What Actually Makes AI-Adjacent Work Meaningful

What actually makes AI-adjacent work meaningful.

A 2025 YouGov survey found a majority of American workers said their job makes a genuine contribution to the world, but not an overwhelming one, and there was a smaller, more useful finding buried in the same data: people whose work involves little contact with anyone else, no emails, no meetings, were less likely to call their job meaningful than people whose work involves plenty of both.

That cuts against the instinct that a meeting-free day is the good day. It suggests meaning -> tracks with visible contact and visible impact! An assessor who reads a whole portfolio, start to finish, and one who only opens whatever a system flags, are doing what looks like the same job. Yet, they are not having the same day… Work does something no pay cheque does on its own:

• it is one of the main places most adults get to practise judgement,
• build a skill over years, and
• feel like part of something bigger.

Take that away from a job and you have removed a fair chunk of what made it worth doing at all, which is why several AI ethics frameworks treat meaningful work as a real sub-clause of respecting human autonomy, not a wellness afterthought bolted on.

Three questions worth asking about any redesigned role

Psychologists Richard Hackman and Greg Oldham built a model of this back in the 1970s, and it has aged better than most workplace theory from that decade. Three things, on their account, do most of the work in making a job feel like something rather than nothing:

• skill variety, whether it draws on a real range of what you can do;
• task identity, whether you see a piece of work through from start to end rather than handing off a fragment; and
• task significance, whether you can see who the work actually helps.

Take our assessor again. Reading a full portfolio uses judgement across several dimensions at once, ends in a decision she reaches herself, and connects directly to a person on the other end of it. Reviewing only the borderline cases a system has already sorted keeps a sliver of the judgement and strips out the other two entirely: she inherits someone else’s triage and never learns what happened to the cases she never saw. Whether that is a worse job depends entirely on how the split was designed, not on whether a system was involved at all. The same automation can free her to spend real time on the genuinely hard calls, or reduce her to rubber-stamping a queue. Nothing about “AI is now part of this role” decides which one happens. The design decision after that point does!

Two more, once other people enter the picture

Researchers Sarah Bankins and Paul Formosa recently added two further ingredients worth naming on their own, because AI-adjacent work strains both in ways older jobs rarely did.

• Autonomy: how much say a person has over how they work, and how closely they are watched while doing it. A role reshaped around a monitoring dashboard, keystrokes logged, screen time tracked, active minutes scored, can hollow out a job that looks unchanged on paper.
• Belonging: whether the work connects you to other people at all. A worker who used to ask a colleague for a second opinion and now asks a chatbot instead has traded a small interaction/exchange for a faster answer and a slightly smaller world.

Neither of these shows up on an org chart. Both show up, reliably, in how a role feels six months in.

The standard describes the person. This is about the seat they sit in

Our own standard is mostly about individual competence: what someone should know, do, and hold to. It has comparatively little to say about whether the role itself was built to let a competent person actually exercise that competence, rather than tick a box next to it. That is a genuine gap, and it sits with whoever designs the role, not with the standard. (check our partner Sentient Staff) If you are redesigning a job around a new AI system, the honest version of the question is not “does a person still do this.” It is closer to: does this person still get to use a real range of their judgement, see enough of the outcome to learn from it, decide some of how they work day to day, and talk to another peer when something does not fit the pattern. Our org self-audit will not answer that for you, but it will tell you, in five minutes, whether the mandate and escalation path around a role are real or only written down.

Written by us at Ethics Directive, drawing on Hackman and Oldham’s Job Characteristics Model (1970s, still the standard reference) and Sarah Bankins and Paul Formosa’s more recent work extending it with autonomy and belonging, plus YouGov’s 2025 survey on the meaningfulness of American work. Summarised and applied in our own words, not reproduced from any single source. If anything here needs correcting, we will say so in the open, dated.