Why frontier AI companies are paying up to $345,000 for editorial judgement

I keep saving editorial job adverts because they are becoming a useful way of seeing where companies think human value still sits as AI changes knowledge work. And, if I am honest, some of them also look like dream jobs.

This week I found a Lead Content Editor role at the DeepMind Institute. It asks for 14 years of experience in technical writing, journalism or editorial management, experience running an editorial desk, the ability to translate scientific and technical concepts, and confidence working with researchers and executive leadership.

The role is not framed as polishing finished thinking. The editor is expected to shape complex arguments, set the intellectual benchmark for the Institute’s voice, own fact-checking and citation checks, and act as the final editorial gate before publication. For US applicants, the advertised base salary is $220,000 to $305,000, plus a 25% bonus target, equity and benefits.

Then I found Roblox’s Manager, Editorial role. The experience threshold is lower at five-plus years, but the work is similarly revealing: write about product innovation, AI, safety, creators and users; work with executives and product leads; move between audiences; resolve conflicting feedback; and help clarify strategy when different parts of the organisation are not aligned.

The advertised San Mateo base salary is $186,620 to $228,550. Then there is Anthropic’s Executive Communications Writer, listed at $295,000 to $345,000, working with senior leaders and subject-matter experts on technically complex, fast-moving material.

Anthropic is also hiring a Developer Copywriter at $200,000 to $255,000. That role sits between technical content and marketing, working with DevRel and Claude Code leadership, developing launch narratives and helping set the developer voice.

A few expensive US technology jobs do not make a labour-market trend. But they do make a useful signal, because the interesting thing is not simply that these companies are paying writers and editors a lot of money.

It is what they are paying them to do.

I used to assume the really big marketing money sat somewhere else

Product marketing, growth, product, revenue and strategy are the functions I would normally expect software companies to pay heavily for because their relationship to commercial outcomes looks obvious. Editorial has traditionally appeared further downstream.

The product gets built, the strategy gets decided, the positioning gets agreed, and then somebody writes the article, speech, report, customer story or campaign copy that explains it. If that is the job, editorial probably should be cheaper than the functions making the underlying decisions.

But that is increasingly not what these descriptions show. The DeepMind editor partners with the Managing Director, shapes arguments with researchers and acts as the quality gate for evidence and citation. Anthropic’s executive writer is expected to help leaders communicate consequential ideas around research, products, strategy and public benefit.

Roblox wants an editor who can operate across AI, product innovation, safety and creator communities while helping different parts of the organisation reach alignment. That is not a production function waiting politely at the end of the process.

It is work happening much closer to the decisions.

The comparison with product marketing is what interests me

I do not think editorial is replacing product marketing, and salary bands are a crude way to compare unlike roles. But the overlap is difficult to ignore.

Roblox is also advertising a Principal Product Marketing Manager for Foundation AI at $263,710 to $300,940. The role owns positioning and messaging, turns technical capabilities into narratives, connects individual launches into a larger arc, brings market signals back into product conversations and is explicitly expected to find the “so what” inside dense technical material.

Anthropic lists senior product marketing roles at around $320,000, including Product Marketing Manager, Knowledge Work - Core Products. Product marketing still owns responsibilities that editorial normally does not: go-to-market strategy, adoption, enablement, commercial positioning and launch execution.

But these editorial and product-marketing roles occupy the same strategic neighbourhood. Both are being asked to understand complex technology, develop a point of view, connect individual developments to a larger narrative, move between expert and non-expert audiences, and create alignment around what something means.

For years, content often sat underneath product marketing. The product marketer worked out the positioning and the writer turned it into assets. At some frontier technology companies, the higher-value editorial roles look much less subordinate to that process.

They operate alongside it.

Maybe the money is moving upstream too

I wrote recently that editorial jobs are moving upstream. My argument was that AI makes production cheaper, which puts pressure on work built mainly around turning an agreed brief into competent copy.

AI can write a credible article, summarise an interview, generate a campaign structure, turn a report into social posts and produce variations faster than a human production team ever could. What it does not remove is the set of decisions that happen before the copy.

What is actually worth saying? What is true? What can we prove? Which evidence matters? What does a technical development mean to somebody outside the research team? How does today’s launch connect to the direction of the company? Which claim sounds persuasive internally but collapses as soon as an expert reads it?

Those questions sit upstream of production, and when I look at these highly paid jobs, that is where the human work seems concentrated.

The money appears to be moving towards the same layer as the judgement.

I want to be careful with that sentence. These vacancies do not prove that AI is causing editorial salaries to rise, and they are disproportionately US roles at wealthy technology companies where geography, company economics, talent competition, executive proximity and regulatory scrutiny all matter.

There are many editorial jobs that pay nothing like these figures. The narrower claim is that, at some frontier technology companies, sophisticated editorial and narrative work is being priced as strategically important work rather than cheap production.

That is worth noticing.

The companies building the technology are an interesting place to look

DeepMind and Anthropic are not companies trying to decide whether generative AI will affect knowledge work. They are building the technology. Roblox is building generative AI into creation itself.

