcile actual departmental expenditures against budgeted figures in the connected workbook.
Identify line items exceeding budget by more than 10%, extract key cost drivers,
and draft an executive memo for the Chief Financial Officer.
```

### 9.3. Cross-Functional Operations Management
The agent excels at multi-stakeholder operational projects that require harmonizing calendar entries, team messages, and open issues into unified status digests or new operational standard operating procedures.

## 10. Security, Governance, and Enterprise Compliance

ChatGPT Work inherits the proven security and compliance architecture developed for ChatGPT Enterprise, ensuring robust institutional governance.

> [!IMPORTANT]
> **Auditability & Compliance API:** All agent reasoning steps, plugin interactions, and file operations are recorded within the enterprise audit log via the Compliance API. Workspaces with Enterprise Key Management (EKM) maintain full control over cryptographic key access.

### 10.1. Centralized Administrative Controls
Administrators can:
- Define allowable plugins and external enterprise data connectors.
- Restrict agent network access and browser capabilities within sensitive subnets.
- Mandate human authorization gates for operations interacting with protected business data.

### 10.2. Sandboxed Desktop Security
The desktop client isolates agent execution: filesystem access is confined strictly to designated directories, and sensitive system commands undergo safety model verification before execution.

## 11. Availability and Pricing Plans

ChatGPT Work is supported across all major operating environments with flexible quota management tailored to enterprise and professional tiers.

### 11.1. Plan Matrix and Capabilities

| Tier | ChatGPT Work Access | Codex Support | Quota & Expenditure Control |
| :--- | :--- | :--- | :--- |
| **Free** | Limited introductory access | Desktop app (basic) | Standard daily rate limits |
| **Plus** | Full access in web and desktop clients | Full access | Generous monthly task allocation |
| **Business / Pro** | Priority access powered by GPT-5.6 | Uncapped desktop client access | High-speed processing with flexible task pools |
| **Enterprise / Edu** | Full security perimeter (EKM, Compliance API) | Deep enterprise repository integration | Centralized admin console management by department |

### 11.2. Resource Allocation and Spending Governance
Because Work executes intensive multi-step operations, consumption is metered by task complexity rather than simple message counts. Administrators can configure team spend limits, assign elevated quotas to high-demand roles, and review ad-hoc capacity requests through the centralized administration console.