Sixteen operating principles for a government whose agents navigate the offices on the citizen’s behalf, ask once and never copy, keep a human holding the pen, and ship within today’s law—grounded in Estonia, Ukraine, Singapore, the United Kingdom, and the Czech digital stack.

For twenty years we digitized the state without changing it. We took the form, the queue, and the counter, and we put them on a screen. The portal replaced the window, the PDF replaced the paper, the login replaced the stamp—and the citizen still did all the work. The deep structure never moved: the human being remained the integrator of the state, the living API who carries a birth certificate from the registry to the health insurer to the tax office, who proves the same fact to four agencies that already know it, who assembles the government by hand because the government will not assemble itself. The agentic state is the moment that job ends. These are its sixteen principles.

The first principle is The Citizen Is No Longer the Integrator. This is the spine from which everything else hangs. In the bureaucratic state and in its digital successor, the citizen is the connective tissue between siloed offices—the one who must know which agency needs what, in what order, by when. The agentic state takes that job away from the human and gives it to an agent, and in doing so changes not the interface but the actor.

The second principle is The Life Event Is the Unit of Service. The agentic state is not organized around forms or ministries but around the moments of a human life—a child is born, a job is lost, a business is started, a flood destroys a home. The citizen states the situation in plain language—“help me after the flood”—and the agent orchestrates insurance, housing, and building permits as one act, not as four queues.

The third principle is Ask Once, Never Copy. The agentic state does not hold a thousand duplicate copies of your address. It queries the authoritative register at the moment of decision and never duplicates the data—Estonia’s once-only principle rendered as architecture, where the agent asks the source rather than maintaining yet another stale copy of the truth.

The fourth principle is Predict, Then Offer—Never Impose. The agentic state detects the need before the citizen asks and brings the offer to them: when a family’s income falls below a threshold, the state proposes help with a pre-filled application before they learn it exists. But it offers; it does not compel. Estonia already pays 99.99% of parental benefits automatically; the agentic state generalizes that to every life event, with consent at the gate.

The fifth principle is Any Surface, One Continuous Conversation. Voice, text, phone, computer—start the conversation in the car and finish it at home. The state adapts to the citizen’s channel and context rather than forcing the citizen to adapt to the state’s opening hours and office geography.

The sixth principle is The Official Holds the Pen. The agent prepares; a named human approves every decision that touches a citizen’s rights. This is not only an ethical commitment—it is the precise legal-engineering move that lets the agentic state ship inside the existing administrative code with no statutory change, because the decision-maker of record remains the official, exactly as today.

The seventh principle is Ship Within Today’s Law First. The agentic state does not wait for new legislation. Phase one runs entirely within existing law—it begins with what can be done tomorrow, and legislates afterward, from proven pilots, rather than theorizing rules for a system that does not yet exist.

The eighth principle is Found It by Resolution, Not Statute. The founding act is a government resolution that defines the goal and assigns responsibility—a political mandate and a clear signal, not a new law. The agentic state is launched by executive will and accountability, not by waiting years for a parliament.

The ninth principle is Officials Before Citizens. The first agent serves the civil servant, not the public—lowest risk, highest feedback. An internal assistant on one ministry teaches the state how agents behave before a single citizen’s case depends on one.

The tenth principle is Agents Compose Across Ministries. The architecture is not one monolithic app per office but agents that assemble themselves dynamically across departments to serve a single request. The orchestration layer—the ability to compose a flood-response or a new-business service from many agencies at once—is the moat, not any individual app.

The eleventh principle is The Register Is the Single Source of Truth. Beneath the agents lies the connected data fund: authoritative registers that the agents read, never shadow copies they maintain. The Czech propojený datový fond is the substrate; the agent’s job is to query it, not to recreate it.

The twelfth principle is Build on What We Already Have. The Czech Republic is not a greenfield—it has electronic identity, base registers, and data mailboxes that many countries are only now creating. The problem is not the absence of foundations; it is that we build on them slowly and in fragments. The agentic state is an acceleration and an integration, not a demolition.

