Frozen teaching version
CO2Calc Lab Snapshot
Clean architecture state for teaching, review, and case-study reference.
Open frozen snapshot ↗Build Fast. Decide Slowly.
Build an MVP in 5 days without losing control of your architecture.
Human-in-the-Middle Architecture is a structured way to build with AI: research is human, structure is human, AI executes stepwise, and nothing ships without review.
The Problem
Most AI-assisted builds fail for the same reason: too much automation, too little structure.
The common pattern is speed first, architecture later. The result is technical debt by day three, silent AI decisions, and a repository developers eventually have to untangle.
What breaks
- Messy repositories
- Silent AI decisions
- Technical debt on day three
- Developers cleaning up chaos
- Founders losing control
The Method
Human-in-the-Middle Architecture is controlled acceleration.
Core principle
- Research is human
- Structure is human
- AI executes
- Every decision is human-approved
- No blind commits
- No silent automation
The alternative to vibe coding
- Define constraints
- Write structured specifications
- Decide what AI may touch
- Generate stepwise
- Approve every change
No magic folders. No automation maximalism. Calm authority.
Architecture Model
Diagram placeholder (v1)
Final version will be a dedicated SVG. This placeholder uses the exact structure and labels from the diagram specification.
Problem / Research
Research is Human
Research is Human
Structured Markdown Spec
Spec Before Code
Spec Before Code
Human Approval Gate
Nothing ships without review.
Nothing ships without review.
Controlled AI Execution
Stepwise Generation
Stepwise Generation
Stable Codebase / MVP
Engineer-respectful output
Engineer-respectful output
Proof of Method
This method is not hypothetical.
I used Human-in-the-Middle Architecture to build a working emissions workflow MVP in two days. The frozen version shows the clean teaching state. The live version continues to evolve.
Live evolving version
CO2Calc (Ongoing)
Current working version as development continues and workflows evolve.
Open live version ↗The lab formalizes the process behind Version 1 so you can build Version 2 of your own product responsibly.
What You Will Build
A shared, realistic example project (not a toy app).
All participants work on the same project to create shared vocabulary, shared debugging, and controlled complexity.
Example project: Spec-to-MVP Tool
- Ingests structured markdown specifications
- Generates a roadmap summary
- Displays feature visualization
- Includes simple authentication
- Maintains clean API separation
- Uses a minimal database
- Includes export functionality
- Contains README + Known Limitations file
Why this matters
- Realistic enough for funding/demo conversations
- Structured enough for developer handoff
- Complex enough to teach boundaries and discipline
- Simple enough to complete in an intensive format
5-Day Intensive
Five consecutive days. Serious work. Small group.
Leave with a clean, funding-ready MVP and an architecture process you can reuse.
Day 1
Architecture & Spec
- Define constraints
- Write structured markdown specification
- Plan folder structure
- Define system boundaries
- Define AI permissions
Output: Clear repo blueprint.
Day 2
Controlled Code Generation
- Stepwise Codex / Claude execution
- Human approval gates
- Commit discipline
- Commenting standards
- Architecture checks
Output: Stable base structure.
Day 3
UI & Asset Layer
- Design-to-code alignment
- Clean component structure
- Optional ComfyUI workflow overview
- Naming governance
Output: Functional interface.
Day 4
Refinement & Documentation
- Error handling
- Edge case review
- Refactoring prompts
- README writing
- Known limitations documentation
Output: Engineer-respectful prototype.
Day 5
Demo & Packaging
- Problem framing
- Architecture summary
- Technical roadmap
- Risk documentation
- Funding-ready demo
Output: MVP ready for funding, handoff, or careful expansion.
Modular by Design
The MVP Lab is the full architecture experience. Modules are the scalable system around it.
Individual modules can be booked separately, delivered in-house, or combined into custom programs. The flagship remains the full Human-in-the-Middle MVP Lab.
Judgment Before Automation — Modular Series
Module 1
Spec Before Code™
Structured Markdown Architecture Workshop
- Constraints definition
- Structured markdown specifications
- System boundary planning
- AI permissions design
Audience: Designers, product thinkers, founders, UX architects
Module 2
Controlled AI Execution
Stepwise Generation & Approval Gates
- Codex / Claude execution discipline
- Human approval gates
- Commit hygiene
- Architecture integrity checks
Audience: Product builders, architects, technical designers
Module 3
Engineer-Respectful Repositories
Repo Discipline & Handoff Structuring
- Folder governance
- Naming systems
- README discipline
- Known limitations documentation
Audience: Founders, designers transitioning into build roles
Module 4
Asset Pipelines with Authority
ComfyUI & Structured Visual Generation
- Controlled asset pipelines
- Design-to-code alignment
- Mac to PC render workflows
- Visual governance and consistency
Audience: Designers, hybrid builder-designers, creative technologists
Module 5
Funding-Ready MVP Packaging
From Prototype to Credible Demo
- Architecture summary
- Technical roadmap
- Risk documentation
- Responsible Version 2 planning
Audience: Founders, indie SaaS builders, product designers
Custom corporate versions available. Modules are secondary pathways, not competing offers.
Who This Is For
Architectural thinkers who want speed without surrender.
- Product designers transitioning into build roles
- UX architects
- Indie SaaS founders
- Technical designers
- Innovation-focused freelancers
This Is Not For
Shortcut workflows and automation maximalism.
- Complete beginners
- Prompt hobbyists
- People looking for shortcuts
- "Just generate everything" workflows
If you want to automate responsibility away, this is not your lab.
Founding Cohort (Intensive Edition)
Small, deliberate, and manually reviewed.
Applications are open. Cohort limited to 6 participants. Seats are confirmed individually upon acceptance and payment.
Early-Bird
€900
First 3 confirmed seats
Founding Cohort
€1,100
Next 3 confirmed seats
Regular Price (future cohorts)
€2,100
Published reference price
Rolling admissions (structured)
Applications are reviewed weekly on a rolling basis until all 6 seats are filled. This creates a clear timeline without fake countdown pressure.
Manual review and seat confirmation
Applications are reviewed manually. This is not an automated enrollment process. Accepted participants receive an acceptance email and payment link. A seat is confirmed upon payment.
Course language can be adapted to cohort preferences (English/German) where group composition allows.
About
Built by a designer moving into product architecture and AI-assisted building.
My workflow is research-first, markdown-spec driven, and repository-disciplined. I use AI for stepwise execution while keeping human approval at every stage.
Using this method, I built a working MVP in two days, clean enough to apply for funding and structured enough to hand off. This lab formalizes that workflow.
Workflow principles
- Research-first
- Markdown specification
- Repo discipline
- Stepwise AI generation
- Human approval at every stage
FAQ
Practical questions
Do I need to be an experienced developer?
No. But you must understand basic product structure and be comfortable with structured thinking.
Is this no-code?
No. This is AI-assisted architecture with human control.
Can I bring my own project?
Not in the founding cohort. Advanced cohorts may allow this later.
Is this anti-developer?
No. The goal is engineer-respectful output.
Will we use a specific stack?
Yes. The stack will be defined before the cohort begins to ensure alignment and maintainability.
Apply
Build fast. Decide slowly.
If you want controlled acceleration instead of chaos, apply for the Founding Cohort. Applications are reviewed manually, and seats are confirmed individually.
Replace the Tally placeholder URL (`https://tally.so/r/REPLACE-ME`) before publishing.