A prototype from a BloomTech AI student: This multi-agentic system writes React components in their entirety (within the context of the company code base.) Generates fully-formed React/JavaScript + CSS + unit tests + Documentation, completely automatically.

In the past few weeks the company used it to generate 290 unique React components. 22 of the 290 (7.6%) were rejected (or recreated) for bad structure, bad prompt, or overly complicated code. After including time for touch-up, the company estimates it saved 205 hours of developer time so far. At a mean cost of $.57/component.

By the way, this methodology works on all aspects of a code base, as well as on aspects of products that aren’t code at all. To learn how:

Brilliant to use a vector to store acceptable models to feed the prompt pre-execution. Never would have thought to use that!

All these models will disappear and eaten up alive by big-techs. Look what happened to all these AI startups.

These are big tech models…

why is this called multi agentic? is the bar for an agent, a python script which calls openai apis?

Huh? He used RAG to build a vector data store of various aspects of the code base and docs, and the data is passing between multiple fine-tuned agents.

very cool!