Public Policy & Civil Service

6 roles live in this world

A policy and programme evaluator's job is to determine whether a government programme actually worked — not whether it launched, not whether it was popular, but whether it caused the outcome it was designed for, and…

Judgement · Undifferentiated → Distinguished

Every law that governs your life, every public service you use, every regulation that keeps your food safe or your money protected, and every pound the state spends was shaped by people you will never see. Behind the politicians who take the decisions and the votes stands a permanent machinery of officials who turn political intention into workable policy and then into services that actually reach millions of people. This is the field of public policy and the civil service: the vast, mostly invisible apparatus of governing. The structural pull at the field level is Organization — the work exists to bring order to the collective problems of a whole society, to take the sprawling, contested, chaotic business of how a country runs itself and give it structure, process, and delivery. A government without a functioning civil service is intention without execution; the field is where execution lives.

Explanation runs beside Organization because so much of governing is making the complicated clear. A policy official's core craft is taking a genuinely hard problem — how to reduce reoffending, how to hit a climate target, how to design a fair benefit — understanding the evidence, and rendering it into advice a busy minister can act on and a public can accept. The civil service is, among other things, an enormous machine for turning complexity into decisions and decisions into things people understand. The official who cannot explain clearly, in writing and in a two-minute briefing, is at a structural disadvantage, because clarity is the currency of the whole enterprise.

🦊

There's a guide here if you want one

Kitsune can talk through anything on this page — whether it might suit you, what to do next, questions this page doesn't answer. Everything here is yours to read either way.

The analytical-capacity barrier for smaller public bodies and generalists

Inference

The barrier

Rigorous quantitative analysis (economic modelling, evaluation, forecasting) required specialist analysts only large central departments could staff — the Government Economic Service and Government Social Research are large CENTRAL professions. A small local authority, small agency, or policy generalist without quantitative training could not access that capability, so their decisions rested on weaker evidence.

What changed

AI analytics and language models let non-specialists run and interpret analysis that once required a trained economist or statistician, and let smaller public bodies do evidence-based policy previously the preserve of large departments.

Behaviours involved

Policy generalists bringing quantitative evidence to advice without a dedicated analyst.
AnalyzingStructuring
Smaller public bodies and local government producing rigorous evaluation in-house.
AdvisingMeasuring

What this is based on

Civil Service AI & Data Challenge (2026 winner: a benefits-caseworker decision-support tool); departmental data-and-AI adoption plans (current_fact on the programmes; inference on the democratisation effect)
named smaller-body exemplars NOT located — evidence is the central programmes, not a verified roster

How this could age

Low risk on direction, moderate on magnitude. The capability to democratise analysis is real; whether it improves decisions or just increases the volume of poorly-grounded confident advice depends on data quality and analytical literacy the tools do not supply.

What it does not cover

Bounded by the infrastructure gap (§3): analysis is only as good as the data, and much government data is siloed and poor-quality, so a generalist with an AI tool and bad data produces confident-looking bad analysis — the false-precision failure Mode 1 warns against. The accountability floor still applies: a generalist can generate analysis, but a decision resting on it needs an accountable human who can defend it.

Assessed July 2026

The case-processing-throughput barrier

Current fact

The barrier

Operational delivery ('passports and tax and benefits' processed by more than half the civil service) was throughput-limited: every routine case consumed staff time, backlogs built, and complex or vulnerable cases competed for the same finite human capacity.

What changed

Agentic AI automates the straightforward, high-volume, rule-bound cases, which — per the UK government's own framing — lets 'staff spend their time dealing with customers with more complex needs, or who cannot access digital services'.

Behaviours involved

Delivery staff shifting from processing routine transactions to helping people whose situations are genuinely complex or who cannot navigate digital services.
SupportingSystematizing

What this is based on

2026 AI & Data Challenge benefits-caseworker tool; Companies House Copilot for routine queries; DWP work-coach personalisation (current_fact)

How this could age

Low risk on direction, moderate on magnitude. Automation of routine casework is happening and will continue; whether it liberates delivery staff or reduces their number is the open political question, treated in §10 as the field's transitional core.

What it does not cover

Genuinely double-edged and the direction of benefit is contested. Optimistic reading (the government's): it upgrades the delivery job from processing to complex support. Pessimistic reading: it is the mechanism by which the 'cut jobs' agenda (§3 economic accelerant) reduces the operational-delivery workforce, the largest entry tier. Both can be true: the REMAINING delivery work moves up-value while the NUMBER of delivery roles falls. The accountability floor is the safeguard against over-automation — the routine case that becomes an accountable decision (a benefit refusal) still needs a human, and the disaster history is why.

Assessed July 2026

Jobs that did not exist five years ago

These are real jobs that exist now and did not exist before the current wave of AI.

