The comfortable version of this argument says that machines will do the routine work and humans will do the creative work. It is comfortable because creative is vague enough that everybody can locate themselves inside it.
We think the honest version is narrower, more specific, and more demanding.
What is actually being absorbed
Be precise about the category. What has become cheap is production: drafting, summarising, restructuring, routine analysis, first-pass code, the competent version of most documents. Work whose difficulty was in the assembly rather than in the judgement.
That is a very large share of knowledge work, and pretending otherwise is not useful to anyone. If your value has been the ability to produce a competent draft faster than a colleague, that value is falling and it is not coming back.
What has not become cheap is deciding whether the competent draft is right, and whether it should have been the draft at all.
Three tiers
| Tier | The work | What automation does to it |
|---|---|---|
| Production | Drafting, summarising, routine analysis | Cost collapses toward zero |
| Verification | Noticing that the plausible output is wrong | Value rises with the volume of output |
| Direction | Deciding what should have been made at all | Unaffected, and increasingly decisive |
The premium moves up the stack. And both of the upper tiers have a property the lower one does not: they require having held the problem long enough to have an opinion.
Verification is real work
This is the part that gets underrated, because it looks like reading.
A system that produces fluent output does not produce a signal about when the fluent output is wrong. Fluency is uniform. The confident correct answer and the confident incorrect answer arrive in the same voice, at the same speed, with the same formatting.
Catching the difference requires something specific: knowing the domain well enough to have expectations, and attending closely enough to notice when an expectation is quietly violated. It is adversarial reading. You are not following the argument; you are looking for the place it stops being true.
And it degrades badly under divided attention. This is the point at which the research in this series becomes directly relevant rather than analogous. Sophie Leroy’s studies of attention residue found that performance suffers when part of your attention remains with a previous task — and verification is precisely the kind of work where diminished performance does not announce itself. A distracted reader does not notice fewer errors and know it. They notice fewer errors and feel finished.
What we are inferring. Leroy's studies measured task performance in general, not error-detection specifically, and were conducted long before the current tools existed. Applying them to the verification of generated output is inferred. The mechanism is plausible and it is not demonstrated. We would rather flag that than let a real finding do work it was not asked to do.
Direction is the harder tier
Verification asks is this right. Direction asks should this exist.
That is the second position on the Arc — deciding what deserves depth — operating at the scale of a team or a company rather than an afternoon. And it is the tier most obviously immune to automation, for an uninteresting reason: it requires having a stake in the answer.
The interesting part is that direction has become harder, not easier, as production got cheap. When making something took three weeks, the decision to make it received scrutiny proportionate to the cost. When it takes twenty minutes, the decision receives no scrutiny at all, and organisations fill with competently produced artefacts that nobody decided to want.
There is more output and less deciding. Those two facts are related, and the second is the expensive one.
Why not just say “creativity”
Because you cannot act on it.
Nobody can schedule creativity, practise it deliberately, or say at the end of a day whether they did any. It functions as a comfort rather than a plan.
Noticing that a plausible answer is wrong is different in every respect. It is observable — either you caught it or you did not. It is practisable. It clearly depends on domain knowledge and on time spent with the problem. And it produces an artefact: the note saying what is wrong.
If you want a concrete answer to what to get good at, that is a better one than creativity, and it is uncomfortable in a way that suggests it might be true.
What this means for a working day
If the valuable tiers require sustained attention, the reactive default is exactly backwards.
The default working day — many channels, short intervals, constant availability — is optimised for a world where the constraint was throughput of production. That constraint is dissolving. The day has not changed to match.
What matches: fewer, longer, deliberately protected blocks, most of them spent not producing. Reading closely. Checking. Deciding what should be made. That looks less busy than the alternative and produces work that is more likely to be correct, which is a trade many workplaces are not yet set up to reward.
We are not going to pretend that is easy to act on inside an organisation that measures visible activity. Protecting time has real costs, and they fall on you rather than on the argument.
The uncomfortable part
An argument that sustained attention is becoming more valuable is also an argument that its distribution matters.
The ability to protect an hour is not evenly available. It depends on the kind of job you have, on whether you have caring responsibilities, on whether your work is measured by responsiveness, on whether you have a door. If the premium moves toward capacities that require uninterrupted time, then people whose conditions do not permit uninterrupted time are on the wrong side of a widening gap.
We do not have a solution to that and we are not going to gesture at one. It is worth stating because a piece arguing that focus is becoming more valuable should be honest that this is a claim with distributional consequences, not simply good news.
What Eonara is betting on
Plainly, so it can be judged: that the scarce human capacity in knowledge work is not producing things, and not having ideas, but holding a hard problem in mind long enough to notice what is wrong with it.
When the cost of producing a plausible answer falls to nothing, the person who can tell that it is wrong becomes the expensive one.
If that is right, then an hour of protected, single-threaded attention is not a productivity technique. It is the working condition for the part of the job that still belongs to you.
That is what the product is. It is also, at this stage, a bet rather than a finding, and the last piece in this series is about what acting on it actually costs.