Copilot as Controller

What’s possible when you make your AI engine in charge of all of your business logic? It turns out it can think creatively about solutions that you might not have thought of. This demo is of a copilot architecture I’m calling multi-brain - a kind of mixture of experts that can run complex planning and development tasks by generating requirements and ensuring the implementation conforms to them by noticing conflicts and suggesting changes. All UI inputs are sent to the copilot and all changes to the data model are made by the copilot, allowing for generative UI and non-web app interfaces like slack or phone. I’ll demo it using Zucca.ai - PSL’s soon to be latest spin out.

Architecture details-
Demo will be creating the requirements, formula, manufacturing process, equipment list, labor plan, and calculate cogs for a 100,000 unit run of a new food product designed on the spot. Takes about 2 minutes. Then I’ll go over the components:

• Copilot AI engine that sits between UI and data model
• Engine is made of a master control router that decides what parts of the expert system to call
• On-the-fly brain building from partials
• use of code interpreter to write code to run unit economics analysis and COGS
• labeled code blocks used to stream UI elements using
• undo-stack built-in because all changes are tied to messages