VOLOVLOAutomation risk & transition, task by task
AskOccupationsMajorsBusinessFoundersChangesMethodSearch occupations, majors…中文
VLO
VOLO

Understanding how automation changes work — task by task, with the evidence shown and the uncertainty admitted.

AskOccupationsMajorsBusinessFoundersChangesMethodAboutRole diagnosisPrivacyTerms
© 2026 VOLO
Occupations
All occupations
AI / software
Translator / InterpreterBank tellerCopywriterCustomer service representativeAdministrative assistantSoftware tester / QA engineerGraphic designerParalegalVideo editorAccountant / BookkeeperMarketing specialistFrontend developerData analystInsurance claims handlerJunior software developerHR / recruiterFinancial analystProcurement / supply chain specialistJournalistSales / account managerReal estate agentAuditorBackend developerProduct / UX designerBusiness systems ownerRadiologistData engineerLawyerMachine learning engineerExperienced software engineerProduct managerPharmacistPartnerships / channel managerArchitectFirst-line manager / team supervisorCounsellor / therapistSchool teacherRegistered nurseAI implementation lead
RPA / self-service
Government service clerkOperations coordinator
Robotics
Retail cashier / shop assistantWarehouse workerAssembly line workerChef / cookElectrician
Autonomous driving
Ride-hail / taxi driverTruck driverDelivery rider / courier
Majors
All majorsEnglish / Foreign languagesComputer scienceAccountingPsychologyJournalism / CommunicationFinance / EconomicsLawVisual communication designMarketingNursingBusiness administrationEducation and teacher trainingArchitecturePublic administration
Guides
Ask VOLOFor businessFor foundersRecent changesRole diagnosisMethod & evidenceAboutSearch
You are reading as:I have a jobI am studyingI run a companyI am building something
On this pageTask breakdownHow it got hereRecent changesWhat it means for youMethod & sources
Occupations›Government service clerk

Government service clerk

Stands between a citizen and a rule — checking that an application is complete, and dealing with the person whose situation the form was not written for.

government-service-clerkSee your options ↓Public administrationAssessed 2026-09-13
Automation impact index
66/100
low confidence · not a job-loss probability
Tasks automating
1of 6
1 being augmented
Still human-led
3of 6
1 new task
Evidence-backed judgements
1of 6
2 verified records
66/100
Automation impact indexLow confidence

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.

Where this applies

Written for front-line government service staff — the counter, the call line, the online back office — who take and check applications against published rules. It is not about the civil service as a whole: policy work, inspection, enforcement and public safety are different jobs with very different exposure. The rules and the pace of digitisation differ enormously by country, and this page is written mainly against mainland China and Singapore, where a national platform is the stated policy.

What is actually changing#

The unit of analysis is the task, not the job title. A role is not replaced — its task mix shifts.

Automating×1Being augmented×1Still human-led×3New task×1

Is this your job? Say so and this page narrows to your share of it.

A job title is a bundle of tasks bought together, and no two people hold the same bundle. Nothing is sent anywhere — it stays in this browser.

Taking the application

Automating✓ Evidence-backed

Checking that the materials are complete, that the identity matches, and that this is the right counter.

RPA / self-service
Why

This step is a checklist against a published rule, and every item on it exists in a database somebody already runs. It is also the step policy explicitly targets: the State Council's 2022 opinion requires cross-provincial services to run through the national integrated platform, with the names, codes and legal bases of listed items standardised nationwide by mid-2023, and pushes mutual recognition of electronic certificates — which removes the document the counter used to check.

What this does NOT mean

A policy requiring adoption is not a measurement of adoption, which is why this site weights a mandate cautiously. It says nothing about how many counters actually closed, nothing about the share of applicants who complete online, and nothing about markets without a national platform. Implementation evidence would be a deployment record, not another policy document.

Checking it against the rule

Being augmented≈ Platform inference

Deciding whether what has been submitted actually meets the requirement, including the part the requirement does not spell out.

RPA / self-serviceAI / software
Why

Where the rule is explicit and the data is structured, a system applies it more consistently than a person at the end of a long queue. What does not transfer is the judgement a written rule leaves open on purpose — most administrative rules have a margin, and the margin is where citizens with unusual circumstances live.

What this does NOT mean

Says nothing about whether the margin is used well: consistent application of a bad rule produces consistent harm, and discretion at a counter is also where unequal treatment lives. Neither is recorded here.

The case the form was not written for

Still human-led≈ Platform inference

The applicant whose documents were destroyed, whose household registration does not match, whose situation has no box.

AI / software
Why

A system can only refuse what does not fit; it cannot decide that the rule should bend, or find the other route through. This is also the case where the state's legitimacy is actually tested — a citizen who cannot get a document they are entitled to does not experience it as a software limitation.

What this does NOT mean

A judgement about where the work sits, not a measurement of how many cases are exceptions. It also does not claim counters handle exceptions well — being told to come back with a document you cannot obtain is the ordinary experience of this failure.

The person who cannot use the app

Still human-led≈ Platform inference

The older applicant, the one without a smartphone, the one whose name the system cannot render.

