This is the home page of Rhetorical AI (https://rhetorical-ai.com) From Persona and Intention to Requirements Creation and Clarification. Below is the introduction.
Origin of Requirements
Requirement Engineering is a structured approach to developing software. Without having good requirements, all subsequent activities are meaningless. This is explained by the Infinite Monkey Theorem – 10,000 monkeys sitting in front of typewriters could not create the complete Shakespeares work even until the end of time. If you don’t believe me, try the Infinite Monkey Simulators at https://tennessine.co.uk/monkey or https://codepen.io/justinchan/full/krOLzV
Hence having good requirements is absolutely critical to all software construction projects.
Without boring you with Blaise Pascal (1623), Charles Babbage (1791) and Ada Lovelace (1815), one could be certain that effective strategies for managing requirements already exist nowadays. Otherwise, we would not have working applications and software.
A generally accepted approach is TOGAF, which provides a framework / method for structured approach to architecture and development. (Certainly there are alternatives, earlier attempts and ongoing / newer developments). Agile approaches provide complementary management processes for project activities.
Central to TOGAF is the Requirement Management. The author may futher add here, that stakeholders and their intents are the impetus to the entire Requirement Engineering activity chains. Without them, subsequent activities and events lead to nowhere.
Effect of Generative AI on Requirement Engineering
Generative AI (GenAI) is a game changer. It allows human to create (things) at unprecedented speed and quantity. While the science and technology behind GenAI are complex (as in Probability Theories, Transformers, Generative Adversary Network, Variation Autoencoders, Large Language Models, …), its fruits are tasty and sometimes good.
The GenAI technology provide values in following areas:
- Efficiency – Speeding up the drafting of new requirements via automation and prototype texting.
- Quality – Improving consistency/traceability by using techniques such as templating and checking (syntax and semantics)
- Adoption – Making requirements easier to understand by using clearer and precise languages
- Summary – Condensing large quantity of text, images and videos into concise summaries
- Compliance – Aligning requirements to reduce risk and cover legal (law and regulation) topics
- Much more …
Potential Drawbacks – The Janus Phenomenon
GenAI is like an idol of Janus, a Roman two-faced god of transitions and doorways. It opens many doors and at the same time creates multiple challenges.
- Hallucinations: GenAI fabricates plausible but invalid requirements, leading to incorrect specs and downstream defects.
- Reproducibility Issues: Inconsistent outputs for identical inputs, hindering reliable validation and auditing.
- Lack of Interpretability: Opaque reasoning erodes stakeholder trust and traceability in decisions.
- Bias and Ethical Concerns: Amplifies training data biases, risking unfair or incomplete requirements; needs domain-specific safeguards.
- Low Industrial Maturity: Nascent adoption (90% early-stage, <2% production), with gaps in tools, datasets, and later RE phases.
Human is the Spark of Creation
Users and their actions, which are modeled after their interactions (existing and expected) with systems, provide the blueprint for requirements. This forms the elements for further elaborations, such as use cases, sequence, activities (behavior), …
When one uses GenAI to rapidly draft prototypes for the requirement documents, one starts a chain reaction of actions (the creation process). You, as the person who drives GenAI (wether via prompts, templates or automations activities), are in fact the creator.
בְּרֵאשִׁית, בָּרָא אֱלֹהִים, אֵת הַשָּׁמַיִם, וְאֵת הָאָרֶץ. – In the beginning God (with knowledge of A to Z – a shovel) created the heaven and the earth (Context & boundary of all work). Gensis 1:1
GenAI response is naturally a reflection of your intent. That response is bounded by a number of factors, such as and not limited:
- Knowledge Envelope – How much content was originally fed to the model
- Optimization and Biases – The model could be skewed by tuning for a number of reasons, such as cultural, legal, linguistic, technical constraints, …
- Hard Boundary – Machine limitations and performance, pertaining to generation speed, depth of search, memory limitations, …
- Probabilistic Parameters – Temperature (Randomness and Entropy), Sampling (such as K and P values), …
- Framing and Contextual Reduction – Making the response fit within a reasonable response window, suitable for human consumption and UI environment, …
- Much more
Communication and Rhetoric (Talents Working Together)
You are the source of creation and GenAI response is a reflection of your intent. One important questions arises, do you know everything about a problem and solution before the implementation has completed?
Most likely, that answer is NO. It is extremely unlikely or perhaps impossible to know every details before a “game” has been played. (Game is in a sense of a project or Unternehmen / Endeavour.)
This is where the knowledge of several people is greater than an individual, (“Two are better than one,” Ecclesiastes 4:9, for they can help each other succeed…)
The benefits include:
- Teamwork
- To be clarified … (ToDo)
(Mathematical) Reasoning behind Rhetorical AI
While Vibe Coding (using GenAI to generate application code) could benefit from optimization by using Newton’s method (also known as Newton-Raphson method), project management for Vibe Coding project requires more comprehensive considerations. I make the following comparison.
| Vibe Coding | Project Management for Vibe Coding | Stewardship for a Portfolio of Multiple AI-Assisted Projects | |
|---|---|---|---|
| Key Activities | – Generating Application Code based on Specifications – Binding Code to Specifications (Test-, Behavior- and/or Spec-Driven Development) – Documentation and Building Knowledge Base – Components Libraries – Statistics of Vibe Coding activities – Optimizing the number of iterations (cycles) during the design, the implementation / generation and the validation process | – Managing and coordinating team members involved in the Vibe Coding project – Resource Consumption and Time Management – Optimization of cost and productivity, in term of best mix between iterations, value creation and risk (uncertainty) determination and control. – Effective communication and coordination (agility & repeatability) – Lessons Learned (Retrospect) from a Vibe Coding development cycle | – Monitoring and supervising a portfolio of AI-assisted projects. – Benchmarking project(s) progress – Governing and charting the portfolio evolution and future growth – Fine tuning projects and balancing the portfolio using realistic strategies to maximize the holistic values |
| Mathematical Equivalence | Newton – Raphson Method and read Newton’s Laws of Motion – (Force and Acceleration to move objects is relevant to changing code base) – Data Set (Context) could be spun around like a physical object with a central of gravity (GenAIcentrus) in the data manipulation process read me presentation at APIDays – Paris | Euler-Lagrange Method – see this on least actions (kinetic and potential energy) – (Derivative and Integral of forces lead to the understanding about Energy – effort for combined team activities) – See project budgets and reserves. – Satisfaction, project progress and adaptability | Hamiltonian Mechanics – read this on symmetry and phased properties (a pair of 1st order differential equations helps us to understand and compare Vibe Projects and make selections for the best working approaches) – Optimization and Portfolio Theories of Vibe Coding or other AI-assisted projects – Benchmarking and Prioritization of two or more projects |
| Discussions | The Study of Forces allow a developer to find the best ways to implement application features (as per specifications aligned to requirements) | Rhetorical-AI is particularly relevant for this topic, because we have to understand and optimize the total energy (communication, coordination and commitment) necessary for success at team level endeavours. | Name yet to be coined – Synergentix – Synergy of AI technologies, team management and scientific human stewardship for tools, processes (frameworks) and methodologies. |
Scope Comparison in Vibe Coding (AI-Assisted) Projects
Rhetorical AI
GenAI is graded via Human-in-loop communication. More information to follow. For now, overview pictures.
