The real AI productivity gain is scope, not speed

AI can make the old task faster. The more consequential change is that it can make more of the surrounding problem economically possible for one experienced person to understand and act on.

Most AI productivity conversations still measure the wrong thing. They ask how much faster someone can write an article, summarise a document, produce a presentation or turn one format into another. Those gains are real, but increasingly they feel like the least interesting part of what is changing.

The bigger change I keep noticing is this: AI is expanding the amount of a problem one experienced person can realistically hold. That is not the same as doing the old job faster. It changes the practical scope of the job.

I saw this recently when I was pulled into a live campaign at work. The campaign itself was already moving: there was a strategy deck, creator activity, paid-media thinking, a content audit, several handovers, multiple stakeholders and a set of recommended articles. The editorial part had not quite been turned into an operating plan.

On paper, my role could easily have collapsed into something simple: here are the content gaps, go and write the articles. But that was not actually the problem. The problem was working out what the campaign needed from editorial in the first place.

Which recommendations were genuine gaps? Which existing pieces could be reused? Which ideas overlapped? Which content needed to exist in the first phase and which could wait? How should practical advice connect to creator work? Who owned the editorial decisions? What did the campaign plan actually commit to, versus what had merely been suggested in an audit?

None of those questions is difficult because the answer requires brilliant prose. They are difficult because the answer is scattered across documents, conversations, timelines and implied ownership. That is the part AI made dramatically easier.

The expensive work was never only the writing

Before AI, I could have done this work, and that matters. I do not want to turn a familiar editorial skill into something that supposedly did not exist until a model arrived.

An experienced editor has always been able to read the deck, interrogate the brief, challenge duplication, ask what the audience actually needs, work out which pieces should be commissioned, and push back when ten suggested outputs should really be four stronger ones. But being able to do something and being able to do it economically inside a normal working week are different things.

The limiting factor was often not capability. It was coordination cost. To reconstruct the campaign manually, I would have had to keep reopening the strategy deck, cross-reference the content audit, trace the handover, hold stakeholder comments in my head, compare proposed articles, map them against campaign stages, work out what was already covered, then translate all of that into messages, decisions and a commissioning plan.

That is possible, but it is also slow, cognitively expensive and easy to abandon when several other pieces of work are competing for attention. That is where a role gets quietly narrowed: you stop asking what the campaign really needs and start doing the visible task in front of you. You write the articles.

AI reduces the cost of making sense of the mess

The most useful thing AI did in this case was not write anything publishable. It helped me reconstruct the shape of the problem quickly enough that I could exercise judgement over it.

I could use it to compare the campaign plan with the audit. I could ask whether the audit was a commissioning list or a pool of recommendations. I could test whether two proposed articles were genuinely different user needs or the same problem expressed twice. I could trace the campaign phases and see that not every piece was required at once.

That let me turn an unclear handover into a much cleaner sequence:

campaign strategy → content audit → commissioning plan → stakeholder agreement → briefs → production

The sequence sounds obvious once it is written down, but it was not obvious when the information was distributed across the project. This is what I mean by scope rather than speed: the old productivity framing says AI helps me produce the article faster, while the more consequential version is that AI makes it feasible for me to look one or two layers above the article and ask whether it should exist at all.

The scope gain is upstream

I have written before that editorial jobs are moving upstream. I have also argued that the unit of capability inside a content team is changing, and that AI makes job descriptions less useful because one person can investigate much more of the surrounding problem than their title suggests.

This experience made the mechanism behind those arguments feel more concrete. The move upstream does not happen simply because AI is clever; it happens because AI removes enough friction from the surrounding analytical work that senior editorial judgement can be applied across a wider surface area.

An editor can move from:

• writing the asset
• to challenging the brief
• to comparing the brief with existing content
• to deciding what should be commissioned
• to sequencing the work against a campaign
• to shaping what stakeholders actually need to approve.

That is not a promotion generated by software, and it is not evidence that every writer should suddenly become a campaign strategist. It is a reduction in the cost of participating intelligently in adjacent decisions. That distinction matters.

This is not the same as doing five jobs

There is a bad version of this argument: AI makes one person capable of replacing the writer, strategist, researcher, project manager, designer and analyst, so organisations should simply collapse the team around them. I do not believe that. What AI gave me in this campaign was range, not omniscience.

I could understand enough of the media plan to see how editorial timing depended on it, but that did not make me the media lead. I could understand enough of the landing-page model to work out what content it needed, but that did not make me the web owner. I could understand enough of the campaign strategy to shape the commissioning response, but that did not make me the campaign lead.

The specialists still own their disciplines. What changes is the quality of the handoff between them, because I can arrive with more of the problem already understood. AI makes professional boundaries more permeable without making them disappear, and that can be a very good thing.

Seniority starts to look like problem ownership

There is a career implication here that I think matters more than the usual “learn AI or fall behind” advice. At a junior level, a lot of work is defined by tasks: write this, research that, update this page, produce these assets.

