First-line manager / team supervisor
The first person above the work: deciding who does what today, noticing trouble before it reaches a report, and answering for a team's output without doing the work yourself.
This is not a probability of losing your job. It combines how much of the role's task load is exposed to automation with how far adoption has actually gone — useful for comparing occupations on one consistent basis, and for nothing else.
The first management layer that still touches the work — shift supervisors, team leads and department heads, typically running 5 to 30 people. Senior executives (strategy, capital, headcount plans) and project managers without direct reports are different jobs with different automation paths and are not assessed here. Confidence is low on purpose: this occupation was added because it appears on every org chart, not because the published evidence about it is thick.
What is actually changing#
The unit of analysis is the task, not the job title. A role is not replaced — its task mix shifts.
Deciding who does what today
Automating✓ Evidence-backedMatching people to shifts, jobs and queues — who is free, who is fast at this, who is owed a better week — and reshuffling the whole thing when someone calls in sick.
Assignment under constraints — availability, skill, fairness, cost — is an optimisation problem with a clean objective, which is the oldest thing software does well. What changed recently is not the maths but the inputs: these systems now read the messy signals a supervisor used to hold in their head, and they are being bought as staffing products rather than built in-house.
A roster the machine produced still has to be defended to the person who got the bad shift, and that conversation stays with the supervisor. What automates is the hour of arranging, not the accountability for the arrangement — so this role can lose its most visible daily task and keep the one that makes it a job.
Seeing it go wrong before the report does
Being augmented≈ Platform inferenceNoticing that a line is slowing, a customer is about to complain, or someone on the team has quietly stopped trying — usually from something that is not in any system.
Anomaly detection on throughput, queue time and quality is standard in the tools these teams already run, and it beats a person at catching slow drift. It does not beat a person at the signals that never become data — who came back from lunch different, which two people have stopped speaking.
More alerts is not more attention. Where detection got cheap, the usual result is a supervisor triaging a longer list of flagged items in the same hours — the work moved from finding to filtering, which reads as no change at all on a job description.
Writing the review and giving the feedback
Automating✓ Evidence-backedAssembling half a year of someone's work into an assessment, saying it to their face, and defending it when it decides their pay.
The assembly half — pulling activity, tickets, output and prior notes into a draft — is summarisation over records the company already keeps, and drafting tools are sold into HR suites for exactly this. Managers reach for them readily, because writing reviews was always the part they postponed.
A drafted review is not a delivered one, and what costs a manager something is saying it out loud to someone who disagrees. It also creates a new failure mode rather than removing one: an assessment assembled from whatever was easy to log quietly promotes whoever is most legible to the system.
The conversation that keeps someone
Still human-led≈ Platform inferenceFinding out why someone is about to leave while there is still time, and either fixing it or telling them honestly that you cannot.
The binding constraint is not prediction, it is standing. Retention models can flag who is at risk; what changes a decision is a commitment from someone with the authority to make it, inside a relationship that makes it credible. Neither transfers to a system, and a flag delivered to a manager with nothing to offer changes nothing.
Being the least automatable task does not protect the headcount around it. A company can halve its supervisors with this task fully intact — each remaining one simply does it with twice as many people. That is how an irreplaceable part gets thinner without ever being replaced.
Telling the floor above what happened
Automating≈ Platform inferenceTurning a week of shifts, incidents and numbers into something the layer above can act on, and deciding what is worth raising at all.
Where the underlying records are already digital, the weekly summary is a reporting query plus drafting, and both are commodity. Management layers exist partly to move information upward; that part of the layer is the cheapest thing in it to remove.
What automates is the summary, not the selection. Deciding that a near-miss is worth the boss's attention this week is a judgement with career consequences attached — and a system that reports everything has effectively reported nothing.
Defending a decision the system made
New task✓ Evidence-backedExplaining to the person in front of you why the scheduler gave them that shift, why the score says what it says, and overriding it when it is wrong — on the record.