The thirteenth principle is Sovereign-European Runtime by Construction. The agentic state runs on inspectable, EU-hostable models, the European identity wallet, and full conformance with the AI Act—so the cognition of the Czech state is auditable, revocable, and sovereign, never rented opaquely from a foreign power that decides what its institutions may say.

The fourteenth principle is Minimization Is the Privacy Firewall. Because the agent asks once and never copies, there is no central super-profile to leak or abuse. Data minimization is not a compliance afterthought bolted onto the system—it is the system’s architecture, the structural reason the agentic state can be proactive without becoming a panopticon.

The fifteenth principle is Contestability with a Named Defendant. Every decision carries an inspectable reason and an affordable appeal to an accountable human—the principle, established in European law by the SCHUFA ruling, that a computation which determines an outcome is itself the regulated decision. There is always an answer, always a defendant, never a faceless machine.

The sixteenth principle is The Manual Fallback Never Dies. A non-digital path always exists, and the state can always be run by hand. The agentic state refuses the monoculture: it preserves the human capacity and the manual procedure so that no failure, no outage, and no excluded citizen is left without recourse.

This article is a field guide to the Agentic State. It states sixteen principles, and dissects each one identically—the Principle itself, its Place in the agentic state across five aspects, the seven reasons it holds, three patterns of how it works in practice, ten building blocks, the four “from → to” shifts it makes, the concrete moves to build it with a real example, and an honest ledger of advantages and risks. It closes with a phased Action Plan anchored to the Czech context and a named deliverable: the Agentic State Operating Charter.

Summary

1) The Citizen Is No Longer the Integrator

• The commitment: the agent, not the human, connects the siloed offices of the state.

• The mechanism: the state assembles itself around a request instead of making the citizen assemble it.

• The payoff: the end of carrying documents between agencies that already hold them.

• The failure it ends: the citizen as unpaid clerk of the bureaucracy.

2) The Life Event Is the Unit of Service

• The commitment: organize service around life moments, not forms or ministries.

• The mechanism: one stated need triggers a coordinated, cross-agency response.

• The payoff: “help me after the flood” instead of four separate queues.

• The failure it ends: the citizen decomposing their own life into agency-shaped tasks.

3) Ask Once, Never Copy

• The commitment: query the authoritative register at decision time; never duplicate.

• The mechanism: the agent asks the source instead of maintaining a stale copy.

• The payoff: the once-only principle made architectural, not aspirational.

• The failure it ends: proving the same fact to agencies that already know it.

4) Predict, Then Offer—Never Impose

• The commitment: detect the need and bring the offer; let the citizen decline.

• The mechanism: threshold-triggered, pre-filled proposals with consent at the gate.

• The payoff: entitlement reaches everyone owed it, without coercion.

• The failure it ends: the take-up gap—help no one knows to claim.

5) Any Surface, One Continuous Conversation

• The commitment: voice, text, any device; resumable across context.

• The mechanism: a single conversation that follows the citizen, not the office hours.

• The payoff: start in the car, finish at home; the state adapts, not you.

• The failure it ends: the citizen bending to the channel and geography of the state.

6) The Official Holds the Pen

• The commitment: agent prepares, a named official approves every rights decision.

• The mechanism: the human decision-maker of record is preserved unchanged.

• The payoff: deployable within the existing administrative code—no statute change.

• The failure it ends: the crumple zone—an unaccountable machine behind a signature.

7) Ship Within Today’s Law First

• The commitment: phase one runs entirely under existing legislation.

• The mechanism: begin with what is legal tomorrow; legislate after pilots prove out.

• The payoff: speed—no waiting years for a new legal regime to start.

• The failure it ends: paralysis-by-regulation that ships nothing.

8) Found It by Resolution, Not Statute

• The commitment: a government resolution defines the goal and assigns responsibility.

• The mechanism: executive mandate and accountability, not a new law.

