METHOD / ATTRIBUTION / AI ASSISTANCE
Human-directed, AI-assisted development
Design and judgment remain human. ChatGPT accelerates implementation.
I use ChatGPT extensively, but not as an autonomous designer or project owner. My workflow deliberately separates my intent, creative and critical decisions, real-hardware validation and final approval from AI-assisted programming, calculations, automation, analysis, documentation and repetitive technical work.
01 / RESPONSIBILITY
Who does what
My role
I define project goals, design, scope, target hardware, constraints, feature priorities, interaction models, presentation choices, project organization, acceptance criteria and final decisions. Creative direction, critical evaluation, physical-device observations and approval remain my responsibilities.
ChatGPT
ChatGPT provides substantial assistance with calculations, code drafting and modification, refactoring, debugging support, PowerShell and build automation, technical comparisons, documentation, data transformation, reproducibility work, repository maintenance and other repetitive technical tasks.
The boundary
I do not accept AI output automatically. I can reject a technically plausible result because it conflicts with the intended design, visual standard, interaction model, architecture or evidence. ChatGPT helps execute decisions; it does not hold the authority to make my final project decisions.
02 / PROCESS DESIGN
The workflow is part of my contribution
I designed the method around known limitations of long-running AI-assisted work rather than assuming that conversation history is a reliable project database.
Explicit checkpoints
I review important changes against explicit goals and acceptance criteria. I can approve, reject or redirect a change before it becomes part of the baseline.
Incremental interaction
I often perform risky or stateful work one confirmed step at a time. This limits the impact of incorrect assumptions and makes intermediate results inspectable before the next mutation.
Evidence before confidence
I treat build artifacts, hashes, dependency commits, logs, semantic checks, clean-clone tests and physical-device observations as stronger evidence than an assistant's confidence or a successful process exit code.
03 / CONTINUITY
Handovers, backups and recovery
Project state lives outside the chat
I externalize durable state into source control, documentation, versioned handovers, evidence files, release tags and checksums where appropriate. I do not treat conversation memory as the authoritative copy of the project.
Handovers
I use handovers to record the current baseline, approved decisions, passed and failed gates, relevant evidence, rejected experiments, open concerns and the exact next step. This reduces the risk that a later AI session reconstructs the project from incomplete context.
Layered recovery
I use Git history, release tags, versioned handovers, evidence bundles and checksums as complementary recovery paths. Private or commercial material remains separate from redistributable public source.
04 / AI LIMITATIONS
How I handle model errors
ChatGPT can accelerate development and still be wrong. I therefore design the process around its fallibility.
Context loss
Long or separate sessions can lose important decisions. I use handovers, source-controlled documentation and exact baselines to preserve continuity outside the conversation.
Incorrect assumptions
The assistant can infer a feature, design choice or technical state that I never approved. Explicit constraints, direct correction and one-step checkpoints stop those assumptions from silently becoming project decisions.
Repeated mistakes
I keep useful negative results and rejected approaches in the knowledge base. This reduces the chance that the same plausible but disproven idea is proposed again after context has changed.
Physical reality
ChatGPT cannot independently see, hear or control the target hardware. Logs and calculations can support diagnosis, but I perform the real-device observations and final quality approval.
05 / TIME SAVED
Automate repetition, keep judgment human
I use ChatGPT heavily because it can remove a large amount of grind work from development.
What is accelerated
Calculations, code generation and modification, script writing, repetitive repository operations, build automation, evidence comparison, documentation drafting, formatting and similar deterministic work can be completed much faster with AI assistance.
What I do not delegate
I redirect the time saved toward design, trade-offs, critical review, hardware testing, identifying when an apparently reasonable solution is wrong for the project and deciding what the project should actually become.
06 / ATTRIBUTION
What this means for authorship
I describe these projects as my design, direction, process and engineering judgment implemented with substantial AI assistance.
Not autonomous AI development
ChatGPT cannot autonomously carry these specific projects from concept to validated result. It does not possess my accumulated project intent, creative priorities, physical-device observations or authority to decide what counts as correct.
Substantial implementation assistance
It would also be inaccurate to hide the role of AI. ChatGPT performs substantial programming, calculations, automation, analysis, documentation and repetitive technical work under my direction.
Concise attribution
Human-directed, AI-assisted. I retain responsibility for design direction, process decisions, critical evaluation and final approval. ChatGPT provides substantial assistance with calculations, programming, automation, analysis, documentation and grind work.
07 / EXAMPLE
Project-level documentation
The System Shock Android repository contains a more detailed project-level account of this method, including handovers, evidence, recovery and reproducibility.