One in three jobs is not a job

Fluxus research note, August 2026

A year ago I argued that AI was not destroying knowledge work but reorganising it, and that three kinds of role were forming as it did: Navigation Experts, whose value is knowing good output from bad; Amplification Natives, who work natively inside AI-assisted workflows; and Integrators and Governors, who build the business integration, evaluation, and compliance that make any of it dependable.

That was an argument from economic history and one scatter plot. A year on, we did some research at Fluxus to see if it’s showing up yet in what employers are advertising. It is, in all three cases. What is striking is the shape each has taken: one became a job title, one became piece-work, and one has no name at all.

Navigation Experts are being bought by the task, not employed

Search for senior AI work and 29% of what comes back is AI training, annotation or model evaluation paid by the unit. A thousand postings in the three and a half thousand we looked at, sitting between permanent engineering roles and senior contract mandates. It is the most common reason a listing turns out not to be a job at all.

The titles are the interesting part. A sample, all of which turned out to be per-task work:

Senior Investment Lead, Remote Investment Banking Analyst, Remote Quantitative Analyst (Finance), Cybersecurity Consultant (Remote), Remote Senior Backend Engineer (Python/FastAPI), Remote Physics Researcher (PhD), Remote Mathematics Expert (PhD), Remote Chemistry Expert (PhD), Machine Learning Analyst (Remote), Security Engineer

One title in twenty-four makes the arrangement explicit: Equity Research Expert (Train AI Models Part Time!)

This is not an attempt to recruit cheap generalist labour. It targets investment banking, quantitative finance, cybersecurity, backend engineering and doctoral-level physics, chemistry and mathematics. It is a direct bid for exactly the expert judgement that makes a Navigation Expert valuable, and it buys that judgement by the hour rather than employing it.

The mechanism has its own infrastructure. One poster describes itself as drawing from more than five hundred sources, including every significant annotation platform, and refreshing every sixty seconds. Among its listings was one presented as a senior cloud engineering contract at 100 to 230 US dollars an hour, naming no client, with responsibilities generic enough to fit any backend role, and the word ‘payout’ where a rate would normally appear.

So the category is real and the demand is real. But the market has found a way to consume the capability without hiring the person, and it reaches those people through the same channels their prospective employers use.

Integrators and Governors have become job titles

This category was the most speculative of the three. It is now the most concrete.

Roles that name AI governance, adoption or enablement directly appear as their own titles:

Director AI Governance; AI Governance & Responsible AI Consultant; Security Governance & Policy Lead, AI; Data Management Analyst, AI Governance (12 Month FTC); Senior AI Adoption Technical Consultant; Senior AI Learning and Adoption Lead; AI Enablement Lead; AI Project Manager, Learning & Enablement; Change and Communications AI Consultant; AI Adoption Lead

These are not engineering roles and they are not strategy roles. They are the operational layer: getting a capability into an organisation, keeping it inside policy, and teaching people to use it. Two years ago most of these titles did not exist. (I was once told a little tongue-in-cheek that longer job titles denote seniority with a new word appended every couple of years, which looks to be the case here.)

But a detail worth noting is how many are temporary. Fixed-term contracts, twelve-month FTCs and interim mandates are common in this group. Organisations are treating the integration work as a project with an end date rather than a permanent function, which is what you would expect during a reorganisation and not what you would expect if these were becoming standing roles.

Alongside them sits a larger body of work that is adjacent rather than technical: Copilot rollouts run as IT projects, AI awareness campaigns, programme management of ‘AI and cloud transformation’, reporting and dashboards. Judged as engineering jobs these look like noise but judged as evidence of reorganisation they are the substance of it. This is what adoption looks like from the inside, and it is where a large share of the spending is going.

The platforms are selling what employers have withdrawn

Employers have spent two years pulling people back to the office, and the job ads show it. Two thirds require office presence: forty percent hybrid, twenty-five percent on-site.

The piece-work listings are going the other way, and doing it loudly. Nineteen percent of them put "Remote" in the job title itself, against five percent of everything else. Almost four times the rate, in the one field a candidate reads before deciding whether to open the ad at all.

Set the two together and the arbitrage is obvious. A quantitative analyst or a doctoral chemist who wants to work from home is being told by traditional employers that the role requires three days in an office, and by the training platforms that it does not require an office at all.

The platforms are not competing on prestige or security, both of which they lose. They are competing on the single term their target market most wants and their competitors have spent two years taking away.

Amplification Natives are in the requirements, not the roles

The third category barely registers as a job, and where it appears it is a line in the requirements rather than the thing being hired for.

Six percent of job descriptions name AI-assisted development directly. ‘Experience with agentic AI frameworks, LLM-powered automation, and tools like Claude Code is a core requirement, not a nice-to-have’. ‘Strong advocacy for AI developer tooling such as Claude Code, Copilot, Windsurf and Gemini’. ‘Help define how AI coding assistants and agentic developer tooling are adopted’.

In job titles it is almost absent: six out of more than two thousand distinct titles, and most of those are Copilot roles, which is a product rather than a way of working.

There is a structural reason for the gap. Navigation Experts and Integrators bolt onto an organisation as it already is. You can hire a governance lead, or buy expert judgement by the hour, without changing how anything else runs. Amplification does not bolt on. Getting the benefit means changing how work is divided, reviewed and scheduled, which is an organisational change rather than a hire, and organisations are slow at those in a way they are not slow at recruitment.

Which suggests this category arrives mostly through new companies rather than through existing ones reorganising. That is the usual pattern during technology shifts, and it is why the effect stays invisible for a long time and then appears to be everywhere at once. Little in a set of job adverts will show it early, because by the time it is written into a job title the change has already happened.

Everything is moving faster than the hiring process assumes

Applicant counts, taken at the moment each posting was first seen (piece-work listings excluded so we can see the picture for the regular job market):

For a full day almost nothing is saturated. The median is five applicants, the top quartile sits at twelve, and one posting in a hundred has reached the point where the count stops displaying.

Then the first day ends and the market arrives at once. One posting in six is at a hundred or more. The top quartile jumps from twelve to sixty-nine. By the end of the week a quarter of everything is capped out and the top quartile stands at ninety-nine, which is to say the whole upper quarter of the market is saturated.

There is no gradual climb. A role is uncontested for a day and then it is over.

What this suggests

The reorganisation argued for a year ago is visible in what employers are advertising, but it is not arriving as a clean set of new permanent jobs.

Expert judgement is being purchased piecemeal, at scale, through channels that look like permanent recruitment. Integration and governance work has acquired names and is being staffed on fixed terms. The way of working that most defines the transition has no title at all, and probably will not have one until the companies built around it are large enough that everyone else is copying them.

The competition for expert judgement is no longer only between employers. It includes a well-funded market buying the same judgement by the hour, from the same people, and offering remote terms that most employers have stopped offering.