• The payoff: a clear political signal that can start immediately.

• The failure it ends: the founding hostage to a multi-year legislative cycle.

9) Officials Before Citizens

• The commitment: the first agent serves the civil servant, not the public.

• The mechanism: lowest-risk, highest-feedback internal deployment first.

• The payoff: the state learns agent behavior before citizens’ cases depend on it.

• The failure it ends: a high-stakes public rollout with no operational experience.

10) Agents Compose Across Ministries

• The commitment: agents assemble dynamically across departments for one request.

• The mechanism: an orchestration layer, not a monolithic app per office.

• The payoff: the cross-agency service is the product and the moat.

• The failure it ends: a hundred disconnected ministry apps the citizen must still wire together.

11) The Register Is the Single Source of Truth

• The commitment: agents read authoritative registers, never shadow copies.

• The mechanism: the connected data fund as the substrate of every decision.

• The payoff: one truth, queried—not a thousand drifting duplicates.

• The failure it ends: contradictory records across uncoordinated databases.

12) Build on What We Already Have

• The commitment: exploit existing eID, base registers, and data mailboxes.

• The mechanism: integrate and accelerate the foundations, don’t rebuild them.

• The payoff: a head start most countries lack.

• The failure it ends: greenfield fantasy that ignores real assets and wastes them.

13) Sovereign-European Runtime by Construction

• The commitment: inspectable, EU-hostable models; the eID wallet; AI-Act conformance.

• The mechanism: auditable, revocable cognition the state controls.

• The payoff: the state’s institutions decide what they may conclude and say.

• The failure it ends: governance rented opaquely from a foreign runtime.

14) Minimization Is the Privacy Firewall

• The commitment: ask-once-never-copy means no central super-profile.

• The mechanism: privacy as architecture, not as a compliance bolt-on.

• The payoff: proactive service without a panopticon.

• The failure it ends: the total-surveillance version of proactive government.

15) Contestability with a Named Defendant

• The commitment: every decision carries a reason and an affordable appeal to a human.

• The mechanism: the SCHUFA principle—the determining computation is the decision.

• The payoff: there is always an answer and always a defendant.

• The failure it ends: mass automated harm with no one to confront.

16) The Manual Fallback Never Dies

• The commitment: a non-digital path always exists; the state can run by hand.

• The mechanism: preserved human capacity, manual procedure, model diversity.

• The payoff: no outage, failure, or exclusion leaves a citizen without recourse.

• The failure it ends: the brittle monoculture with no way back.

The Sixteen Principles

1) The Citizen Is No Longer the Integrator

The Principle

The agentic state removes the citizen from the role of integrator—the human being who connects the siloed offices of government by hand—and assigns that role to an agent, so that the state assembles itself around a request instead of demanding that the citizen assemble it.

It functions as the master reframe of the agentic state: the change is not a better interface but a new actor doing the integration work that the human has always, invisibly, done for free.

Place in the Agentic State: 5 aspects

1. The hidden labor of citizenship

   • Every interaction with the bureaucratic state requires the citizen to know which office needs what, in what order.

   • This navigation is unpaid, expert labor the state has always offloaded onto the governed.

2. The digital state did not remove it

   • Twenty years of portals moved the counter to the screen but left the citizen as the connective tissue.

   • A better PDF is still a PDF the citizen must route.

3. The agent as the new connective tissue

   • The agentic state inserts an actor that holds the map of the bureaucracy so the citizen does not have to.

   • The agent knows the agencies, the sequence, and the deadlines.

4. The inversion of burden

   • The work of integration moves from the citizen to the state.

   • The citizen states an intent; the agent executes the orchestration.

5. The spine of every other principle

   • Life-event service, ask-once, proactive offers, cross-ministry composition—all are consequences of this one move.

   • Remove this principle and the rest collapse into “a nicer website.”