That does not make these companies infallible. Being close to the technological frontier does not give an organisation magical knowledge of the labour market, and frontier companies can overhire, misread markets, copy one another and waste money like everyone else.

But hiring is still a form of revealed belief. Companies spend serious money on capabilities they believe they need, and organisations living closest to a technological shift can sometimes provide an earlier signal of how work may reorganise around that shift.

So I find this combination striking:

The companies closest to making content production cheaper are simultaneously paying heavily for human judgement around what gets said, how technical work gets interpreted and how the organisation’s story holds together.

That is not proof of where every company will end up. But if your organisation is assuming AI means editorial should simply become a cheaper production function, it is worth asking why some of the companies closest to AI appear to be organising differently.

They may be wrong. Or they may be earlier.

Narrative becomes more important when capability becomes harder to differentiate

There is a broader software problem underneath this. For a long time, product differentiation could carry a lot of the story: build something meaningfully better, explain the feature, show the benefit, sell it.

AI complicates that because capabilities move quickly, features get copied and almost every software company can claim some version of intelligence, automation, agents or transformation. Technical distinctions are also difficult for many buyers to assess, while products can change faster than the market’s mental model of what the company actually does.

In that environment, narrative is not simply decoration around the product. It becomes part of the interface between technical capability and market understanding.

What is this actually for? Why does this development matter? Why should I trust this company to do it? How does one product announcement connect to another? What does the company believe about the problem it is solving, and where is it going?

Those are partly product-marketing questions, partly corporate-communications questions and partly editorial questions. The organisational boundaries matter less to me than the capability underneath them: someone has to turn complexity into a coherent point of view without flattening the complexity in the process.

That is higher-leverage work than producing another asset from an already settled brief.

This does not mean writers win

That would be a comforting conclusion, and I do not think it follows. If anything, the shape looks more polarised.

AI can make routine production dramatically more accessible while organisations place a high premium on people who already know how to decide what good looks like. That connects directly to another argument I have been worrying about in Raising the floor is not building a ladder.

AI can help less experienced people produce professional-looking work sooner. That is the floor rising. But if routine production disappears, some of the repetition through which judgement used to develop disappears with it.

Put the two arguments together and you get an uncomfortable labour-market possibility:

competent production gets cheaper while accumulated judgement gets more valuable.

That would explain why the future can contain both pressure on ordinary content roles and unusually valuable editorial roles at the same time. The occupation does not have to disappear; it can stretch, with the bottom and middle compressed while the upstream end becomes more consequential.

Some companies may still be organising around the old bottleneck

This is the part I think software companies should pay attention to. If the editorial request still begins with “we need a blog”, “we need four articles”, “we need some social posts” or “we need a writer to support this campaign”, production is still being treated as the central problem.

That made sense when production was expensive. It makes less sense when an organisation can generate competent words almost instantly, because the harder problem becomes everything required to make those words worth generating.

Evidence, point of view, technical understanding, customer knowledge, editorial standards, market context and narrative coherence all sit before the asset. So do the harder judgements: knowing what not to say, recognising when the product claim is ahead of the evidence, and seeing when five separate launches should be understood as one strategic story.

This is why I increasingly struggle with “content” as the centre of the profession. Content is the visible output; the upstream work is what determines whether the output has any value.

This is increasingly what I mean by Editorial Intelligence

My working model has become:

signals → evidence → judgement → narrative → AI-assisted production → activation → response → new evidence

The expensive human work in the roles I am looking at appears heavily concentrated on the left-hand side. Understand the signal, get close enough to the technical work to know what it actually means, interrogate the evidence, exercise judgement and shape the narrative.

Then let increasingly capable production systems help that narrative travel further.

That is not really a defence of writing. It is an argument that the unit of editorial value is getting bigger than the writing itself.

Perhaps the dream editorial job is changing too

There is a personal reason these adverts caught my attention. For most of my career, the apparent route upwards from writing looked fairly predictable: write, become a senior writer, edit writers, manage a content team, then move further away from the intellectual work as your title gets more senior.

I have never been completely convinced by that definition of progression. These roles suggest another possibility: stay close to the work, but move closer to research, evidence, products, executives and strategy.

Become more valuable not because you supervise more production, but because you improve the quality of the decisions that happen before production. That is a much more interesting version of senior editorial work to me, and apparently, in some companies, a much more valuable one.

I would watch what the frontier companies do next

I am not predicting that $300,000 editorial jobs are about to become normal. They are not, and I am not arguing that narrative matters more than the product or that every software company should copy Anthropic, DeepMind or Roblox.

But companies close to the technological frontier are useful places to look for early changes in how knowledge work is valued. Right now, some of them are telling us something interesting: they are not treating the human contribution around writing as simply the production of sentences.

They are paying for technical translation, argument, evidence, judgement, executive counsel, cross-functional alignment and narrative. In other words, the things immediately upstream of the words.

If your response to AI is mainly to make writers produce more content faster, there is a chance you are optimising the layer that is becoming cheaper.

The more interesting question is what you are doing with the layer above it, because some of the companies building the future of the production technology appear increasingly willing to pay for that.