Gabriella here again from the product team Assist is live — and now it’s your turn to play.
We’re kicking off the Prompt Pro Challenge: a community thread for sharing your best (or most unexpected) Assist prompts. Quick wins, creative hacks, or hilarious fails… it all counts.
How to participate:
Comment with the prompt you used
Share what Assist gave you back (screenshot encouraged!)
Bonus: Did it save you time or unblock you?
Need inspiration? Try things like:
Styling apps → e.g. “Make this look like Netflix”
Editing → e.g. “Add a column”
Every post earns you the Prompt Pro badge Standout prompts may also get featured in our new Builder Digest newsletter
Note: Assist is still evolving, so some results will be great, some will flop. Both are useful — and fun to share.
I wanted a tool to keep track of all the fantastic demos our EPD team presents every week.
This is a tool that probably would have taken 2-3 hours to build myself with all of the various and polish and things I got automatically thanks to the AI assistance, but all in all, it took me 21 minutes and 57 seconds (70% generated app and probably 30% manual tweaks).
It allows me to paste in a transcript from our team Zoom call and uses Claude to parse out each specific demo.
Here's the prompt I used:
I want to build an EPD demo tracker. This will allow me to keep track of
all the various demos that EPD has put on.
Use the epd_demos table in Retool DB to source your data.
UI
- A table for all of our data, with the ability to search by presenter
name or by fuzzy search in title, description or notes. When
one of these inputs is populated, it should filter down the table.
- An add modal with a large text area to paste a transcript of a demo
call as well as an input for the recording link and for the
date of the recording
- An edit modal that allows the user to edit anything about
a particular demo. This modal is triggered by a row action
on a particular row of the table.
Screenshots attached for how it turned out! It's been working great!
I've found a few things here and there that I didn't capture in the initial prompt that I've needed to modify, but honestly I probably wouldn't have built this at all if I had to build it completely on my own. I find I'm much more likely to start building apps if I know I don't have to start from a completely blank slate.
+1 to Keanan's demo tracker — such a cool use case
I challenge you to give Keanan’s prompt a try and let us know how it goes. Did it work the same way for you? Did you have to tweak it? Always interesting to see how Assist handles the same request in different hands.
Brand new badge alert!
We just dropped the Prompt Pro badge... and you’ll be the very first to earn it! All you have to do is share a prompt here (whether it’s a quick one-liner, a win, or even a weird result).
I just started exploring with Assist. I asked "Hi Assist. I need to build [redacted....but extremely vague ask] ...information, most of which is from our database. Do I need to have an existing query for the table that we want to build or do you help me create that query here? It did answer my question, then it proceeded unprompted to build an entire app on its own making many assumptions. So...A for effort - though seeking more clarity on goals or awaiting more specific asks / instructions might be helpful. [Update] I am now hand-holding it through more detailed asks and so far so good.
I wanted to share something I have been adding to my prompts that has been quite helpful. Often times I miss providing details that are important for the agent to really be helpful. Asking the model to ask questions is something I started doing a long time back when Cursor was first released and I find it really reduces the back and forth and often makes the models first attempt much stronger.
The following is just an example you can add. For particularly complicated changes I might also mention something about "interviewing" or "interrogating" me to learn requirements, but I find this results in a bit of a verbose output for regular use.
Ask for clarification when there is ambiguity. DO NOT make assumptions. Before starting work or making changes ask questions to ensure you understand the requirements and intended output.
@kimweller I think you may find this helpful as you mentioned needing to hand hold the model. Adding something like this lets the llm hand hold you, while still getting good results!