The presentation folder contains a Jupyter notebook that is a demonstration of a DisC Agent.
What if software could be built by typing up the requirements without writing any code? This talk introduces a new paradigm in software development that redefines the relationship between natural language and application creation. Moving beyond the limitations of traditional No-Code platforms, we explore a revolutionary approach: Documentation is Code (DisC). By leveraging large language models, DisC enables written specifications to be transformed directly into working solutions—whether through AI-driven multimodal workflows, traditional software, or hybrid applications.
DisC takes to an extreme the promise of Agentic AI. It decreases time from ideation to execution and dramatically reduces cost. However, acknowledging limitations of AI agents and DisC is paramount to maximizing the utility when using them. Some constraints, such as runtime performance, are inherent, while others have theoretical and practical workarounds.
The session features a live demonstration showcasing how an existing implementation can be transformed to a DisC solution by using a prebuilt ReAct agent from LangGraph. Instead of generating code, the Reason–Act–Observe loop replaces traditional algorithms and data structures reshaping the way we think about software engineering. Instead of the agent generating code, the agent replaces code. We will discuss challenges such as ensuring consistent outputs, enabling iterative development, and agentic tooling.
Despite these challenges, DisC paves the way for billions of people to build billions of solutions—without ever writing a single line of code. Ultimately, this innovative approach brings us to a future where product descriptions become live applications.