Why it holds: 7 reasons

1. The integration work is real — someone must connect the offices; today it is the citizen, at great cost.

2. Agents can hold the whole map — a model can know every agency, rule, and sequence at once.

3. The citizen cannot — no human masters the full topology of the state they must navigate.

4. The data already exists — the offices already hold what the citizen is forced to re-supply.

5. It is the true differentiator — this, not chat, is what makes the state “agentic” rather than “more digital.”

6. It compounds — once the agent integrates, every downstream principle becomes possible.

7. It is humane — it returns time, dignity, and certainty to people at their most vulnerable moments.

Three patterns of how it works

1. Intent → orchestration → result

   • The citizen states an intent in plain language

   • The agent decomposes it into the agencies and steps involved

   • The agent executes and returns the result

2. Map → sequence → execute

   • The agent holds the topology of the bureaucracy

   • It computes the correct sequence of actions

   • It carries them out across offices

3. Silo → bridge → whole

   • Offices remain internally siloed

   • The agent bridges them at the moment of need

   • The citizen experiences one coherent state

Ten building blocks

1. A citizen-facing agent (the new integrator)

2. A map of the bureaucracy (agencies, rules, sequences)

3. Cross-agency orchestration (the composition layer)

4. Authoritative registers (the data the agent queries)

5. Identity and authentication (who the citizen is)

6. Intent understanding (plain-language need → structured action)

7. Task decomposition (need → agency-shaped steps)

8. Status tracking (where each step stands)

9. The official’s approval gate (for rights decisions)

10. An audit trail (what the agent did, on whose behalf)

The shift it makes: four “from → to” moves

1. From citizen-as-clerk to citizen-as-principal

   • The human states intent rather than executing process.

2. From human integration to agent integration

   • The connective work moves from the governed to the state.

3. From interface upgrade to actor change

   • The novelty is a new worker, not a new screen.

4. From navigating the state to being served by it

   • The state assembles itself around the person.

How to build it

A. Put one agent in front of the whole state

• Give the citizen a single conversational entry point that reaches every agency, not one app per office.

• Example: Ukraine’s Diia consolidated services into one experience and helped move the country from 102nd toward the top of the UN e-government ranking; the agentic step adds an agent that orchestrates across them.

B. Hold the map centrally, keep the offices as they are

• Build the orchestration map without forcing every agency to re-platform.

• Example: the Czech propojený datový fond lets an agent query base registers without each office surrendering its system.

C. Measure success as steps removed from the citizen

• Track citizen-initiated inter-agency steps and drive them toward zero.

• Example: the UK’s Tell Us Once lets a death be reported a single time and propagates it to every relevant office—an early, narrow instance of the citizen ceasing to integrate.

Advantages and risks

Advantages

1. The end of the citizen’s hidden, unpaid navigation labor.

2. Time and dignity returned at life’s hardest moments.

3. A genuine generational change, not a cosmetic one.

4. The enabling move for every other principle.

Risks

1. Concentrates enormous orchestration power in one layer—governance is essential.

2. If the agent errs, it errs across many agencies at once.

3. Requires real cross-agency cooperation, which is politically hard.

4. Can mask, rather than fix, broken underlying processes if used as a veneer.

2) The Life Event Is the Unit of Service

The Principle

The agentic state organizes itself around the events of a human life—birth, job loss, starting a business, disaster—rather than around forms, departments, or legal procedures, so that a single stated need triggers a coordinated response across every relevant office.

It functions as the organizing grammar of the agentic state: the citizen describes a situation, not a procedure, and the state translates it into action.

Place in the Agentic State: 5 aspects

1. Life is not shaped like the org chart

   • A flood, a birth, a layoff cuts across many agencies at once.

   • The bureaucratic state forces the citizen to slice their life into agency-shaped pieces.

2. The life event as the interface

   • The citizen states the event—”I had a child,” “I lost my job”—and that is the whole request.

   • The decomposition into tasks is the state’s job, not the citizen’s.

3. Coordination as the deliverable

   • The value is the coordinated bundle: insurance plus housing plus permits, together.

