One of the biggest “a-has” for me when generating PowerPoint slides with AI tools was the simple thought: almost all software is code, including PowerPoint. Think about it — a PowerPoint is just a map of visual and text-based objects within a confined physical canvas. Every single position, color, font, and graphic can be turned into code. And now with AI tools, you can command that code with text prompts.

Many people don’t realize that there are many open-source and free tools available for doing this already. They’ve mostly been used by developers, but now with agentic coding platforms like Codex and Claude Code/Cowork, almost anyone can access them.

Here are a few of them:

Pro tip: Feed this post to your favorite AI tool and have it give you ideas of what you can do with it.

Python: python-pptx

python-pptx is the most widely used library for creating and editing PowerPoint files in Python. Steve Canny created it and has maintained it actively since 2013. It works without PowerPoint or Office installed. The library reads and writes the underlying .pptx XML directly.

A minimal example:

from pptx import Presentation

prs = Presentation("template.pptx")
slide = prs.slides.add_slide(prs.slide_layouts[1])
slide.shapes.title.text = "Q3 revenue by segment"
slide.placeholders[1].text = "Software grew 12% YoY"
prs.save("output.pptx")

That is the entire pattern: open a template, add slides, fill placeholders, save. From there you can iterate. The same script can produce one deck or a hundred.

What it does well:

  • Reads and writes .pptx files
  • Adds text, tables, images, charts, and basic shapes
  • Modifies existing decks (handy for batch updates)
  • Works with corporate templates as-is

What it doesn’t do:

  • Render slides to images or PDF (you need LibreOffice or Office for that)
  • Support every PowerPoint feature. Some chart types, SmartArt, and animations aren’t available
  • Replace design judgment

For most deck-automation work, python-pptx is enough. It forces you to design a clean template upstream, which is the right place to do that work anyway.

Other Python options

If you need to convert .pptx to PDF or images without Office installed, run LibreOffice in headless mode from the command line:

libreoffice --headless --convert-to pdf deck.pptx

You can also drive LibreOffice from Python via the UNO bridge, though the API is famously awkward. Most people use python-pptx for the slide work and LibreOffice only for conversion.

Aspose.Slides for Python is a commercial library that does what python-pptx does plus the things python-pptx can’t: rendering, format conversion, full chart support. Free trial, then paid. Worth knowing about if you hit python-pptx limits and need a fix without standing up Office.

JavaScript: PptxGenJS

PptxGenJS is the JavaScript equivalent of python-pptx. It runs in the browser or in Node. The browser case matters: if you have a web app where the user clicks a button and downloads a generated deck, PptxGenJS produces the file in the browser without a server round-trip.

The API is similar in shape to python-pptx: open a file, add slides, save. The syntax is JavaScript-flavored. Active project, well-documented, free.

Pick PptxGenJS over python-pptx when the generation has to happen in the browser, or when your stack is already JavaScript. Stick with python-pptx when you’re doing batch work from data pipelines and your data lives in Python.

Java: Apache POI

Apache POI is the Java workhorse for Microsoft Office files. The XSLF subproject handles .pptx. HSLF handles the older binary .ppt format.

POI is verbose. A snippet to create a slide with a title is twenty lines of Java where python-pptx would be two. But it covers more of the file spec than any other open library, and if you work in a Java enterprise shop, it’s already on the approved-vendor list.

For new projects, there’s no real reason to choose POI over python-pptx unless you’re stuck inside the JVM.

Markdown-first tools

The libraries above let you build slides as code. A different family of tools lets you write slides as markdown. The tradeoff is straightforward. You give up fine-grained control over layout, but you write decks the way you write notes.

Pandoc is the universal document converter. pandoc deck.md -o deck.pptx produces a real PowerPoint file from a markdown source. Layouts are limited. You get title slides, content slides, and not much else. For a quick deck where the content matters more than the design, that works.

Quarto is Pandoc’s better-organized successor, built by Posit (the company formerly known as RStudio). It supports Python, R, and Julia code blocks that execute and embed their output into the deck. Good for analyst presentations where the chart needs to update when the data does. Outputs to .pptx, revealjs, beamer, and PDF.

Marp is markdown to slides with output to HTML, PDF, or .pptx. Simpler than Quarto and faster to learn. Best for technical talks where you’d otherwise be using a slide tool you don’t enjoy.

