Warm Body Problem
Oversight on paper, absent in practice.
AI adoption advisory
Your AI works in the demo. Make it survive production.
Everyone has advice on adopting AI. Almost none of it explains why pilots succeed and production deployments quietly fail, or what to do about it. I help AI leaders find the real failure points and scale systems that hold up under real-world pressure, audits, and growth.
40+ production deployments audited · Forbes Tech Council · Creator of the HITL Maturity Model™
Most organizations think they have human oversight. They don't.
When the AI is wrong, someone has to catch it. In most systems I audit, that someone does not exist, is drowning, or has no way to prove they caught anything.
Burnout by Design
Reviewers flooded past capacity.
No Measurement Standard
Nothing to benchmark against.
The EU AI Act and NIST AI RMF both mandate meaningful human oversight but never define what "meaningful" looks like in practice.
Why this keeps happening
Production fails on a question nobody is measuring.
The AI industry has produced endless advice on models, prompts, and tooling. Almost none of it addresses the question that decides whether production succeeds: when the AI is wrong, does a qualified human catch it, and can you prove it? Companies that can’t answer that are running on an oversight illusion. Closing it starts with measuring something nobody has been measuring.
Here’s how to measure it ↓The Framework
The HITL Maturity Model™: five levels
Know your level in one working session. Leave with a ranked fix list.
A shared language for where an AI system actually sits, not where the deck says it does.
- L0 · Automation-First
The system runs. Oversight is a story people tell themselves.
- L1 · Reactive HITL
Humans intervene after something goes wrong, not before.
- L2 · Designed HITL
Review is scoped and routed, but capacity is the ceiling.
- L3 · Collaborative Intelligence
The model and the reviewer make each other better in the loop. Incidents drop; reviewer expertise compounds.
- L4 · Adaptive Governance
Oversight quality is measured, monitored, and improving: a number you can put in front of a board, a regulator, or a customer.
The HITL Health Index (HHI™) scores where you sit on this ladder, so progress is a number, not a feeling.
Proof
Where the work shows up
Advisory roles
“Su builds clarity where others see chaos. In 12 months, she helped a team reach $2M revenue with no sales team, cutting dev time by 40% and driving focused execution.”
“We took your recommendations seriously. Most of the jump is in Governance, the dimension you flagged, with real gains in human capacity and intervention quality too.”
“The maturity model gave me a practical framework to evaluate where my product stood and identify the highest-impact opportunities to improve. It helped me move beyond intuition, make more strategic product decisions, and build with greater confidence.”
Speaking
Talks on AI oversight, for rooms that make decisions.
I speak on human oversight design at conferences, internal leadership offsites, and board sessions. Formats run from a 20-minute keynote to a 60-minute session with Q&A.
Topics I speak on
- The warm body problem: why AI oversight fails on paper and in practice
- Where does your AI oversight actually stand? A five-level reality check
- Burnout by design: building oversight around human cognitive limits
- What the EU AI Act and NIST AI RMF actually require of you
- Collaborative Intelligence: designing the next decade of human-AI work
How we work together
How I bring the framework into your organization.
Most clients start with a workshop or a diagnostic. Start wherever you are: every engagement ends with something you can use the next day.
Learn the framework
Bring the model to your team, or start with the free material.
Half-Day Team Workshop
A working session for the team that owns your AI oversight. We run your live systems through the maturity model, name the failure modes you actually have, and leave you with a level and a ranked fix list. Delivered virtually or on site.
- Live walkthrough of all five levels
- Applied scoring against your own workflows
- Ranked priority list your team leaves with
- Virtual or in person
Full-Day Design Intensive
Everything in the half-day session, then a working design block. Your team leaves with drafted routing thresholds, escalation paths, and a reviewer capacity model for one live workflow.
- Everything in the half-day workshop
- Confidence-based routing design for one workflow
- Escalation thresholds and reviewer capacity model
- Written session output and next-step plan
- Virtual or in person
Free Lessons
Short lessons on human oversight design, published free. Start with Field Notes and the newsletter.
- Field Notes on this site
- The Framework Effect newsletter
Bring me in
Direct engagement when the work is bigger than a workshop.
HITL Diagnostic Assessment
Structured evaluation, scorecard, level classification, regulatory gap analysis.
Book a diagnostic call →HITL Design & Implementation
Confidence routing, escalation thresholds, feedback loops, agentic checkpoint gates.
Book a diagnostic call →Executive & Board Advisory
Board-ready governance presentations, investor due diligence, regulatory roadmap.
Book a diagnostic call →The Product: Humyn Pulse
A health monitor for your AI oversight.
Frameworks tell you where you stand once. Humyn Pulse measures your HHI™ score and shows you exactly where oversight is weakest, so you can fix it before it becomes an incident.
Writing & Talks
Where to read and hear more
Get the newsletter
Case studies and frameworks on AI governance, roughly weekly.