If capability rather than old activity is what matters, what should people still be able to do after AI helps?
This is where a concern can easily turn nostalgic.
Keep handwriting because handwriting used to matter.
Keep drafting from scratch because that is how we learned.
Keep calculating manually.
Keep searching the old way.
Keep the difficult task because difficulty builds character.
That is not the argument.
Some effort is pointless.
Some work should disappear.
AI can remove friction that never deserved protection.
It can open access to people who previously needed specialist technical skill just to begin.
It can let an expert spend less time on routine production and more time on judgement.
It can help a novice enter a problem that would otherwise remain closed.
Taking AI away would preserve the wrong thing.
The harder task is to identify the capability underneath the activity.
What was the old work doing for the person?
Was it building memory?
Pattern recognition?
Source judgement?
Tolerance of uncertainty?
The ability to frame a question?
Experience of unusual cases?
The knowledge needed to notice when something is wrong?
If AI removes the activity but the capability still matters, then the capability needs somewhere else to develop.
That was one of the central questions emerging from the wider Neighbourhood inquiry, and it sits behind the earlier question, The Work Got Better. What Happened to the Human?
It also sits directly behind another piece in the wider series: Can We Design Against Deskilling Without Taking AI Away?
The critical-friend journey gives both questions another foundation.
A good helping relationship does not merely make today’s task easier.
It should leave the learner increasingly capable of seeing, learning and judging.
The AI-era educational problem is therefore not:
How much AI should students use?
It is closer to:
What should the person still be able to understand, notice, explain, verify, adapt, recover and own in a world where AI can do more of the visible work?
That question has gradually become an educational response in this project.
USE AI
because the gains are real.
Learn what the systems can do.
Give them bounded roles.
Use them to extend capability.
Automate what deserves automating.
Do not build an educational future around artificial abstinence.
But then:
KEEP THINKING
because assisted performance is not the same thing as retained capability.
Keep enough contact with the reasoning to recognise error.
Protect opportunities to frame before accepting a frame.
Verify when verification matters.
Practise the capabilities that still have to exist somewhere.
Know what has been delegated.
Be able to explain what you are standing behind.
And then the third move:
STILL THINKING?
because even that arrangement will move.
The learner changes.
The model changes.
The task changes.
The consequence changes.
A sensible boundary today can become ceremonial tomorrow.
A task that once needed protected practice may cease to matter.
A new failure mode may appear.
A better tool may make an old rule pointless.
So learners also need to be able to judge the arrangement itself.
Not only:
Did I use AI well?
But:
Is this still the right division of work?
What evidence would make me change it?
What happens if the system fails?
Where is authority sitting now?
What did I stop practising?
What became possible?
What became invisible?
Who is affected?
When should this rule be retired?
This is not a claim that a thirty-lesson curriculum has been proven effective.
It has not.
The evidence supports the problem more strongly than it supports any particular educational intervention.
That distinction matters.
A curriculum can be evidence-informed because it responds to a recurring, well-supported problem.
Its own effectiveness still has to be tested.
Learners have to experience it.
Demonstrate capability.
Transfer that capability elsewhere.
Retain it over time.
Change behaviour.
Expose where the curriculum itself is wrong.
At some point, the educational response has to become part of the inquiry.
That may be the most important meaning of Still Thinking?
The question mark applies to the curriculum too.
So the answer to what people should still be able to do cannot be a permanent list.
The current direction is more modest.
After AI helps, a person should retain enough capability to participate meaningfully in the judgement they still own.
Enough to understand what matters.
Enough to challenge.
Enough to verify.
Enough to adapt.
Enough to stop.
Enough to recover.
Enough to know when more AI would help.
And enough to know when it would not.
That is not a defence of human effort.
It is a defence of usable human judgement.
And if we are serious about that, the final question has to apply to our own answer:
What would make us revise even this?