4 minute read

Extreme ideation and AI protoyping

Home » Blog » Extreme ideation and AI protoyping
Home » Blog » Extreme ideation and AI protoyping

Authored by Ian Gotts and Brooke Mohnkern

Imagine this.

You run a restaurant. One of your team sees AI as an opportunity to offer a personalized Michelin-starred quality of service. And what if you were able to copy and offer the best of the best within a day? Granted, you cannot provide the personalized service of a waiter in white gloves standing at your table waiting for every one of your diner’s demands. But what if you could offer their approach and knowledge to make the your dining experience significantly better?

So the team members builds a prototype of your Expert Waiter agent in less than an hour. The new mantra is:

Don’t tell me about your idea. Show me!!

Enter the Expert Waiter agent

The diner scans a QR code on the table and it brings up an agent. They can talk or chat to the agent. It knows about every dish and how it is prepared. It knows the best wine pairing for dishes. It knows the price of everything. And it is constantly up to date, so if dishes are not available they are not offered. 

They can have a conversation about what they feel like eating and it can give recommendations. But part of the experience is how it talks to them. How it is empathetic to their tone and matches it. Their answers are single-word or abrupt. Clearly in a rush. Let’s get on with this.  Can’t decide and want to chat through the options. Take all the time in the world. Ask as many questions as you want. 

And when the diner is ready, it will take your order and send it to the kitchen. This is clearly not the same as a real waiter, but it is affordable for every restaurant, bar, cafe, and take-out. It will raise the quality of the lowest to close to the highest. Want to emulate the ambiance or feel of the Four Seasons or Michelin-starred restaurant? Get AI to listen to several videos (https://www.youtube.com/watch?v=CD_b8bZxk6g)  where they talk about the service and the way the waiters describe the dishes, and get AI to write a prompt so the agent mimics this behavior. 

 Every day, the menu changes, the drink options change, and the stories change. Normally your waiters who have to deliver those stories need to memorize them. In the best restaurants they have staff who adapt all the time to this. But your agent can do this effortlessly.

These agents are aimed at all but the very exclusive restaurants where having a white-gloved waiter waiting at your shoulder for your every whim is part of the experience. For everyone else the benefits are:

  • Personalized service: so will tailor language to diner’s demeanor and needs
  • Unlimited resources: can deal with an infinite number of diners at once
  • Infinitely patient: diners can take as long as they want
  • Constantly up to date: it knows about every dish and availability
  • Multi-lingual: can deal with diner in their language
  • Lower cost and scaling: Scales as demand grows, so no need to have waiters hanging around looking at empty tables, or disgruntled diners when there is a rush

Rapid prototyping and AI coaching

Best of all, in less than an hour you can build any prototype agent and start testing the behavior and user experience. You don’t need to build the full agent and connect it to all your “knowledge”.  Then prototype and imitate the experience and make up dummy data. Building the technology is arguably the easy part.  Getting the “feel” right is way harder.  So get AI to coach you on how to improve it. 

Sounds amazing. Can’t wait for it.

You don’t need to.  We’re already doing it.

We built the “Expert Waiter” prototype as an OpenAI CustomGPT. We could ask the GPT when in “role play mode” to make up data or we could ground it with some data in a document. Let’s face it, LLMs are pretty good at making stuff up. BTW Have you tried voice mode?  It is impressive. We then asked it to role play. When we weren’t happy with its result, we asked it to stop role-playing and tell us why it behaved that way and which instructions to fix. So it rewrote the instructions. We incorporated the changes into the process diagram, re-generated the instructions and pasted them into our prototype. Rapid iteration with governance.

Here’s an audio of me chatting to the Expert Waiter.

Below is the process diagram in Elements.cloud that generated the complete Expert Waiter agent instructions and guardrails. This is why we could build it so quickly. Here is a Solution Guide on how to build agents using a process-led approach. LINK

Cover image Photo by Ibrahim Boran on Unsplash

Back to News