Automating post-workshop tasks with Cursor and AI (no dev experience required)

After a UX workshop, there’s a lot of information to process - audio, post-its, scribbles, photos. Documenting it all takes time. Doing it properly takes even more. I wanted a way to speed that up without compromising quality.

Earlier in my career I was a developer. These days I focus on UX, not writing code, but I still know how a few well-placed scripts can take the grunt work off my plate so I can focus on what matters.

The goal

Take a folder of raw workshop output - audio recordings, photos of post-its or cards - and get back usable, structured artefacts I can work with.

Automations I’ve played with so far:

• Audio transcription using different services (OpenAI Whisper, Assembly, etc.)
• Image-to-text conversion with OCR libraries and APIs to pull out content from handwritten post-its
• Mermaid syntax generation so I can recreate structures like user flows and journey maps directly in digital tools

How I built it

I used Cursor, an AI-assisted coding environment that lets you describe what you want, and then generates and debugs the code for you.

I don’t write Python. I never have. But I can describe a task clearly, and I understand the flow of data from input to output. That was enough. Cursor filled in the rest.

Each script started with a natural language prompt. I’d test the output with real workshop data, then refine based on the results. Within minutes I had working scripts, within a couple of hours, I had working tools that could save me hours of manual effort.

An example prompt

Write a Python script that processes a workshop photo containing multiple coloured cards (e.g. Post-its) arranged on a whiteboard. The script should:

1. Detect and isolate each individual card in the image
2. Extract the dominant colour of each card
3. Extract any handwritten or printed text from each card using OCR
4. Output a structured list of dictionaries where each dictionary includes:
   • the card’s text
   • the card’s dominant colour (in RGB or HEX)
   • the bounding box coordinates

What’s worked, what hasn’t

The transcription workflow has had the biggest impact so far. I can now revisit full conversations instead of relying solely on post-its or my memory. It gives me a richer picture of what was said and makes it easier to cross-check and spot gaps.

OCR has been hit-and-miss. Handwriting recognition, especially using open-source tools, is fragile. Colour-based card detection also needs refinement - lighting, card type, and pen choice all impact accuracy. I’ve started playing around with the offline factors in workshops (e.g. brighter cards, thick pens, good lighting) to improve results before anything hits a script.

Worth a try?

For anyone with a technical background who’s moved into UX or strategy: this kind of automation is within reach. Tools like Cursor bridge the gap. You don’t need to get back into spending hours on Stack Overflow. To be honest, even if you don’t have a technical background, I’d encourage you to have a go.

You just need a clear goal, a methodical approach, and a willingness to test and refine.