Focused decision session
Bring one real AI use case. We map what AI may do, what remains human-owned, and which review points matter—then leave with a clear go, modify, or do-not-automate direction.
Human in the Middle
Decide what to automate. And what not to.
Practical teaching, training, and AI-governance advice for organizations that want useful AI without giving away judgment or responsibility.
Human in the Middle is my AI framework. It defines how I think about AI-shaped work: what should be automated, what should stay human-owned, where review belongs, and how systems can preserve judgment, responsibility, and control instead of hiding them behind automation.
Human in the Middle is the framework. HITM Labs is the teaching practice. Arthur's AI Lab is the video layer: short tutorials, practical walkthroughs, and examples that let future clients see the teaching style before they book a workshop or course.
The decision problem
AI pressure arrives before organizations know how to judge it
A board asks for an AI plan. A vendor promises automation. A team has already started experimenting. The difficult question is rarely whether AI can produce something. It is whether the organization can inspect, govern, and stand behind what happens next.
- What should be automated, assisted, or kept fully human?
- Where must a person review, approve, stop, or reverse?
- Who owns the result when AI contributed to it?
- What evidence would justify a go, modify, or do-not-automate decision?
Teaching is the front door
Learn the method by seeing it work
The teaching layer is deliberately tangible. It uses working examples, transparent experiments, and unfinished tools whose progress is shown honestly rather than disguised as finished software.
Work in progress
HITM Blueprint Studio
A guided environment for mapping ownership, approval gates, risks, and the boundary between human and machine work.
Work in progress
Agentic Office
A playful teaching device that makes agentic workflows, delegation, and human checkpoints easier to understand.
Work in progress
Whetstone
A public philosophy demo for testing claims, sharpening judgment, and resisting confident but weak AI output.
Ways to work together
Start with one real AI decision. Build from there.
HITM Labs
Applied and Quest formats teach teams through real material, visible checkpoints, and reusable workflows rather than abstract lectures.
AI-governance advice
Clarify ownership, approval, escalation, documentation, and adoption before an AI-supported workflow becomes difficult to change.
HITM in action
Case studies before claims
Aerial View, card&board, and Spiegel des Universums show HITM as a development practice: finished, working products built with AI while product intent, design authority, review, and responsibility remain human-owned.
DONE, GLASSBOX, and Panic Defense show how HITM principles change an interface: consequential actions become visible, review happens at the right moment, and a human can understand what the system is asking them to own.
The evidence sits at different levels of maturity: Aerial View and card&board are shipped products in private, daily use, while Spiegel des Universums is live. DONE, GLASSBOX, and Panic Defense remain design case studies; working prototypes are their next step. None is presented as proof of validated operational outcomes.
See the design case studies
Working product · used daily
card&board
A native macOS app I conceived, specified, designed, tested, and now use every day—implemented in SwiftUI with Claude Code.
Read case study
Working product · open source
Aerial View
A native file-curation app that organizes one file across many projects without moving or duplicating the original.
Read case study
Finished & live
Spiegel des Universums
The first product I built entirely through an AI-assisted development process. AI is not in the runtime; it helped make the specification, architecture, theme, and workflows buildable.
Read case studyOpen live site ↗
Case study
GLASSBOX
Visible intent, live execution, consequence gates, audit, and reversibility for automated work.
View project →
Case study
Panic Defense
A low-cognition crisis flow where support, human choice, and a clean exit matter more than engagement.
View project →The unfair advantage
HITM is backed by thirty years of design craft
Most AI-governance offers stop at principles, policies, or technical architecture. Code:Emotion adds the missing interface layer: the actual screens, states, gates, and decision points through which people retain control.
The two practices remain distinct sibling brands under schmidtpabst.com. They cross-promote because strategy and design are stronger when the governance model can be made visible and usable.
Customer discovery
Bring the AI decision you cannot yet make confidently.
I am currently speaking with organizations about where AI decisions stall, what vendor claims are difficult to evaluate, and where responsibility becomes unclear. This is a discovery conversation, not a disguised software pitch.