ChatGPT’s Agent can now make PowerPoint slides. But …
It uses a convoluted JavaScript-based process with the pptxgen library:
The Process:
1Script Generation: Wrote/modified a Node.js script (answer.js) using the pptxgen library
2Template Reference: Used slides_template.js as a base template with pre-defined constants (fonts, colors, margins)
3Manual Slide Construction: Built each slide programmatically by:
◦Adding shapes (addShape)
◦Adding text boxes (addText)
◦Embedding images (addImage)
◦Positioning everything with exact coordinates (x, y, width, height in inches)
4Icon Handling: Converted FontAwesome SVG icons to base64 data URIs
5Conversion Pipeline:
◦Run node answer.js to generate answer.pptx
◦Run python3 pptx_to_img.py to convert slides to PNG images
◦Review images in browser
◦Find spacing/overflow issues
◦Edit the JavaScript code
◦Repeat (it did this cycle 10 times!)
This Is Absurd:
•It's like building a house by writing assembly instructions for robots instead of using blueprints
•Every tiny adjustment required editing code, regenerating, and re-converting
•The agent spent more time debugging pixel-perfect positioning than thinking about content
•For a task about rewriting content, it chose the most technically complex approach possible
This is a perfect example of choosing the wrong tool - instead of focusing on content transformation, it got lost in the weeds of programmatic slide generation.