A weird thing about AI tools is that the best indication something's working isn't the fancy metrics or user stats - it's when using your own tool starts driving you mad.
Huh?
After creating our video script prompt in Part 2, you might find yourself copying and pasting it into ChatGPT twenty times a day.
You’ve built a basic tool and it does the job. Well in fact. But having to load up ChatGPT and copy and paste in news articles manually is now annoying.
It’s only annoying because the tool works so well. And you are now hitting up against limitations of HOW we are using AI - we’re still stuck at Levels 1 and 2.
Sound familiar? That's when you know you're ready for Level 3 - turning your prompt into a proper AI assistant.
Why (and when) to build an assistant
Converting prompts to assistants
The art of simple interfaces
Testing and refining your assistant
Common mistakes to avoid
Prompts are brilliant. Hell, they are my bread and butter. People pay me $4000/hour to go and teach about writing good prompts.
They're quick, flexible, and perfect for testing ideas. But they have limitations:
Repetition: Copy-pasting the same prompt gets old fast
User Experience: Teaching others (ie, your VA) to use your prompt correctly is painful
Consistency: Even small changes to prompt format can affect results
Accessibility: Not everyone understands prompt engineering and what to do if it “goes wrong”.
This is where assistants shine. They're like putting your prompt into a nice, friendly package that anyone can use. We are wrapping up a prompt -hence the name that is often (wrongly) pejoratively used - “ChatGPT wrappers”.
An assistant is a nicely packaged set of prompts and knowledge that, and this is important, anyone can use. Not just us.
Before we dive into building, let's check if you're ready for Level 3. You might be ready if:
Your basic prompt is working reliably
You're using it frequently (10+ times a week)
Other people want to use your tool
You want to sell access or use it to collect emails
But stay at Level 1-2 if:
Your prompt still needs lots of tweaking
You're the only user and it's occasional use
You're still figuring out the best format
You need maximum flexibility for testing
Don’t cement into an assistant what isn’t worth cementing!!
Here's where Level 3 gets interesting - we're no longer just building for ourselves. Assistants open up new possibilities for sharing and, if you want, monetisation. You can charge for access - this is the beginning of turning it into a product.
You could sell at Level 1 and 2:
Level 1 prompts? Sure, you can share or sell prompt libraries, but it's a crowded space with (let’s be honest) limited value. This is why I’ve never sold “prompt packs” - it’s not terribly valuable.
Level 2 projects? Great for team collaboration if you're all on the same platform, but not easily shareable publicly. And forget about adding a paywall - I’ve tried and goodness me it’s a faff!
But Level 3 assistants? Now we're talking about something deployable. Something that can:
Be shared easily with anyone
Include proper access controls (ie. users make accounts)
Have a payment gateway (if that's your plan)
Collect emails for lead generation
Track usage and gather feedback
etc.
We're shifting from "tool" to "product". Not that everything needs to be a product - this is one of my bad habits! I try to make everything I do into a product or service. Plenty of assistants should stay as internal tools!
But if you're looking to expand your reach, collect leads or build a business, this is where things get interesting.
For Level 3, I’d suggest using Launch Lemonade - and there's a good reason why. Think of it as a way to package everything we've built so far into something others can easily use, while extending what's possible.
The technical learning curve is basically zero. Anyone can build.
The real power of Launch Lemonade comes from two things:
Multi-modal- we can use different AI models for different tasks
Web scraping capability - automatically pulls content from your website or knowledge store
Deployment ability - easy to add a paywall, user access, publish as a page or even embed on your own website
I’ve written 4+ Playbooks on building an AI assistant using Launch Lemonade. I’m also kicking off a free accelerator in early February to get you started so won’t go into the details here. If you want to join the free live accelerator here’s the waitlist.
Here's where most people go wrong - they try to add too much.
AI assistants are powerful. So it’s tempting to try to get it to do more.
Remember, we're trying to make this simpler than using raw prompts, not more complicated!
Just because it can do more doesn’t mean it needs to!
For our script assistant example, we really only need:
A clear welcome message explaining what it does
A way to input the article
Basic formatting options (length, style)
The output script
That's it! Don't add bells and whistles just because you can. (Not yet anyway!)
In Part 4, we'll look at taking your assistant to the next level - building it into a standalone application. But for now, focus on creating a simple, reliable assistant that makes your life easier. We can use this assistant to prove market demand before rushing off to build a full app!
And again - if you want to hop on our free AI builders’ accelerator here’s the waitlist.
Keep Prompting,
Kyle