Making AI real
The AI Sandbox Needs a Gate
Experimentation matters. So does deciding what should leave the sandbox.
I like AI sandboxes.
Give people space to experiment.
Let them try things.
Build something quickly.
Break it.
Learn.
Ask:
What could this technology actually do for us?
That’s often how useful ideas emerge.
But eventually somebody has to build the gate.
Because an interesting prototype and a system that should operate in the real world are not the same thing.
Once an AI experiment begins interacting with real users, organisational processes or real decisions, the questions change.
Not only:
Can it do this?
but:
Which tools and systems can it access?
What decisions can it make?
When does it need human approval?
What happens when it’s uncertain?
What level of risk is acceptable?
Who is responsible when something goes wrong?
This becomes particularly important with AI agents.
An assistant that generates a suggestion is one thing.
An agent that interprets an intention, creates a plan, accesses tools and performs actions is another.
The UX is no longer simply:
- click
- response
It may become:
- intent
- interpretation
- planning
- actions
- decisions
- human checkpoint
- further action
- outcome
Much of that process may be invisible to the person using it.
That’s why I’m increasingly interested in the experience design of agentic systems.
Where should the gate be?
When does the human need to appear?
What should they see?
What should they approve?
How can they interrupt?
How do they regain control?
Experimentation helps us discover what is possible.
Human judgement helps us decide what should become real.
- Agentic AI
- AI Prototyping
- Human Oversight
- AI Governance
- AI Experience Design