Once allocation and assessment run through software, someone has to stand between the output and the people it lands on. This duty did not exist in the role before the tools did, and it is rarely written into anyone's job description — it falls to whoever is physically nearest the affected person, which is this layer.
New duties are not new authority. A supervisor asked to explain a system they are not allowed to change carries the blame for it without the power to fix it — a different job from the one they were promoted into, heavier, and invisible on the org chart.
Which technologies matter here#
Four separate signals. They are deliberately not added together — a job exposed to two technologies is not twice as exposed.
How it got here#
The index is not a static number. This is where it would have sat at each capability checkpoint since ChatGPT — reconstructed, and labelled as such.
● 3 verified events for this occupation, plotted at the date it happened — the parts of the line near a marker are anchored to something checkable.
Starts higher than most office roles because the biggest task in this job — building the roster — had real software long before language models: workforce-management systems have been assigning shifts against availability, skill and cost since the 2000s. The gentle slope after 2023 is a second wave arriving on the writing side: weekly reports, then first drafts of performance reviews. It is a slope and not a jump because none of this removes a supervisor, it removes hours from one — and that shows up as more people per supervisor rather than fewer supervisors, which moves at the speed of reorganisations.
A flat line is not a forecast of safety. It says which tasks automation has reached so far — the occupations that moved least here are the ones where the constraint is physical or regulatory, and both of those can change.
Recent changes#
The European Union. The clause names exactly two things this job does every day: allocating tasks based on individual behaviour, traits or characteristics, and monitoring and evaluating the performance and behaviour of people in work relationships. The application date comes from Article 113, read in the official text: the Regulation applies from 2 August 2026, with Article 6(1) and its corresponding obligations deferred to 2 August 2027 — so this is not something coming, it has been in force for forty days. What it does not establish: high-risk is not prohibition. It brings obligations — risk management, data governance, logging, transparency, human oversight — and this record cites the classification itself, not the detail of that obligation set. It also does not say any employer has yet been penalised under it, and says nothing about jurisdictions outside the EU.
Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.
Colorado only. The statutory definition of a consequential decision names employment, so an adverse decision an employer makes with an automated system falls inside this law. One thing this record does not establish needs saying plainly: the law puts the explanation and the review on the deployer, and says nothing about who inside an employer discharges it — it never mentions supervisors. Attaching it to this occupation's tasks is our inference about where the duty lands in practice: on the layer standing closest to the affected person. The law also does not require the review to change the outcome, only that it be meaningful. Duties phase in, with the developer documentation requirement starting 1 January 2027 and the attorney general's rules on post-adverse-outcome disclosure due by the same date.
Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.
California. The bill would have required three things: notice to a worker before deploying an automated decision system that makes employment decisions; a prohibition on relying solely on such a system for a disciplinary, termination or deactivation decision; and a right for the worker to request the data the system used. The Governor's veto message gives his reasons: the notification duty was unfocused and would fall on any business using even innocuous tools; the restrictions were overly broad (his example is that barring customer ratings as the primary input removes a tool for rewarding high performers); and the disciplinary and termination scenarios are, in his words, partially covered by forthcoming California Privacy Protection Agency regulations. That last point matters: this record establishes that the specific guardrail SB 7 proposed does not exist, not that California has no protection — separate privacy and anti-discrimination rules are outside its scope.
A rule that would have constrained or required automation was formally proposed and did not come into force — vetoed, voted down, struck down, or allowed to lapse. It establishes something real and checkable about the legal environment: a guardrail many people assume exists does not. It never moves a task's assessment, because it says nothing about what employers can do or are doing.
What this means for you#
The usual first rung into management was being trusted with the roster and the weekly report — the two things that automate first. Expect to be promoted straight into the hard half: the conversations. That is a harder start than your own manager had, and the practice they got from doing the easy half is not on offer to you.
What you know about this team — who covers for whom, which two people cannot work the same shift, what your best operator is actually worried about — is in no system, and it is a larger share of why you are paid than it was. The risk is not being replaced; it is being handed two more teams because the roster stopped taking time.
