Skip to content

RFC: Founders waste time on heavyweight AI process or ship broken IaC from stale LLM training data #116

Description

@ppgmdev

Summary

Founders using AI coding tools either drown in process frameworks that generate a dozen design docs before any code, or get confidently-wrong CDK/CloudFormation from stale training data that fails at deploy. This skill (aidlc-lite) is a lightweight AI development lifecycle — understand → sketch → build → iterate — that verifies constructs against current AWS docs via the toolkit's MCPs before generating, keeping the align-before-building value of AI-DLC without the ceremony.

Motivation

Founders want working AWS code fast, not a stack of design documents. Full-process AI development frameworks are overkill for a solo founder, and generating IaC from stale LLM memory produces deprecated CDK/CloudFormation that fails at synth or deploy.

Proposed Solution

A process skill at skills/aidlc-lite/SKILL.md: a four-phase loop (understand → sketch → build → iterate) that asks max 3 clarifying questions, sketches services + trade-offs before coding, and verifies constructs with the awsiac/awsknowledge/awspricing MCPs before generating. Opinionated anti-patterns discourage over-process and generating IaC from memory.

Alternatives Considered

customer-ideation covers idea → architecture (service selection); this skill is the downstream build loop once services are chosen. Full AI-DLC fits regulated/enterprise contexts but is too heavy for a founder shipping today.

Open Questions

Whether to add a references/ file with a longer worked example, or keep it single-file.

Drawbacks

Conceptual overlap with customer-ideation on the "ask good questions, don't over-engineer" philosophy — the description must keep the ideation-vs-build-loop distinction clear.

Metadata

Metadata

Assignees

Labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions