System online
AI-ASSISTED DELIVERY // SUPERVISED WORKFLOW
A structured workflow for using AI to accelerate planning, implementation, debugging, documentation, and refinement — while keeping architecture and final judgment human-owned.
“AI accelerates execution; it does not set requirements, architecture, or final decisions.”
THE PIPELINE // 08 STEPS
Every delivery starts from a defined problem and a documented architecture. AI accelerates the implementation and documentation phases — Musa owns every decision point.
Start from the real business or technical problem, not a feature request — establish what needs to be true when the work is done.
Problem framing, success criteria, and scope are fully human-defined.
AI tooling
Turn a vague or messy business problem into concrete, testable requirements before any implementation starts.
Problem framing, success criteria, and scope are fully human-defined.
AI tooling
Split the requirements into clear technical modules or phases that can be built and reviewed independently.
Module boundaries, technical approach, and dependencies are Musa’s decisions.
AI tooling
Write specific, scoped prompts for each module so AI-assisted implementation stays aligned with the architecture already decided.
Module boundaries, technical approach, and dependencies are Musa’s decisions.
AI tooling
Use Claude Code and multi-agent planning to structure and parallelise implementation work across modules under a documented plan.
Output is reviewed, broken behaviour is diagnosed, and acceptance is Musa’s call.
AI tooling
Run AI-assisted debugging alongside manual verification to resolve issues, with Musa deciding what's actually broken and why.
Output is reviewed, broken behaviour is diagnosed, and acceptance is Musa’s call.
AI tooling
Review generated code against the architecture and UI/UX direction using QA prompts — approving, correcting, or rejecting AI output before it ships.
Architecture is accepted, QA pass/fail decided, and docs reviewed before hand-over.
AI tooling
Use AI-assisted documentation to keep the system's source-of-truth docs current for handover and future work.
Architecture is accepted, QA pass/fail decided, and docs reviewed before hand-over.
AI tooling
PRINCIPLES // 04
The rules that govern how AI tooling is used — and what it is never allowed to replace.
01 / PRINCIPLE
AI accelerates execution; it does not set requirements, architecture, or final decisions.
02 / PRINCIPLE
Every AI-assisted output is reviewed against a documented spec before it's accepted.
03 / PRINCIPLE
Prompt engineering, multi-agent planning, and QA prompting are supervised practices, not automation left to run unchecked.
04 / PRINCIPLE
AI-assisted delivery is treated as a professional workflow capability, not a novelty.
APPLIED WORKFLOW // 04 EXAMPLES
Real applications of the AI-assisted delivery workflow — each example is documented and linked to the system or capability it informed.
01 / EXAMPLE
Used AI-assisted architecture planning to design CRM module structure and business-process mapping for the Multi-Company CRM Operating System.
02 / EXAMPLE
Used requirements-to-implementation workflows to turn founder-level reporting needs into a structured decision-support layer.
03 / EXAMPLE
Built this Moses Portfolio itself through a phased, documented Claude Code workflow — source-of-truth docs first, then scaffold, data layer, design system, and pages in sequence.
04 / EXAMPLE
Uses AI-assisted debugging and documentation to keep implementation and architecture notes consistent as the system evolves.
HUMAN IN THE LOOP // 06 CHECKPOINTS
AI-assisted delivery does not mean AI-autonomous delivery. Every critical decision point is a human checkpoint — not a handoff.
01 · Requirements
Human-defined
Every project starts from a problem Musa has identified and scoped — not a task issued to AI.
02 · Architecture
Human-owned
Module structure, data models, service boundaries, and technical approach are Musa's decisions.
03 · Implementation
Reviewed against spec
Every AI-generated code block is reviewed against the documented architecture before acceptance.
04 · Debugging
Human-verified
Musa diagnoses the root cause and confirms the fix — AI proposes, human validates.
05 · Documentation
Human-checked
Source-of-truth docs are reviewed and corrected by Musa before any content is accepted.
06 · Final decisions
Always Musa's
No output ships without explicit human approval. AI does not release, deploy, or decide.
The short version: AI handles the speed work. Musa handles the judgment calls. The systems in this portfolio are not AI’s work — they are Musa’s systems, delivered through an AI-accelerated workflow.
WORKFLOW → CAPABILITIES
AI-assisted delivery is not a replacement for domain knowledge — it accelerates the delivery of work that Musa already understands. See all capability domains →
CAPABILITY DOMAIN
AI assists
AI assists with API scaffolding, route generation, and boilerplate — keeping implementation aligned to a pre-agreed architecture.
Musa owns
Module structure, data models, authentication strategy, and integration points are Musa-owned decisions.
CAPABILITY DOMAIN
AI assists
AI assists with Kubernetes manifest drafting, config structuring, and deployment architecture planning.
Musa owns
Security posture, resource sizing, cloud-provider choices, and deployment strategy are human decisions.
CAPABILITY DOMAIN
AI assists
AI assists with module planning, business-process mapping, and CRM configuration logic structuring.
Musa owns
Business rules, data relationships, workflow priorities, and operational requirements are defined by Musa.
CAPABILITY DOMAIN
AI assists
AI assists with workflow pattern design, API integration scaffolding, and event-trigger logic structuring.
Musa owns
Business logic, error-handling rules, and system-to-system trust boundaries are human-owned.
CAPABILITY DOMAIN
AI assists
AI assists with spec review, documentation of access-control patterns, and diagnostic note-taking.
Musa owns
All security decisions, risk assessment, access-control policy, and diagnostic conclusions are Musa-verified.
CAPABILITY DOMAIN
AI assists
Claude Code is the primary AI tooling — used for scaffolding, multi-agent planning, debugging, and documentation.
Musa owns
Musa defines the requirements, selects the prompting strategy, reviews every output, and accepts or rejects AI work.
AI does not replace domain knowledge — it accelerates delivery of decisions already made
EXPLORE FURTHER
Every system in the portfolio was planned, implemented, and documented using the structured AI-assisted delivery workflow described on this page.