   • A single life event resolves into a single coherent service.

4. The demonstrable difference

   • Life events are where the gap between the digital and the agentic state is instantly visible.

   • They are the proof, not the slogan.

5. The pilotable unit

   • A single life event (a birth) is a clean, bounded first pilot.

   • It generalizes outward to every other event once proven.

Why it holds: 7 reasons

1. It matches reality — people experience events, not procedures.

2. It bundles correctly — one event implies a known set of services that belong together.

3. It is legible — “help after the flood” is a request anyone can state.

4. It is demonstrable — the contrast with today is immediate and visceral.

5. It is bounded — a single event is a tractable pilot scope.

6. It compounds — events share components (identity, registers, approval) reusable across all.

7. It is humane — it meets people in the language of their lives, not the state’s.

Three patterns of how it works

1. Event → bundle → delivery

   • A life event is declared

   • The state maps it to its bundle of services

   • The bundle is delivered as one

2. Trigger → orchestration → completion

   • A trigger (birth registered, income dropped) fires

   • The agent orchestrates the relevant agencies

   • The service completes with minimal citizen input

3. One event → many agencies → one experience

   • A single event touches many offices

   • The agent coordinates them

   • The citizen sees one seamless response

Ten building blocks

1. A catalog of life events (the service grammar)

2. Event-to-service mappings (what each event implies)

3. Triggers (registrations and thresholds that fire events)

4. Cross-agency orchestration (the bundle delivery)

5. Pre-filled applications (from authoritative data)

6. Consent capture (the citizen’s go-ahead)

7. The official’s approval (for rights decisions in the bundle)

8. Status and notification (keeping the citizen informed)

9. Reusable components (identity, registers, approval, shared across events)

10. Outcome tracking (did the bundle actually help)

The shift it makes: four “from → to” moves

1. From form to life event

   • The unit of service becomes the human moment, not the document.

2. From citizen-side decomposition to state-side decomposition

   • The state slices the event into tasks, not the citizen.

3. From sequential queues to one coordinated bundle

   • Services arrive together, not one office at a time.

4. From procedure-first to need-first

   • The citizen states a need; procedure becomes the state’s internal concern.

How to build it

A. Start with one high-clarity event

• Pick a single, emotionally clear life event and deliver it end-to-end.

• Example: the birth of a child—the maternity ward registers the birth, and the state proactively offers everything the family needs in a few clicks—is the canonical first pilot.

B. Map the bundle before building the agent

• For each event, enumerate the agencies and services that belong together.

• Example: Singapore’s LifeSG organizes government around life moments and cross-agency data sharing, bundling services a citizen would otherwise chase separately.

C. Generalize by reusing components

• Build identity, register-query, and approval once; reuse them across every event.

• Example: once “birth” works, “job loss” and “starting a business” reuse the same orchestration spine.

Advantages and risks

Advantages

1. Service that matches how people actually live.

2. Immediate, demonstrable improvement over the digital state.

3. Clean, bounded pilots that generalize.

4. Reusable components that compound across events.

Risks

1. Event bundles can embed wrong assumptions about what people need.

2. Edge cases and unusual life situations may be poorly served.

3. Bundling can over-reach, offering more than the citizen wants.

4. Requires cross-agency agreement on what each event entails.

3) Ask Once, Never Copy

The Principle

The agentic state collects a fact from the citizen at most once and thereafter queries the authoritative register at the moment of decision—never duplicating, syncing, or hoarding a private copy of data the state already holds.

It functions as the data constitution of the agentic state: the once-only principle rendered as architecture rather than aspiration.

Place in the Agentic State: 5 aspects

1. The citizen is not a data courier

   • In the bureaucratic state, the citizen re-supplies the same facts to every office.

   • Ask-once ends the courier role: the state already holds the truth.

2. Query, do not copy

   • The agent reads the source register when it needs a fact, then forgets it.

   • No new master copy is created to drift, leak, or contradict.

3. The register as truth

   • The authoritative register is the single source; everything else queries it.

