Is prompt writing really that hard?
Well, that depends. Seriously, it depends. I’m not just being a consultant, I promise.
Let’s back up a little.
Since we know my day-to-day is in Power Platform, let’s start there.
My recent talk at PPCC25 was about using Copilot to accelerate your data modeling buildout. I mean seriously, did anyone ever actually like the tediousness of manual table creation? Click here, type this, add that, change the thing over here.
With Copilot assisting, we can negotiate our data model and just let it be built in the background while we move on to the next task, or grab a coffee, or if it’s a huge data model, take the dog for a walk. And then tadah, it’s done.
Back to prompting.
For my session at the conference we used a scenario for a gated community and how to improve their management of gate traffic.
As long as we give at least some context for our business use case, the LLM does its thing and make a pretty good data model for us. Each of these prompts actually returns darn near identical results.
So I’m trying to build something for gates and cars and people and like tracking stuff? Can you make a thing for that? Make it smart, k?
I need to track residents, vehicles, gate access events, and security personnel.
Create a data model with Residents, Vehicles (linked to Residents), Gates, Access Logs (timestamp, vehicle, gate, granted yes/no), and optional Security Staff.
Even this one full of typos gives pretty good returns
I need a data modle for a gated comunity app. It shoud track residnets and their vehicals. Each vechile can be loged enterring or exiting a gate. Gates have names and loctions. Secuirty staff may be assinged to log evnets. I want to know when acces was granted, wich vehical it was, and wich gate.
These are all great starting points. You can also give a serialized, detailed list like this:
Create the following tables and appropriate relationships:
Resident table- Name, specific address details, contact information, notification preferences, vehicles owned/leased
Vendor table- access duration, name, vehicles, relationship to resident
Vehicle table- make, model, year, color, plate, type (resident or vendor)
Access log table- accessed by (vendor or resident), vehicle information, in or out, timestamps, logged by staff or automated
Where you get further, is when you combine the LLM smarts, with your explicit instructions.
Let’s combine the LLM-dependent prompt, with our explicit serialized list.
I need a data modle for a gated comunity app. It shoud track residnets and their vehicals. Each vechile can be loged enterring or exiting a gate. Gates have names and loctions. Secuirty staff may be assinged to log evnets. I want to know when acces was granted, wich vehical it was, and wich gate.
Create the following tables and appropriate relationships:
Resident table- Name, specific address details, contact information, notification preferences, vehicles owned/leased
Vendor table- access duration, name, vehicles, relationship to resident
Vehicle table- make, model, year, color, plate, type (resident or vendor)
Access log table- accessed by (vendor or resident), vehicle information, in or out, timestamps, logged by staff or automated
This gives me the best results yet.
We hear all the time about the need for human oversight with AI, and I think this is a great use case for that. Give your Copilot the right context, and your list of tables and columns. This reduces the time to build tremendously.
Bad prompts offer conflicting information. When you do that, then Copilot has to decide what you REALLY mean. And while I love me some good Copilot help, I’m not sure I want it to make those decisions on my behalf.
Prompts aren’t poetry; they’re specs. Start messy if you must, but add context and a tidy list and you’ll get 80–90% of the way there on the first pass. The magic isn’t the wording—it’s the clarity. Let Copilot do the clicking while you do the thinking.
(Yes, I left the typo in the AI-generated image for so many reasons)
Leave a Reply