I’m not trying to turn every specialist into a developer. I’m trying to make sure they don’t have to become one.
As AI becomes more capable, an odd thing is happening.
To use that capability seriously, professionals are increasingly pushed toward databases, APIs, schemas, permissions, agent orchestration, tool configuration and technical environments that were never part of their profession.
A researcher wants to investigate a question.
A teacher wants to design learning.
An analyst wants to work with evidence.
A methodologist wants to understand how a system behaves.
They should not first have to become part-time software engineers.
The specialist should define the professional problem:
What matters?
What evidence is acceptable?
Which relationships need to be preserved?
What must remain uncertain?
Which functions should different specialised agents perform?
When does a human need to make the decision?
The AI system should handle much more of the technical detour required to make that possible.
That does not mean giving professionals a simplified, child-sized version of advanced AI.
It means giving them the full capability with an interface and scaffolding appropriate to their actual expertise.
And there is an important distinction here: hiding technical machinery should not mean hiding epistemic structure.
A professional still needs to see what is:
fact, inference, hypothesis, source evidence, unresolved conflict or system-generated suggestion.
That is part of their work.
Managing schemas, plumbing and orchestration usually is not.
We already have increasingly powerful models. The next step should not be requiring every serious user to learn more technical infrastructure just to reach that power.
It should be translating that power into professional environments where domain expertise remains the scarce and valuable resource.
Keep the specialist specialised.
Let the AI handle the technical detour.