   • Copies are the enemy of consistency and of privacy alike.

4. Privacy by minimization

   • Because nothing is copied, there is no central super-profile to abuse.

   • The data constitution is also the privacy firewall.

5. Consent and purpose at the query

   • Each query is purpose-bound and, where required, consented.

   • Access is logged at the point of use, not buried in a warehouse.

Why it holds: 7 reasons

1. The state already has the data — re-collection is pure waste and indignity.

2. Copies drift — duplicated data becomes inconsistent and wrong.

3. Copies leak — every duplicate is a new attack surface.

4. Querying is now cheap — connected registers make real-time lookup feasible.

5. It minimizes by design — no hoard means less to protect and abuse.

6. It localizes truth — one authoritative source ends contradictory records.

7. It is auditable — purpose-bound queries log who saw what, when, and why.

Three patterns of how it works

1. Need → query → forget

   • A decision needs a fact

   • The agent queries the authoritative register

   • The fact is used and not retained

2. Source → authority → consistency

   • One register is authoritative for a fact

   • All consumers query it

   • The whole state stays consistent

3. Consent → purpose-bound access → log

   • The citizen consents to a use

   • Access is bound to that purpose

   • The access is logged for audit

4. (patterns held to three)

Ten building blocks

1. Authoritative base registers (the sources of truth)

2. A connected data fund (the query fabric)

3. Real-time query interfaces (lookup at decision time)

4. No-copy data policies (prohibition on duplication)

5. Purpose-binding (each access tied to a reason)

6. Consent management (where consent is required)

7. Access logging (who queried what, when)

8. Data-quality governance (the source must be correct)

9. Identity resolution (linking citizen to record)

10. Selective disclosure (reveal the minimum needed)

The shift it makes: four “from → to” moves

1. From re-supply to ask-once

   • The citizen provides a fact at most once.

2. From copy to query

   • The state reads the source instead of hoarding duplicates.

3. From data warehouse to data fund

   • Truth lives in authoritative registers, queried on demand.

4. From bolt-on privacy to architectural privacy

   • Minimization is built into how data is accessed, not added later.

How to build it

A. Forbid the copy

• Make “query the register, do not duplicate” a hard architectural rule for every agent.

• Example: Estonia’s once-only principle and X-Road data exchange let agencies ask the source rather than maintain copies—the model the Czech propojený datový fond extends.

B. Make every query purpose-bound and logged

• Bind each data access to a stated purpose and record it for the citizen to inspect.

• Example: a citizen-visible access log so anyone can see which office queried which fact, and why.

C. Invest in the authority of the source

• A query architecture is only as good as the register it queries—fund data quality.

• Example: designate and maintain base registers as the legally authoritative source for each class of fact.

Advantages and risks

Advantages

1. The end of re-supplying facts the state already holds.

2. Consistency—one truth, queried, not many copies drifting.

3. Privacy by minimization—no central hoard to abuse.

4. Auditability—purpose-bound, logged access.

Risks

1. A wrong fact in the authoritative register propagates everywhere.

2. Real-time query availability becomes mission-critical infrastructure.

3. Centralized query fabric is itself a high-value target.

4. Purpose-binding must be enforced, not merely declared.

4) Predict, Then Offer—Never Impose

The Principle

The agentic state detects a citizen’s need before they ask and brings them a ready-made offer—a pre-filled application, a calculated benefit—while leaving the decision to accept entirely with the citizen.

It functions as the proactivity clause of the agentic state: anticipation with consent, help that arrives early but never imposes itself.

Place in the Agentic State: 5 aspects

1. The take-up gap is a failure

   • Benefits that people are entitled to but never claim are rights in name only.

   • Proactivity closes the gap by bringing the offer to the person.

2. The state acts first

   • When income drops below a threshold, the state offers help before the family knows it exists.

   • The initiative shifts from citizen to state.

3. An offer, not an order

   • The state proposes; the citizen disposes. Acceptance is always the citizen’s.