At a more senior level, the work becomes increasingly ambiguous. The brief is incomplete. Stakeholders disagree. Two plans contradict each other. Nobody has quite claimed ownership. The obvious output is not necessarily the right output.

The valuable person is often the one who can make sense of that without creating more confusion. That ability existed before AI; what AI changes is how much ambiguity one person can process before the coordination overhead becomes the work itself.

So the senior AI advantage may not be “I can produce more”. It may be: I can get to the real shape of the problem faster, and therefore take responsibility for a more useful part of it. That is a very different productivity story.

The gain is not output. It is avoided waste.

There is another reason conventional productivity metrics miss this. Suppose an audit recommends ten possible articles, and AI helps me write those ten articles 40 per cent faster. That is easy to measure: ten assets, fewer hours.

But suppose AI helps me establish that the campaign really needs four new articles, two refreshed pieces and existing content for the rest. The productivity gain is no longer visible as output; it appears as work that never had to be done.

No duplicated briefs. No unnecessary articles. No extra editing rounds. No unnecessary design or publishing work. No later content clean-up because the estate became bloated.

This is why I increasingly think the most important AI productivity number may be impossible to see in a dashboard. It is the amount of low-value work an experienced person was able to prevent because they had enough time and context to challenge the task before production began.

This depends on judgement being real

There is an obvious weakness in the argument: AI can also make someone confidently wrong at a larger scale. A model can rationalise a campaign into four pieces when ten really are required. It can flatten meaningful distinctions, produce a neat ownership map that does not match the politics of the organisation, or make a speculative interpretation feel settled because the prose is organised.

That is why the human role does not shrink as the scope expands. It gets more demanding. You need to know which conclusions come from evidence and which are inference, recognise when a document describes a recommendation rather than a commitment, and know when to stop analysing and ask a stakeholder.

You also need to understand enough of the organisation to know that the cleanest answer on paper may not be the answer that can actually move. The models make more of the terrain visible; they do not decide where the boundary of your authority sits.

Managers still define what matters

This is also why I do not think AI makes management less important. In the campaign example, I could reconstruct a plausible editorial route, but I still needed people around the project to confirm campaign timing, creator dependencies, ownership and what had already been agreed.

AI can help me answer:

What does the material suggest we should do?

A manager or stakeholder still needs to help answer:

Which of these things actually matters here, now, inside this organisation?

That is a healthier division of labour than using managers as document routers and task distributors. The individual gets more autonomy to investigate and propose, while the manager spends more time setting direction, priorities and boundaries. AI makes the route easier to find; it does not choose the destination.

Organisations may be underpricing this capability

There is a structural problem hiding here. If someone has the title “content writer”, an organisation may still evaluate them mostly on content output even while the most valuable work they are increasingly doing happens before the writing starts: synthesising evidence, preventing duplication, shaping briefs, connecting campaigns, identifying missing information, proposing architecture, making work easier for specialists downstream.

That creates a mismatch between the visible role and the useful role. The person appears to be producing fewer articles because they are spending more time making sure the right ones exist, and a crude productivity measure could interpret that as lower output.

A better measure might notice that the campaign became clearer, duplicated work disappeared, stakeholders had a recommendation to respond to, production happened in the right order and the final content fitted the wider system. That is much harder to count, but it is also much closer to the value of senior editorial work.

Scope needs a stopping rule

The danger is obvious once you see the opportunity. If AI lets you understand more of the problem, every problem starts looking like yours.

It is very easy to become the person who can “just help” with the strategy, the workflow, the research, the site, the measurement and the stakeholder messaging until you are effectively running six workstreams without formal authority over any of them. That is not leverage. It is uncontrolled scope creep.

The useful rule for me is becoming: use AI to understand more of the system than you own, but be explicit about which part you are actually responsible for changing. In the campaign, that meant making the editorial layer coherent, not solving the entire campaign. That boundary is what turns expanded capability into useful seniority rather than burnout.

The real productivity gain is optionality

I still use AI to write faster. I would be pretending otherwise if I said production speed did not matter, but speed is increasingly the secondary gain. The more important one is that work which used to be technically possible but operationally unrealistic becomes feasible.

That can mean reading across a campaign instead of only your brief, testing whether ten commissions should really be ten, connecting a content audit to the paid-media sequence, turning a messy handover into an editorial operating plan, or arriving at a stakeholder conversation with a recommendation instead of a request for instructions.

None of those capabilities began with AI. What changed is that an individual can now use them often enough, quickly enough and cheaply enough for them to become part of the everyday job rather than the occasional luxury of having a quiet week.

That is close to the same economic shift I described in AI didn’t create Editorial Intelligence. It made it economically viable., but at the level of one person’s day-to-day operating range rather than an editorial system as a whole. It is also why I keep returning to working in the order the problem demands: once the cost of understanding the surrounding problem falls, the old sequence of brief → asset → production stops being the only economically realistic one.

That is why I think the most consequential AI productivity question is not:

How much faster can I do my current task?

It is:

How much more of the real problem can I responsibly take on before I need to hand it to someone else?

That is a harder thing to benchmark. It may also be the thing that changes the job most.