The pattern is the same across all three. Write the content in markdown, let the tool render the visuals. If you’re a developer giving a talk, this is usually the right path. If you’re a consultant building client decks, the limited layout control will frustrate you.

Even if you want to go directly to slides, I recommend asking your agent or AI tool to give you a quick text-based overview of each slide: title, content, positioning, and the overall message it’s trying to convey.

Putting it in markdown is useful because it’s very compatible with most AI tooling.

HTML-based slides

A different family entirely. Instead of a .pptx file, the deck is a web page.

reveal.js is the granddaddy. HTML, CSS, and JavaScript control everything. It scales well for technical content and looks clean. Slides live in a Git repo and version like code.

Slidev is the Vue-based modern alternative. Better tooling, hot reload, code-friendly. Designed for developers presenting to developers.

Spectacle is the same idea built on React.

If your audience expects a .pptx attached to an email, none of these matter. You can convert at the end, but you lose most of the visual polish in the conversion. If your audience is technical and the deck can live on the web, these tools are usually better than PowerPoint.

I suspect that creating HTML websites on the fly is going to become increasingly popular.

API-based: Google Slides and Microsoft Graph

If your org runs on Google Workspace, the Google Slides API lets you create and modify presentations directly in Drive. Same idea for Office 365 via the Microsoft Graph API, though Graph’s PowerPoint support is more limited than the Slides API.

Templates are still hit or miss

The library’s strength is filling in templates from data. Its weakness is anything that requires aesthetic judgment. I highly suggest tinkering with a master template you want it to use before asking for output.

Iterating with an agent on design is just not great right now (trust me, I’ve tried).

A good template has defined layouts (title, content, two-column), a consistent color palette set in theme colors, placeholders sized and positioned for the kinds of content the script will produce, and a few master slides that handle the 80% case. Do this once and your automation script becomes thirty lines of code. Skip it and you’ll spend a week fighting font sizes.

For the layouts themselves, see the fourteen slide layouts consulting firms use for everything. For thinking about what to put on each slide, the SCQA and Pyramid Principle approach still applies whether you build the deck by hand or by script.

Want more guidance?

A lot of what makes a consulting deck good is upstream of the tools. It’s the structure of the argument, the synthesis of the data, and the choice of what to put on the slide. The libraries above solve the production problem. They don’t solve the thinking problem.

But AI can be a great sparring and feedback partner. To make them more effective, I’ve created custom skills which guide you through structured processes to drive toward high-quality content. I keep updated based on model releases and my own attempts to get better at working with AI.

AI skills can be a way for you to get in the loop of problem solving and iteration more quickly than you can on your own. I’ve built these based on my course content from Think Like A Strategy Consultant that are now available.

Through trial and error, I landed on four component skills:

  • Problem Framer turns a messy situation into a structured problem
  • Structure & Synthesize takes scattered notes, research, or raw data and turns them into an insight or argument
  • Strategy Communicator presents clear thinking in structured ways that align with the audience
  • Strategy Slides handles slide design and PPTX creation based on the principles in the course

And two overall skills you can use more loosely or to build your own processes:

  • Strategy Coach bundles all four component skills and routes you through the full problem-solving process
  • Strategy Writing strips out AI patterns, forces specificity, and gets prose to read simply and clearly

These skills work with any AI chat or agentic coding app. You can use them three ways.

As a coach when you want to learn the frameworks. The AI walks you through SCQA, MECE, or the Pyramid Principle, asks the right questions, and helps you build the skill yourself.

As a sparring partner when you’ve done the work and want feedback. Paste in your draft, your issue tree, or your slide outline, and have it pressure-test the logic. Where’s the weak branch? What’s missing? What would a skeptical executive push back on?

To do the work for you when you give it enough context. Hand it the situation and let it produce the SCQA, the issue tree, the synthesis, or the deck outline. You review and edit instead of starting from scratch.

You can grab them at strategyu.co/strategy-consulting-aiskills:

Or just bootstrap your own

Or skip all of that. Point Claude (or whatever AI tool you use) at strategyu.co and give it this prompt:

Scrape strategyu.co and help me build custom tools and processes to build McKinsey-style PowerPoint slides. Use resources on the site and from lectures on YouTube. Also consider the free tools and resources at https://strategyu.co/free-tools-to-build-powerpoint-slides-with-ai/. Ask me what formats and style I'm aiming for, then walk me through the process.