The Interim Species

The Line That Wouldn't Leave

A line from an advertisement lodged itself in my head this week: that we have always had an AI partner, but never an AI doing partner. It is a good line — you can almost hear the marketing department exhale — but the reason it stuck is that it names a real threshold. A partner you talk to is a tool. A partner that acts on your behalf is something else. It is the difference between a calculator and a colleague.

The moment you notice that shift, a second question arrives on its heels, and it is a less comfortable one: if the machine is now doing, what is the human doing?

The instinctive answer is "less" — that this is the beginning of the replacement we have all been told to expect. I want to argue the opposite. What if the software engineer was never the destination at all — but an interim stage, a pupa, that we are only now emerging from? And what if the thing emerging is more human, not less?

The Musician Who Becomes a Conductor

There is a pattern in the way expertise moves that predates software entirely, and it is worth sitting with before we reach for the alarm.

When a gifted instrumentalist becomes a conductor, something is genuinely lost. The hours of scales, the fingertip calluses, the reflexes that only come from thousands of repetitions — those fade. The conductor no longer plays. Measured against the specialist they were, they are, in one narrow sense, less skilled. They have traded depth at one instrument for command of many.

We have watched the same trade in clinic after clinic. The senior technician who is promoted to run the department spends their days in coordination, not on the bench. The artisan who becomes the foreman stops holding the chisel. The surgeon who pioneered a technique a decade ago now spends most of his week in administration — a scalpel surgically replaced with a pen.

We call it task shifting, and we tend to frame it as a reward: a promotion. But the people living through it often describe a quiet grief. It is not simply that they loved the work. It is that they were the work. The surgeon did not merely enjoy operating; "surgeon" was the answer to the question of who he is. When the theatre gives way to the boardroom, what is lost is not a skill but an identity — the reps, and, with them, a piece of the self.

The fear about artificial intelligence is this same grief, scaled up and stripped of the promotion. If the machine takes over the instrument, the musician does not get to become the conductor — the musician leaves the orchestra. That is the story we have been told to expect. It is also, I think, a misreading of what is actually happening.

The Interim Species

Consider why humans were ever programming at all.

For most of computing history, machines could not understand us. They could only follow instructions written in their own language — rigid, literal, unforgiving. So the human who wanted to build something had no choice but to become bilingual: to translate a human intention ("let the nurse see which patients are at risk of a dosing error") into the machine's dialect (a data model, a loop, an error handler). The programmer is not a natural species. The programmer is a translator, forced into existence by the inadequacy of the machine's ears.

Hold that thought and the whole anxiety reorganises itself. We have been treating "software engineer" as an eternal job description, a final form. But look at what the job actually is: the human spends most of their working life thinking like a computer — managing memory, tracking state, arguing with a compiler about a semicolon. That is not the human contributing what humans are uniquely good at. That is the human doing the computer's job, because for decades, no one else could.

The software engineer — in the specific sense of a person who manually instructs the machine — may be the caterpillar stage. Not because the caterpillar is worthless. Metamorphosis does not replace the caterpillar; it is the caterpillar, transformed. Every future capability was always latent in the crawling thing. What changes is that the stage the technology was waiting for has arrived. The machine can now understand intent directly enough that the human no longer needs to be fluent in the machine's grammar to get anything built.

We were never meant to be programming specialists. Programming was the interim — the holding pattern while the technology caught up to us.

Let Computers speak "Computer language"

If that is right, then agentic AI is not the replacement of the software engineer. It is the specialisation of the work the engineer was always doing too much of.

Watch what an agent actually absorbs. The agent handles the computer's dialect: the boilerplate, the state, the syntax, the long chain of mechanical steps between an intention and a running system. It does the computer talk. It has, in effect, the machine's native fluency that a human can only ever borrow.

What remains with the human is the part the machine has never been able to source from within itself: knowing what should be built, and why, and for whom. Specifying intent. Judging trade-offs against values a computer does not hold. Reading the messy human context — the nurse's actual workflow, the patient's actual fear, the regulator's actual posture — and translating it into something precise enough to build against. The human translates the human world into instructions. The computer translates instructions into systems.

This is why the conductor analogy breaks down in the right direction. The musician who becomes a conductor does lose their instrument — but the software engineer who stops hand-writing every line does not lose their judgement; they finally get to use it. For most of the profession's history, the judgement was rationed, because so much of the day disappeared into translation overhead. Agentic AI returns that time. It does not hollow out the role; it removes the part of the role that was never the human's to lose.

We have been asking the wrong question. "Will AI replace software engineers?" assumes the engineer's value lived in the typing. It did not. It lived in the translating. And translation — from the human world to the machine, and back — is the one thing that gets more valuable as the machine gets more capable.

Making Healthcare More Human

Nowhere does this distinction matter more than in healthcare, which is where we spend our days.

Medicine is, at its core, one of the most stubbornly human disciplines there is. The best outcomes grow out of things no machine has ever held: a clinician who notices what the patient did not say; a judgement call made at 3 a.m. with incomplete information; the trust that lets a frightened person follow a difficult treatment plan. And yet the way healthcare is actually run has drifted steadily away from the human and toward the machine's terms.

A single nurse can spend a third of a shift translating — feeding the record, hunting the alert, reconciling the chart. A regulatory officer in South Africa, as we have written before, loses hours a week to visiting six browser tabs because no one built a better way to be told. This is not a shortage of compassion. It is a surplus of computer talk, performed by humans who trained for something else.

Our vision at Medicaputare is a more human healthcare. For a long time, that phrase sounded like it could only mean less technology — fewer screens, fewer systems, a retreat. We no longer think that is what it means.

A more human healthcare is one where the human-facing work is the work people actually do, because the machine-facing work has been handed to the machines. Where the agent watches the SAHPRA pages so the compliance officer can think about the submission. Where the model tracks the dosing trend so the clinician can sit with the patient. Where nobody spends an afternoon converting their knowledge into a form the computer can stomach, because the computer has learned to digest it directly.

That is not the replacement of the specialist. It is the specialisation of everyone. The computer does computer talk. The human does human. And the patient gets both, unrationed.

None of this requires us to believe the machine has become wise. It only requires us to stop mistaking the interim stage — the caterpillar, the coder, the human forced to speak machine — for the final one. Metamorphosis is not a loss. It is the whole point of having crawled.