RPA / self-service
Why

Every policy that moves services online creates this category, and the same policies say so: the 2022 opinion requires online and offline channels to be deeply integrated rather than the counter simply closing. So the counter survives specifically for the people the platform does not reach — which makes this task a consequence of automation rather than a survivor of it.

What this does NOT mean

It establishes that the category exists and that policy acknowledges it, not how many counters are actually kept open for it or how well they are staffed. Where a service closes its counter anyway, this task does not become human-led; it stops being done.

Getting departments to move together

Still human-led≈ Platform inference

One life event that touches four agencies with four rules, four systems and four definitions of the same word.

RPA / self-serviceAI / software
Why

The platform moves the data; it does not settle whose rule wins when two agencies disagree about what a resident is. That is negotiation between institutions, and it is exactly what the 'one thing, done once' policies are asking for and exactly the part a platform cannot supply.

What this does NOT mean

Nothing here says who does this coordination today, and in many places nobody is assigned to: the citizen does it, by visiting four counters. That is the failure this task describes, not evidence against it.

Answering for what the platform decided

New task≈ Platform inference

Explaining to a citizen why the system rejected them, and being the route by which that gets fixed.

RPA / self-serviceAI / software
Why

New work created by moving services onto a platform. An administrative decision carries a right to reasons and a route to challenge it in most systems, and those attach to the agency regardless of what produced the decision. Somebody local has to be able to say why this application failed — and when the reason is inside a shared national system, that is harder than when it was a form on a desk.

What this does NOT mean

New work appearing is not new headcount, and this site holds no verified record of how such challenges are handled in any market. It also does not establish that a reason is actually available — in several systems the answer a counter can give is the same code the applicant already saw.

Which technologies matter here#

Four separate signals. They are deliberately not added together — a job exposed to two technologies is not twice as exposed.

Process & self-service
Taking the applicationChecking it against the ruleThe person who cannot use the appGetting departments to move togetherAnswering for what the platform decided
Cognitive automation
Checking it against the ruleThe case the form was not written forGetting departments to move togetherAnswering for what the platform decided

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.

Reconstructed · platform inferenceEstimated today for each past checkpoint — not measured at the time. 40 → 66.
1007550250
not assessed
2022 H22024 H2Now

A high start, because this work was already being moved online before generative AI: portals, e-filing and self-service kiosks had been taking the routine counter transaction for a decade. The 2023-2025 climb is policy rather than capability — a national platform, standardised item lists with deadlines, mutual recognition of electronic certificates, each of which removes a document the counter used to check. It flattens because the same policy that moves the transaction also requires the counter to stay open for the people the platform cannot reach. Read the shape as a mandate rather than a market: what is unusual here is that the direction is written down, and what is still unknown is how completely it was implemented.

2022 H240General-purpose text generation reaches the public. Before this point, exposure came from automation that was already deployed — OCR, RPA, machine vision, self-checkout, dispatch algorithms. ChatGPT research preview (2022-11-30) ↗
2023 H143A general model that passes professional exams. First-draft quality crosses the threshold where professional work starts using it. GPT-4 (2023-03-14) ↗
2023 H248Vision input, long context and tool calling. Models can be pointed at documents and connected to systems, which is what moves process work rather than writing work. GPT-4 Turbo:128k 上下文、视觉、工具调用(DevDay) (2023-11-06) ↗
2024 H154The same capability gets much cheaper and faster. Nothing new becomes possible; a lot becomes affordable at volume, which is when deployment decisions change.
2024 H259Reasoning models that work through multi-step problems, and the first models that operate a computer by looking at the screen. The second one is what reaches software-operating jobs. OpenAI o1(推理);同期 Claude 的 computer use 进入公测 (2024-09-12) ↗
2025 H162Agents begin operating real software end to end rather than producing text for a person to paste. This is also when the first public reversals appear — organisations that automated and partly undid it. Claude 3.7 Sonnet 与 Claude Code:混合推理 + 命令行编码代理 (2025-02-24) ↗
2025 H264Long context and tool use become the default rather than a feature. Capability gains continue; the visible constraint shifts from what models can do to liability, procurement and cost. GPT-5(2025-08-07);Claude Opus 4.5(2025-11-24) (2025-08-07) ↗
2026 H166Long-horizon agents land inside specific industry workflows. Adoption becomes sector-specific rather than general. GPT-5.5:「专为实际工作打造」 (2026-04-23) ↗
Now66The current assessment — this point is the impact index published on the occupation's page, so the curve is anchored to a number the site already stands behind. Worth noting for the flat curves: in the same weeks, a research preview of a shared specification for AI agents to operate physical devices was opened to research labs and manufacturers. That is the first capability class pointed at the physical occupations whose lines here barely move. GPT-6 Astra(2026-09-03);Claude Fable 5.1 / Mythos 5.1(2026-09-01);Model Hardware Standard 研究预览(2026-08-27) (2026-09-03) ↗

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#

Policy mandate2022-09-28Verified 2026-09-12
China's State Council General Office required cross-provincial government services to run through the national integrated platform, with listed items standardised nationwide by mid-2023

Mainland China, issued by the State Council General Office. It requires adoption and sets a deadline for standardising the names, codes and legal bases of listed cross-provincial items; it also pushes mutual recognition of commonly used electronic certificates and adds 22 items in its annex. It is a requirement, not a report of what happened: it says nothing about how many counters closed, what share of applicants complete online, or staffing. The same document requires online and offline channels to be deeply integrated rather than counters closing.

