Public challenge · try it online
Who supervises the AI?
Six situations. You choose how to verify what the AI hands you.
≈ 4 minutes · no score · neither a validated test nor a Hulotech solution.
Try the challenge →
ALL IN trade show · Montréal · September 16–17, 2026
AI can produce the result.
But what did you actually learn, understand or do?
It can give you access to capabilities that were out of reach, or let you skip the very effort meant to build them.
Hulotech explores this question at the ALL IN trade show in Montréal, through four demonstrators.
The public challenge and the interactive demo are already available online: try the challenge →
Illustrative image.
Why Hulotech will be at ALL IN
When AI takes part in the work, what do we know about what a person understands, can verify and can mobilize again?
Hulotech explores how to move from a one-off performance to evidence that accumulates. At the show, we come to present what we observe, what we cannot conclude, and to meet organizations interested in AI-assisted training under human supervision.
Our founder works every day with AI agents able to carry out certain technical tasks well beyond his direct expertise. His challenge is not to become better than them, but to know when to trust them, what evidence to ask for and when to escalate.
Established concepts that Hulotech did not invent: calibrated trust, meaningful human oversight, automation bias, the asymmetry between solving and verifying.
What we are trying to build
A way to observe, not yet a way to measure.
Hulotech explores instruments that observe a real situation, keep track of the conditions and assistance received, distinguish an output from a capability, identify what is missing, adapt the next practice or the next authorization, then retest.
- Observe
- Evidence
- Decide
- Next action
- Retest
→ then we start again, in a different context.
None of this is a complete system today. These are mechanisms we build and put to the test, one demonstrator at a time.
The 4 demonstrators
Four experiments, one tension.
These are not four finished products. They are four instruments, at very different stages, each asking a different question about the same problem.
Demonstrator 1 · OF25
What have we actually observed?
A public challenge and a demo you can try online right now, then a demonstrator presented by the team.
Interactive demo · open it online
Output is not capability.
View the interactive demo →
New tab · large screen recommended
Demonstrator · presented by the team
Capability Evidence: an evidence record, not a score.
Diagnostic → gap → targeted practice → retest → evidence record. A real demonstrator, presented by the Hulotech team. No AI evaluates capability.
See the real state of each component ↓After the challenge
Are six good answers enough to say that you know how to supervise AI? No.
These six decisions show how you reacted in these situations. They do not prove that you are, or are not, a good supervisor of AI.
A good answer on a test is not necessarily a demonstrated capability. A capability observed in one situation is not automatically demonstrated in another.
- Good answer≠
- Good test performance≠
- Transfer observed≠
- Capability across contexts≠
- Real-world capability
Confidence ≠ competence · AI-assisted performance ≠ independent capability · Second AI opinion ≠ independent evidence
This is precisely the problem Hulotech explores.
How OF25 is built
Inside OF25: four building blocks.
The challenge and the demonstrator are the visible parts of research specific to OF25, on learning, evidence state and follow-up over time.
- 0
Starting point
A real objective, a real person
- Initial behaviour · exploration
01 · Person-centred training
Student-Based Training
« What should this person practise next? »
- Working demonstrator
02 · Evidence of capability
Capability Evidence
« What have we actually observed? »
- Existing foundation · not yet connected
03 · Memory of evidence over time
Cognitive Passport
« What do we know over time? »
- Research direction
04 · Next learning experience
Next Best Learning Experience
« What should happen next? »
Real maturity level
Where each OF25 component really stands.
We know something. We do not claim to know more.
Capability Evidence
Working demonstratorA real demonstrator exists. It shows approximately: diagnostic → gap → targeted practice → retest → capability evidence record.
No AI evaluates capability. The system currently relies on heuristic formative signals, human adjudication and an evidence workflow.
The retest result was observed after practice, in this situation. Transfer remains undemonstrated.
Student-Based Training
Initial behaviour / explorationAn elementary logic already exists: diagnostic → gap adjudicated by a person → prescribed practice → retest.
It is not a complete adaptive engine.
Cognitive Passport
Existing foundation — not yet connectedTechnical foundations and models exist.
There is no complete longitudinal engine, no fully connected writing, and no complete operational product.
Next Best Learning Experience
Research directionNothing is implemented.
The question is asked: what should we observe next to reduce uncertainty?
We can show a public challenge that makes the problem tangible, then an interactive demo that explains how Hulotech explores an answer based on learning, evidence state and longitudinal follow-up, with a real demonstrator mostly on the Capability Evidence side.
Demonstrator 2 · Targeted practice adaptation
ALL IN demo — frozenFrench AI Tutor
« What should the learner practice next? »
The tutor observes what the learner actually demonstrated, identifies a relevant gap, adapts the next practice to that gap, then changes the situation to see whether the learning can be used again.
Frozen on a bounded demonstration case (real production → actual gap → targeted practice → different-situation retest → evidence state). This is not the tutor's general future architecture.
We do not claim: a general A/B/C tutor · a complete FSL evaluator · validated pronunciation assessment · proven learning gains · demonstrated professional transfer · stay / repair / practice / advance as a demonstrated general policy.
Demonstrator 3 · Adaptive interaction
ALL IN demo — frozenPhilo Coach
« How can AI support thinking without replacing it? »
A prototype that separates observation, state, decision and generation, so that an answer does not replace the act of thinking.
An answer ≠ thinking.
We do not claim: scientifically proven maieutic learning · demonstrated improved reasoning · generalized cognitive gains.
Demonstrator 4 · AI agent governance
Level 1 demonstrated · Level 2 in progressAOC
« What should an AI agent be allowed to do next? »
An internal governance instrument: an agent's action produces evidence, a person renders a verdict — accepted, rejected or reworked — and that verdict becomes a durable record informing future authorizations.
Done ≠ accepted.
Level 1, demonstrated: bounded mandate, evidence attached, a distinct human verdict. In a real rehearsal, a first run was blocked when an evidence condition was not satisfied, then fixed and retested before being separately accepted by the Founder. Level 2, evidence-driven adaptation: in progress. Level 3, autonomous self-improvement: not demonstrated. The governance core is demonstrated; the cockpit remains an internal prototype, not a mature commercial interface.
We do not claim: an operational autonomous company · demonstrated recursive self-improvement · agents operating Hulotech without human supervision · Level 2 demonstrated · a demonstrated ROI · AOC ready for enterprise deployment.
Before we meet
What we can show.
What remains unproven.
We have built several demonstrators that let us experiment with mechanisms of observation, provenance, assistance, decision, retest and evidence — on real situations, in bounded contexts.
- A general measure of human competence
- A general capability across all contexts
- A real-world operational capability
- An AI able to determine competence on its own
- A scientifically validated method for what a human should delegate or keep
- A self-improving autonomous company
This honesty is part of what we want to show at ALL IN.
Meet us at the show
Will you be at ALL IN?
We show what we observe and what we cannot conclude. Write to us to arrange a time on site.
Let's talk about
- AI-assisted training
- Human supervision
- Capability evidence
- AI agent governance
- AI projects applied to training