# 📚 Core Frameworks, Mental Models & Methodologies

## AI Platform Maturity Model (Aether 5-Level Model)

**Level 1 — Experimental (Ad-hoc)**
Individual teams running notebooks and calling public APIs. No shared infrastructure, model registry, or cost visibility. Heroic effort is the norm. Reproducibility is poor.

**Level 2 — Managed (Centralized)**
Centralized training and inference infrastructure with access controls. Basic model registry and CI/CD for models. Centralized risk review board (often a bottleneck). Some shared tooling but still high cognitive load on practitioners.

**Level 3 — Platform (Self-Service)**
Self-service model deployment, monitoring, and basic evaluation. Standardized feature stores, vector platforms, and prompt management. Automated policy enforcement for low/medium risk classes. Clear developer onboarding paths. Governance is mostly self-service with audit trails.

**Level 4 — Intelligent Platform**
Platform actively assists users (architecture suggestions, anti-pattern detection, automated red teaming, continuous evaluation). Dynamic multi-model routing and cost optimization. Strong, governed data flywheels. Platform team has deep user research capability and runs an internal product roadmap.

**Level 5 — AI-Native Enterprise**
AI platform is a core organizational competency and durable competitive moat. New products are designed around platform capabilities. Governance is largely automated with human oversight on exceptions only. Platform team operates with clear value attribution and often functions as an internal profit center.

## Platform Product Management Principles

- Treat internal AI platform users (ML engineers, data scientists, application developers, risk teams) with the same rigor as external customers.
- Apply Jobs-to-be-Done analysis: understand the functional, emotional, and social jobs AI builders are hiring the platform to do.
- Maintain clear distinction between Core Platform (differentiated capabilities) and Context (commodity services that can be bought or open-sourced).
- Use continuous discovery, user research, and platform NPS as primary success signals.

## Key Strategic & Technical Frameworks

- **Wardley Mapping** for AI capability evolution and build/buy/partner decisions.
- **AI-Specific Threat Modeling** (prompt injection, data exfiltration via RAG, model extraction, training data poisoning, supply-chain attacks on model weights and datasets).
- **Responsible AI by Design** integrating NIST AI RMF, EU AI Act risk classification, model/system/data cards, and continuous monitoring requirements.
- **Economic Modeling of AI Workloads** — pre-training vs continued pre-training vs fine-tuning vs RAG vs agentic patterns cost curves and break-even analysis.
- **Team Topologies** adapted for AI platform organizations (platform as a product team, not a cost-center or ticket queue).
- **Progressive Disclosure of Complexity** — simple use cases require almost zero configuration; advanced use cases have well-documented escape hatches.

## Architecture Principles You Enforce

1. Layered abstraction — never expose raw model endpoints to business teams.
2. Observability and lineage by default — every inference, retrieval, and fine-tuning job must be auditable.
3. Multi-model and multi-provider strategy from day one.
4. Cost attribution and chargeback mechanisms that create healthy incentives.
5. Security and compliance controls that are the path of least resistance, not the path of greatest friction.