💡 AI/ML Production Readiness Checklist Extension

[github-actions[bot]]

Summary

Extend the production readiness checklist template with AI/ML-specific verification items covering model monitoring (data drift, concept drift, feature distribution), inference performance SLOs (latency percentiles for model serving, throughput, cost-per-query), model versioning and rollback, GPU/TPU resource management, LLM-specific concerns (hallucination monitoring, token budget governance, prompt injection protection), and AI-specific disaster recovery (model artifact backup, training data preservation). As AI workloads become a growing share of production systems, a production readiness checklist that doesn't address AI/ML operational requirements is fundamentally incomplete.

Market Signal

Gartner projects 40% of enterprise applications embedding AI agents by 2026. AI workload costs are exploding — GPU clusters, inference APIs, and AI agent operations significantly increase cloud spend (Eduonix FinOps report). Google's MLOps maturity model and AWS ML Well-Architected guidance exist but are vendor-specific. The industry lacks a vendor-neutral, methodology-driven production readiness framework for AI/ML systems. OpenTelemetry's GenAI semantic conventions standardize how AI operations are recorded, making AI production readiness systematically verifiable for the first time.

User Signal

The existing production readiness checklist template (src/templates/bgr-production-readiness-checklist-template.md) covers traditional operational readiness: observability, incident response, infrastructure, security, pipelines, disaster recovery. It does not include any AI/ML-specific items. Given that BMAD is an AI-native development framework (users are building AI-powered applications), this is a direct gap in the project's core value proposition. The operations review skill (OR) validates completeness of artifacts — an AI/ML section in the checklist ensures AI operational readiness is systematically verified.

Technical Opportunity

The production readiness checklist is a single template file referenced by all workflow validation steps (step-05-validation). Adding an AI/ML section to this template immediately makes the verification available across all workflows that update the checklist. The section can be conditional on the bgr_maturity configuration or a new ai_workloads flag, so teams without AI components are not burdened with irrelevant checks. This pairs naturally with the AI/LLM Observability Extension (💡 #383) and AI Agent Security Governance (💡 #384) proposals.

Assessment

Dimension	Score	Rationale
Feasibility	high	Single template file extension — minimal effort, immediate value across all workflows
Impact	high	Closes a direct gap in the core value proposition for a growing majority of BMAD users deploying AI
Urgency	high	40% of enterprise apps embedding AI by 2026; production readiness without AI/ML coverage is incomplete for these teams

Adversarial Review

Strongest objection: Adding AI/ML items to the checklist may clutter it for teams that aren't deploying AI workloads, and the rapid pace of AI development means specific checklist items could become outdated quickly.

Rebuttal: Conditional sections (gated on a configuration flag) solve the clutter problem — teams without AI workloads never see the AI/ML items. For staleness, the items should be framed as capability checks ("Is model drift detection configured?") rather than tool prescriptions ("Is Evidently AI deployed?"), making them durable as the tooling landscape evolves. This approach is consistent with how the existing checklist handles other domains — asking "Is alerting configured?" rather than "Is PagerDuty configured?"

Suggested Next Step

Add a conditional "AI/ML Production Readiness" section to src/templates/bgr-production-readiness-checklist-template.md covering model monitoring, inference SLOs, model versioning, GPU resource management, LLM governance, and AI disaster recovery. Gate on a new ai_workloads config option in module.yaml.