What We Actually Know and What We Do Not — Handling Career Predictions

Sentences That Deliver Predictions as Facts
What the 47 Percent Figure Actually Said
The Same Warning Was Made in 1964
Why Predictions in This Field Miss
Separating the Known from the Unknown
Acting Without Knowing
Further Reading

Sentences That Deliver Predictions as Facts

Search this topic and two kinds of writing come back. One tells you not to worry: tools have always changed, developers have always survived. The other tells you that unless you do a particular thing now, you will be left behind. The expressions are opposites, but the structure is identical. Both speak about a future nobody can verify as though it had been verified.

Readers can feel this. So the reassurance does not last and the warning does not convert into action. What remains is fatigue.

This series stands somewhere else. Nobody knows how this plays out. The uncertainty is not a rhetorical hedge; it is the actual state of things. And there is still work worth doing inside that state. The job of this first part is to draw the line between what we know and what we do not as precisely as we can, before choosing any of that work.

What the 47 Percent Figure Actually Said

Start with the most cited number in this field. In 2013, Carl Frey and Michael Osborne of the Oxford Martin School published "The Future of Employment," analyzing 702 detailed occupations in the United States and reporting that about 47 percent of total US employment falls in the at-risk category (Oxford Martin School publication page, read 15 August 2026).

For the decade that followed, that number traveled under a different sentence: half of all jobs will disappear. But what the study computed was how susceptible each occupation is to computerisation — not how many people would be let go and when, but how close the tasks composing that occupation sit to a technically automatable shape.

Susceptibility and forecast are different objects. Susceptibility is a statement about a property; a forecast is a statement about an event. Between "this can be automated" and "headcount fell" lie adoption, regulation, cost, and organizational redesign. The popular sentence skipped every one of those steps. That does not make the study wrong. It means the thing we have been quoting is not the thing the study said.

The Same Warning Was Made in 1964

This warning is not new. On 22 March 1964, a memorandum titled "The Triple Revolution" was delivered to President Lyndon B. Johnson. It warned that automation would create a system of almost unlimited productive capacity while cutting the number of workers needed and raising the skill level required, producing steadily rising unemployment (overview of the Triple Revolution document, read 15 August 2026).

What is interesting is that the verdict is still split. By the same source, the sociologist Daniel Bell judged those projections illusory, while Martin Ford argues they were not wrong in direction, merely early. Sixty years on, one prediction still has no agreed scorecard.

The lesson to take is not "they were wrong then, so they are wrong now." That is another baseless prediction. The lesson is twofold. First, this class of prediction is extremely hard to grade. Second, there has never been a shortage of people willing to state it with confidence.

Why Predictions in This Field Miss

Past automation predictions have gone wide for reasons that repeat. This is a retrospective on those forecasts, not a claim about the future.

They treat tasks and occupations as the same thing. An occupation is a bundle of tasks. Remove a few and the occupation is usually reorganized rather than deleted. Industrial history contains many cases where verification, coordination, and exception handling flowed into the space an automated task vacated. Which effect is larger varies by case and is hard to know in advance.
They leave out the demand response. When making something gets cheaper, more people try to make it. Whether the added volume offsets the removed hands differs by field, and that too goes uncomputed beforehand.
They treat job boundaries as fixed. A backend engineer ten years ago and a backend engineer today do different work under the same name. What disappeared was not the occupation but its old definition.
They extrapolate from the most visible capability. The distance between what is impressive in a demo and what an organization will actually pay someone to own has consistently been longer than expected.

None of this ends at "so nothing will happen." This time may genuinely be different. But when a confident number arrives, this list gives you grounds to ask which of the four traps it has a foot in.

Separating the Known from the Unknown

So let us divide, honestly, what can be treated as known right now.

Closer to observation. The cost of producing a first draft of a well-defined piece of work has fallen sharply. Familiar code patterns, formulaic documents, and exploratory lookup are handled far faster than a few years ago. As a consequence, the share of time spent on review and verification has gone up. These are not forecasts; they are changes you can find in your own work log today.

On the unknown side. Whether the total number of engineering jobs rises or falls, on what timescale, in which layers and which domains first, and how the path from newcomer to senior gets rebuilt. Every sentence currently available on those four is a prediction. Forcefulness of tone has no relationship to strength of evidence.

Drawing the line this way changes how you read. From writing that reports observations, take the facts. From writing that offers predictions, take only the author's assumptions.

Acting Without Knowing

That leaves the practical question. If you cannot predict, what do you move on?

There is one method. Do not try to be right about the future; choose actions that retain their value across many futures. The test fits on one line: does this action still pay off if my prediction turns out wrong?

A study plan built on the assumption that a particular tool will dominate the next decade is mostly discarded when that assumption fails. Meanwhile, the ability to read code someone else wrote and find what is wrong with it, a feel for how systems break, a financial runway measured in months, people outside your company who know your work, and an identity that is not entirely staked on a job title — these retain value under every scenario. They pay whether the industry convulses or nothing much happens at all.

Anxiety is often unbearable not because the risk is large but because there is no answer, so people go looking for sentences that sound like answers. This series will not hand you one. Instead, across nine more parts, it works on how to choose the next step while no answer exists. The goal is not to recover certainty. The goal is to be able to move without it.