1. Identify Business Challenge
We observed that many professionals struggle to grasp Solution Architecture through static content. Reading about suitability checks, AI type selection, and governance is intellectually useful — but not experientially clear.

There was a need to demonstrate what architecture looks like by first showing what happens when it is missing. AIBadBot was designed as a contrast tool to make architectural gaps visible through interaction rather than explanation.

2. Conduct Suitability & Readiness Check
This challenge is well-suited for AI because the tool must interpret natural-language problem statements across many industries and functions, detect the likely intent behind vague inputs, and respond in a consistent interaction pattern every time.

It is also feasible because Benchmark already has the architecture framework (CAISA), the teaching narratives and failure-pattern library needed to design controlled “bad” responses, the delivery competence to map each failure to an architectural lesson, and a ready web interface to deploy and iterate the experience

3. Select Appropriate AI Type
A combination of Conversational AI and Generative AI is used.

Conversational AI manages flow, routing, and mode switching (BadBot vs GoodBot). Generative AI produces both the intentionally misaligned responses and the structured architectural alternatives.

4. Input Capture & Structuring (Problem Framing Layer)
The agent captures the user’s business situation in natural language, cleans and classifies it into predefined context buckets (e.g., prediction, automation, governance), and converts it into a consistent internal structure.

In BadBot mode, clarification is intentionally minimal. In GoodBot mode, the same structured context supports disciplined architectural reasoning.

5. Assumption Logging & Clarification Loop
In BadBot mode, vague inputs are accepted without clarification and assumptions remain implicit — intentionally reflecting poor architectural discipline.

In GoodBot mode, the response explicitly states the assumptions that should be validated and outlines the clarification steps that a properly architected system would require before proceeding.

6. Feasibility & Approach Selection (LLM + Rules Hybrid)
The agent maps each user scenario to an AI context category and then activates one of two response strategies.

In BadBot mode, it deliberately misapplies an AI pattern and skips critical layers. In GoodBot mode, it explains the appropriate AI pattern (e.g., automation, retrieval, predictive modeling, human-in-the-loop) and outlines how it should be applied with architectural discipline.

7. Risk, Controls & Human Oversight
In BadBot mode, guardrails such as data sensitivity checks, hallucination controls, approval points, audit trails, and escalation logic are intentionally absent — illustrating governance failure.

In GoodBot mode, the response outlines the control mechanisms that should be built into the solution, including human review checkpoints, boundary conditions, and monitoring requirements.

8. Output Generator (Boardroom-Ready Justification Pack)
In GoodBot mode, the tool presents a structured architectural justification: clarified problem statement, scope boundaries, recommended AI approach, required inputs or data, key risks with mitigations, expected value, and indicative implementation effort.

In BadBot mode, this structure is intentionally missing — reinforcing the contrast between enthusiasm and engineered design.

9. Architectural Integrity Check
In GoodBot mode, the response implicitly evaluates whether the proposed approach aligns with core architectural principles — clarity of objective, suitability of AI type, defined components, control mechanisms, and governance structure.

In BadBot mode, no such integrity check exists. The absence of architectural validation is intentional and forms part of the learning contrast.

10. Interaction Loop & Learning Continuity
The tool is designed as a cyclical learning experience. After presenting the GoodBot alternative, it intentionally resets to BadBot mode for the next scenario, reinforcing that architecture must be consciously applied each time.

Rather than capturing leads or routing to CRM, the primary objective is architectural awareness — encouraging users to test multiple scenarios and observe recurring failure patterns.

Architecture Flow
User inputs business situation → Light clean-up & context classification → Intentional wrong AI-type selection (BadBot) → Confident misaligned response → Italic confession naming the missing architectural layer → User chooses (GoodBot or continue chaos) → If GoodBot: structured architected explanation (objective, suitability, AI type, components, controls, governance) + humorous invite to continue → Reset to BadBot mode → Next scenario.