Regulation, subsidy or public procurement is requiring or funding adoption — the mirror of a constraint. It shows adoption is being required, not that it has happened, so one mandate is never enough on its own; two independent ones are.

国务院办公厅 · 国办发〔2022〕34号(中国政府网) ↗Full impact card →
Deployment2021-05-11Verified 2026-09-12
Singapore's communications ministry told Parliament on 11 May 2021 that the Digital Government Blueprint target of 90–95% of government transactions completed digitally end to end had been met

Singapore only, and it is the ministry's own assertion that a target was met — the answer does not publish what counts as a transaction, which agencies are in scope, or how it was measured, so the figure is a policy statement rather than an audited statistic. Note the date above all: May 2021, before generative AI existed as a public tool. What moved the routine counter transaction online in this market was ordinary digital government — forms, national identity, pre-filled data — and it had largely happened already. That matters for how the rest of this page reads: the open question for this occupation is not whether the routine application leaves the counter, but what happens to the cases that could not.

An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.

Ministry of Digital Development and Information (then MCI), Singapore — written answer to Parliamentary Question, 11 May 2021 ↗Full impact card →

What this means for you#

If you are starting out

The visible half of this job — taking an application and checking it is complete — is the half a national policy is explicitly moving onto a platform, with deadlines. What remains at a counter is the exception and the person the platform cannot reach, and both are harder than the routine work, not easier. If you are choosing this path for stability, look at what your local service centre is being measured on this year.

If you are experienced

This occupation is unusual on this site: the automation is required rather than chosen, and the same policies that require it also require the counter to stay open for people the platform does not reach. That is a real protection and a narrow one — it protects the function, not the number of people performing it, and it protects it only where the policy is actually implemented that way.

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.

Stay and strengthen

Own the exceptions counter

Routine intake is what policy is moving onto the platform. The applicant whose case has no box, and the one who cannot use the app, are what the counter is explicitly being kept for.

Real constraints

It is the harder work with the same pay, and in many services it is not a named role — you would be absorbing it rather than being assigned it.

Test this week

Count how many applicants last week were told to come back with something. Then check how many of those came back at all.

Reshape the role

Be the one who can say why the system refused

As decisions move into a shared platform, the right to reasons does not move with them — it stays with the agency. Someone has to be able to reconstruct a refusal, and in most services nobody can.

Real constraints

It requires access to how the platform actually decides, which front-line staff usually do not have.

Test this week

Take one online rejection and try to explain it from the record alone. Note where you have to phone someone.

Adjacent move

Move to the side that designs the service

The people writing the item lists and the forms rarely watch someone fail to complete them. Having stood at the counter is the qualification that side is short of.

Real constraints

It is a different kind of institution to work inside, and the distance from the applicant is the point of the job and also its cost.

Test this week

Take the item your counter gets the most repeat visits for, and read its published requirement as if you were the applicant.

Common questions#

How long do I have?

Unusually for this site, the direction is written down in policy rather than inferred from company behaviour: a national platform, standardised item lists with deadlines, and mutual recognition of electronic certificates. What the same policy also says is that online and offline channels must be integrated rather than the counter closing. A signal you can check yourself: what share of your counter's visits last month were people who had already tried online and could not finish. That number is what the counter is being kept for.

Is a policy requiring digital services the same as jobs going?

No, and keeping them separate is the whole method of this site. A policy establishes that adoption is required; it does not establish that it happened, how completely, or what happened to staffing. This site weights that kind of evidence cautiously for exactly that reason — a mandate describes something not yet completed, while a deployment record describes something that has. If you want to know what changed where you work, the number to look for is not in the policy.

If everything can be done on a phone, why keep counters at all?

Because the policy that moves services online also says the two channels must be integrated, and because of who is left over. Every service that goes online produces a group it does not reach — no smartphone, a name the system cannot render, documents that were destroyed, a situation with no box on the form. Those cases are harder than the routine ones, not easier, and they are the reason an administrative service is judged legitimate or not.

Is this still a stable job?

This site does not rate stability and a page that did would be worth less than one that shows you where to look. What it can tell you is the shape: the intake half has a policy with deadlines pointed at it, and the exception half is being kept by the same policy but staffed by whoever is already there. Ask what your service centre is measured on this year — if the target is the share of items completed online, that is the direction, stated in the only place it counts.

Studying towards this?

These majors lead here. Their pages break down which of their competencies transfer and what graduates typically lack.

Public administration →

Method and sources#

Assessment date
2026-09-13
Basis of the task judgements
1 evidence-backed · 5 platform inference · 0 not enough evidence
Verified events
2

How we assess an occupation →