The history of software can be understood as a progression in what the human must specify.
First, we specified the steps.
Then, we specified the outcome.
Next, the system will predict which outcome is needed and simply produce it.
That progression is the movement from tools to agents to absorption.
Traditional software begins with instructions. The person must translate an intention into the sequence the machine understands. Open the application. Find the correct screen. Enter the information. Choose the settings. Press the button. Check the result. Move it somewhere else.
The software may execute every operation perfectly, but the human remains responsible for decomposing the outcome into steps.
This is why sophisticated software can still feel like work. The machine performs the clicks after we tell it exactly which clicks should happen. Execution has moved to the computer. Coordination remains in the mind.
Artificial intelligence changes the unit of instruction.
Instead of specifying the steps, we can specify the outcome.
Write the report. Reconcile the accounts. Find the scheduling conflict. Publish these articles as drafts.
The human no longer has to know every operation required to produce the result. The machine can choose the steps, use the necessary tools, recover from ordinary problems, and verify that the requested state exists.
This is a major transition. But it is not the end of the progression.
The system still waits for us to recognize that an outcome is needed.
We must notice the unfinished work, remember the obligation, open the interface, formulate the request, and review what happens. We have stopped managing the steps, but we are still managing the arrival of the outcome.
The next transition occurs when the system can predict the outcome itself.
Not predict in the mystical sense. Not consciousness. Not unrestricted authority. Prediction means that the system maintains enough context to recognize the expected next state.
It knows the twelve articles are finished. It knows the publishing policy says completed articles become WordPress drafts. It knows which category and tags belong to the series. It knows draft creation is reversible and inside its authority. It creates the drafts, verifies all twelve, and returns only if something genuinely surprising happens.
Nobody has to remember to ask.
That is absorption.
The progression looks like this:
You specify the steps.
The machine executes your procedure.
You specify the outcome.
The machine determines the procedure.
The machine predicts the outcome.
The machine recognizes what the situation requires and maintains the expected world.
Each stage removes a different burden.
The first removes physical execution.
The second removes procedural coordination.
The third removes the need to remember, initiate, and supervise the ordinary outcome.
This is why the future of AI cannot be measured only by the quality of its answers. A perfect answer still leaves the human responsible for asking the question at the correct moment.
The deeper measure is how much obligation disappears from attention.
Of course, prediction must have boundaries. A system should act quietly when the outcome is highly predictable, authorized, reversible, and easy to verify. It should surface uncertainty when the consequence is meaningful, the evidence is weak, the action affects another person's rights, or recovery would be difficult.
Absorption does not mean that the system does whatever it wants.
It means that the system learns which outcomes do not require us.
That is also how the subconscious works. It does not ask permission for every breath or heartbeat. It maintains the predictable body beneath consciousness and interrupts only when the expected state can no longer be preserved.
AI will move through the same design progression.
At first, we will admire how quickly it follows instructions.
Then, we will value how reliably it delivers outcomes.
Eventually, we will stop noticing the predictable outcomes altogether.
The interface disappears.
The outcome remains.