New title

AI governance / assurance / algorithmic-transparency officer

Government's own AI use must be documented (UK Algorithmic Transparency Recording Standard), risk-assessed (EU AI Act high-risk conformity), and made defensible against judicial review — creating dedicated oversight roles. The best-evidenced new role.

Industry/OECD coverage: 'new government roles in AI governance, oversight, and cybersecurity require master's-level credentials' (survey_self_report, careers/OECD 2026). · seen in the wild

Familiar title, new shape

Government data scientist / AI-analyst (expanded)

The analytical professions are growing an AI-and-data-science layer; the analyst role is augmented and the profession hires for AI fluency.

Expansion of analytical functions with AI/data-science capability; AI-skills wage premium ~25% (survey_self_report). · early signal

Familiar title, new shape

The 'AI-supervisor' caseworker (reconfigured)

OECD: 'the role of the government worker is moving from task executor to AI supervisor' — the caseworker who validates, oversees and corrects AI output rather than processing from scratch. The operational-delivery role reconfigured, and the field's transitional core.

OECD Governing with AI / Digital Government Outlook 2026 (survey_self_report on the framing). · seen in the wild

Genuinely substantial and positive relative to the recent corpus, for the same reason that protects the field: government must GOVERN AI, including its own. The opposite of the gfx/nfp finding ('a field with no regulation gets no compliance jobs') — gov has the MOST regulation of any recently authored field and gets the MOST compliance jobs. The AI-skills wage premium (~25% more for AI-skilled government workers, survey_self_report) confirms real, priced demand. CAVEAT: these roles concentrate at the specialist/master's-credentialed end and do not offset the transactional-tier compression at the entry end — the new roles are senior and technical, the disappearing roles junior and administrative.

The barriers that recently fell are the other half of this story. They are just above on this page.

What stays yours · where AI reaches

Bars above the line are the parts of this work that still need a person. Bars below it are what AI can already do. Tap any column to see the actual work behind it.

high ground · holds stronglydeep water · reaches furthest

yours, by strengthAI reach, by depth

The honest read. gov is one of the best-protected fields in the corpus, and its protection is unusually robust because it is LEGAL as well as relational and judgemental. Three dimensions are strong, and the strongest — accountability — is constitutionally grounded rather than merely professional. The transactional and analytical TASKS are genuinely exposed, but the ACCOUNTABLE CORE of governing is protected by a principle about who may exercise state power that no model release changes. Mode 1's 'judgement under political pressure, negotiation, accountability to the public... remains firmly human' is not aspiration; it is administrative law.

A meta-field: government's subject matter is every other sector, but the matrix scores the WORK OF GOVERNING (submissions, delivery, analysis, negotiation, enforcement), not the AI relevant to the sectors governed. Two cores plus an agentic layer: language_reasoning (the submission/briefing/casework-letter craft; Copilot trial saved ~26 min/day), predictive_ml (government is a huge employer of analysts — 3,500+ economists, 2,500+ social researchers), and agentic_ai significant and rising in operational delivery (case-processing automation). No dual_use: the regulator regtech arms race is real but sits in one sub-context, not the field core, and the field's distinctive dynamic is the public-law accountability brake (§4), a regulatory finding not an arms race — noted in intra_field_variation instead.

How AI is changing the way in

5 ways into this field, and AI is not doing the same thing to each of them.

Operational-delivery entry (gov_arch_002 — AO/EO case-processing)Harder to enter
The largest entry tier and the accessible, meritocratic route in (Mode 1). Routine casework is both most automatable (gov_cc_002) and the target of the fiscal cut-jobs agenda — 'data entry, benefits processing, basic citizen services face the highest automation risk'. Cushioned from severe by the accountability floor (gov_ncc_001) and service to vulnerable people, and the work shifts up-value rather than vanishing, but this is where entry compression concentrates.
Policy Fast Stream / policy-officer entry (gov_arch_001)Harder to enter
Junior drafting and research tasks exposed (the Copilot saving); the accountability-protected core is senior work, so entry exposure is real but the ladder leads toward protected work.
Analyst / economist entry (gov_arch_003)Harder to enter
Junior data-processing exposed; the profession is augmented and hiring for AI fluency, so the intake profile shifts toward data-and-AI-literate entrants rather than shrinking.
Regulator entry (gov_arch_005)Harder to enter
Compliance-monitoring tasks exposed; enforcement judgement and site work protected.
Diplomatic entry (gov_arch_004)Slightly harder to enter
Relational and representational from early on; the least exposed entry point. Mode 1: 'the whole role is built on relationships'.

That is everything we currently know about AI in Public Policy & Civil Service. It shows where things are moving so you can choose which way in suits you.