8 Insane ChatGPT Work Use Cases! (automate anything)

Summary of Technological Concepts & Product Features (ChatGPT Work)

The video demos ChatGPT Work, an AI agent running in a dedicated desktop app (not the web version). Key capabilities highlighted:

• Desktop-based agent that can see/access local files on your machine.
• Includes a built-in browser to navigate websites and perform actions (with tool/browse skills).
• Supports skills/plugins, such as:
  • custom YouTube skills,
  • data analytics,
  • spreadsheets,
  • data warehouse connectors.
• Can take action in tools you use daily and run long tasks.
• Offers Remote control from the phone/cloud to monitor and continue tasks started on the desktop.

---

Use Cases Showcased

1. Interactive Reporting Dashboards (YouTube Analytics)

Workflow:

1. Start from an inspiration dashboard screenshot.
2. Paste the screenshot into ChatGPT Work.
3. Ask it to recreate it as a shareable live site.

Uses:

• A custom skill to connect to the YouTube channel and pull analytics.
• A data analytics plugin to query BigQuery (data warehouse).
• A sites plugin to build an interactive dashboard.

Iteration:

• The presenter tweaks interactivity (e.g., hover tooltips showing numbers).
• Updates arrive in minutes (reported as ~4 minutes).

Claimed benefit:

• Much faster than traditional BI request backlogs (minutes vs. weeks).

---

2. Web Automation Agent (Download and organize assets)

Workflow:

• Provide a link to a blog listing.
• The agent uses its own browser to open pages and download the first 20 feature images.

Output:

• A folder of downloaded images.
• A CSV mapping: blog article title → corresponding image.

Emphasis:

• Useful when tools lack APIs/connectors.
• Example mentioned: schedule an agent to log into invoice platforms and download invoices monthly.

---

3. Excel Cohort Analysis Model Builder

Inputs:

• Multiple exports (about ~35MB) from different tools, such as:
  • community stats,
  • event platform activity,
  • Stripe payments,
  • legacy system export.

Prompt requirements:

• Must generate an Excel workbook using formulas only (no hard-coded numbers).

Uses:

• Spreadsheets plugin (supports templates and emphasizes formula traceability).

Results:

• Agent takes ~1 hour and produces a workbook with:
  • raw data tabs,
  • cohort retention analysis,
  • Excel-editable graphs,
  • split retention by acquisition source,
  • findings/recommendations.

Note:

• Desktop app can handle larger datasets than web (web likely errors due to size).

---

4. Lead Generation / Lead Research Agent

How it works:

• Uses a custom “lead research” skill to find companies matching criteria, including:
  • region (Europe; separate US search implied),
  • hiring signals (AI automation / ops transformation),
  • announced AI initiative,
  • company size (50–500),
  • excludes companies already known in the “lead research project folder”.

Execution setup:

• Uses higher “effort” level: ultra.
• Spawns multiple sub-agents concurrently (e.g., Berlin/Hamburg, Munich/Vienna/Zurich, UK/Ireland).

Output:

• An Excel file with 25 companies.
• Direct URLs for verification.
• Potential decision-maker names/titles.
• Confidence ratings and summaries.
• Finds LinkedIn profiles for outreach.

---

5. Receipt Expense Processing Automation

Setup:

• Photos of receipts are placed into a local receipts folder.

Agent behavior:

• Moves images to processed automatically.
• Sorts by receipt date.
• Extracts key receipt fields.
• Appends categorized expense rows into a receipt log Excel file.

Demo note:

• Shown as a manual run, though designed for scheduling.

---

6. Bulk Editing YouTube Descriptions (Context-aware link updates)

Problem:

• Adding a new tracked short-link to 130+ video descriptions is too time-consuming and error-prone.

Approach:

• Agent connects to YouTube via custom skill.
• Step 1: Downloads and saves existing descriptions (backup in markdown/structured files for revert).
• Step 2: Live-edits each video’s description using rules:
  • If the description begins with a lead magnet link, insert the CTA below it.
  • If it has no CTA at the top, insert the CTA as the first line.
  • If it starts with a subscribe CTA, replace it with the training CTA.

Links:

• Uses Short.io via custom skill to generate dynamic short links with tracking parameters, using video ID as a dynamic field.

Output/verification:

• Shows updated descriptions.
• Confirms link placement for a sample video.

---

7. Image Generation Agent (Style replication + prompt capture)

Technique:

• Agent downloads and analyzes existing blog feature images (e.g., 20 examples) to infer:
  • format,
  • style,
  • resolution.

Task:

• Generate multiple feature image suggestions for new blog posts (3 articles).

Behavior:

• Produces several variations per article (9 total suggestions in the demo).
• Saves outputs in a folder.
• Outputs the exact generation prompts to a markdown file for review/audit.

Benefit:

• Integrates creative generation into workflows.
• Supports “try multiple variations.”

---

8. Remote Control for Long-Running Jobs (Phone + cloud)

Feature:

• “Remote” in the mobile app (with Face ID).

Purpose:

• Monitor tasks started on desktop (e.g., long edits, long analysis).
• Check completion while away.
• Trigger additional actions (example: “run it for all videos from 2025”).

Constraint noted:

• The desktop machine must remain on for local file access, but the session can be managed remotely.

---

Main Speakers / Sources

• Yann (narrator/presenter) — referenced by name (“Yann, we already know…”), presenting the use cases.
• OpenAI — source of the ChatGPT Work product referenced throughout the demo.