Your options#
Four directions, each with its real constraints and one thing you can test this week. Continuing as you are is a legitimate choice — it just has to be a chosen one.
Own the override, not the roster
Building the schedule is leaving this job; being allowed to overrule it is not. The supervisors who stay valuable are the ones whose override sticks — and who can say afterwards why it was right.
Override authority is granted by people who have been burned by a bad one, and it usually arrives only after a run of small correct calls. It also makes your mistakes attributable in a way that following the system never is.
For one week, write down every time you changed what a system proposed — the roster, a flag, a rating — and what happened afterwards. Five entries tell you whether your overrides are judgement or habit, and that page is the case for being given more of them.
Decide what the system is allowed to decide
Somebody has to set which calls the scheduler makes alone, which need a human, and what happens when it is wrong. In most companies nobody owns this, and it is being settled by default — usually by whoever configured the tool.
It is a governance job dressed as a settings screen, it competes with your operational number, and it only becomes a real role if someone above agrees it is one. The common failure is doing it for a year unpaid and then asking.
List the last ten decisions your scheduling or rating system made without a human, and mark each: fine, should have been checked, caused a problem. Bring the three counts to your manager. Counts move this conversation; opinions do not.
Move to where the work is physical and variable
Allocation automates fastest where the work is uniform and the constraints are clean. Sites where the job changes with weather, terrain, a customer's building or a safety call keep more of the decision with the person standing there.
It usually means shift work, a site rather than an office, and a first year where you are worse than the person you replaced because you do not know the site. Domain knowledge here is physical and takes time you cannot compress.
Ask one supervisor on a site like that a single question: what did you have to change this week that no schedule could have predicted? If they cannot think of one, that site is closer to yours than it looks.
Cross into workforce planning
Deciding how many people a site needs, at what skill mix, against what demand curve, is becoming a function rather than a spreadsheet someone keeps. Supervisors who have actually run the roster know the constraints that never make it into the model.
It pays in salary rather than in the authority you have now, you stop knowing anyone's name, and you will be measured on other people's numbers. It is also a small function: most companies have one of these people, not five.
Take last quarter's staffing and work out, from the raw records, how many people you actually needed each week versus how many you had. Whatever took you longest is the job — and whether that hour was interesting tells you more than any job description.
Common questions#
Not the part where someone has to be answerable to the people in front of them — a commitment has to come from someone who can be held to it. But rostering, reporting and drafting assessments are a large share of a first-line manager's week, and those are automating. The realistic reading is fewer supervisors each carrying more people, not a floor with nobody in charge.
The signal for this job is your span of control, and you can read it yourself. Write down how many people you were responsible for twelve months ago and how many today. If that number has grown while nobody was promoted into the layer beside you, the change has already happened in your building — it arrives as more reports before it ever arrives as a redundancy, and the two are usually a year or more apart.
Rostering first, then the weekly report, then the first draft of performance reviews. All three are assembly work over records the company already keeps. Spotting trouble is being assisted rather than removed. The conversation that keeps someone, and being the one who answers for a decision, are the parts nothing on the market currently does.
For the exceptions and for the answering. A scheduler optimises against what it was told; it does not know that one of your two best people is covering for the other this month, and it cannot be the one who explains the outcome to the person it disadvantaged. That is real work — but notice it is work that scales, which is why the honest concern here is span of control rather than replacement.
Use it for the draft, never for the decision, and read what it assembled before it reaches the person. The failure here is not a wrong number, it is a rating built from whatever happened to be logged — which systematically favours the people whose work leaves traces. If you cannot say, in one sentence and without opening the tool, why someone got the rating they got, it is not yours yet and you should not be delivering it.
Method and sources#
- Assessment date
- 2026-09-11
- Basis of the task judgements
- 3 evidence-backed · 3 platform inference · 0 not enough evidence
- Verified events
- 3