   • Proactivity without consent is coercion; the line is absolute.

4. Pre-filled, not pre-decided

   • The application arrives complete, but the citizen confirms and submits.

   • The work is done; the choice remains.

5. Timed to the moment

   • The offer arrives when it is useful—at the birth, at the layoff, after the flood.

   • Relevance is a function of timing, and the agent gets the timing right.

Why it holds: 7 reasons

1. Entitlement should reach its target — unclaimed help is a policy failure, not the citizen’s fault.

2. Friction is regressive — application burdens fall hardest on those most in need.

3. The data enables it — thresholds and triggers are computable from registers the state holds.

4. Consent preserves autonomy — offering, not imposing, keeps the citizen sovereign.

5. Timing multiplies value — help at the right moment is worth far more than help eventually.

6. It builds trust — a state that anticipates and offers earns legitimacy.

7. It is already proven — Estonia pays 99.99% of parental benefits automatically.

Three patterns of how it works

1. Trigger → calculate → offer

   • A threshold or event fires

   • The agent calculates the entitlement

   • It offers a pre-filled application

2. Detect → propose → consent

   • The need is detected

   • The state proposes a remedy

   • The citizen consents or declines

3. Profile → match → time

   • The citizen’s situation is understood

   • It is matched to relevant support

   • The offer is timed to the moment of need

Ten building blocks

1. Eligibility triggers (thresholds and events)

2. Entitlement calculators (compute the benefit)

3. Pre-filled applications (from authoritative data)

4. Consent gates (acceptance is the citizen’s)

5. Timing logic (deliver at the useful moment)

6. Profile matching (situation → relevant support)

7. Notification channels (reach the citizen)

8. Decline and opt-out paths (refusal is easy)

9. The official’s approval (for the rights decision)

10. Outcome monitoring (did the offer help)

The shift it makes: four “from → to” moves

1. From apply to be offered

   • The state brings the offer; the citizen need not chase it.

2. From reactive to proactive

   • Help arrives before the request, not after.

3. From paperwork to confirmation

   • The application is pre-filled; the citizen confirms.

4. From eventual to timely

   • Support lands at the moment it matters.

How to build it

A. Compute the trigger from data you already hold

• Define thresholds and events that the connected registers can detect automatically.

• Example: when a family’s income falls below a defined line, the state offers help with a pre-filled application before they learn of it.

B. Make declining trivial

• Every proactive offer must be as easy to refuse as to accept.

• Example: a single “no thanks” that the agent respects and logs—proactivity that never becomes pressure.

C. Generalize from a proven automatic service

• Start where automation is already accepted and extend the pattern.

• Example: Estonia’s near-fully-automatic parental benefit (99.99%) is the template for proactive offers across every life event.

Advantages and risks

Advantages

1. The take-up gap closes—entitlement reaches everyone owed it.

2. Help arrives at the moment it is most useful.

3. Friction and indignity are removed for the most vulnerable.

4. Trust grows in a state that anticipates and offers.

Risks

1. Proactivity without strict consent slides into coercion.

2. Detecting need requires data use that must be tightly minimized.

3. Wrong triggers offer the wrong help to the wrong people.

4. “Helpful” anticipation can feel like surveillance if not transparent.

5) Any Surface, One Continuous Conversation

The Principle

The agentic state meets the citizen on any channel—voice, text, phone, computer—and treats every interaction as one continuous conversation that follows the person across devices and time, rather than a series of disconnected sessions on the state’s terms.

It functions as the access clause of the agentic state: the state adapts to the citizen’s surface and context, not the reverse.

Place in the Agentic State: 5 aspects

1. The channel is the citizen’s choice

   • Voice in the car, text on the phone, a screen at home—whichever suits the moment.

   • The state is present wherever the citizen is.

2. One conversation, not many sessions

   • A request begun on one device continues on another without restarting.

   • Context persists; the citizen does not repeat themselves.

3. Plain language, not forms

   • The citizen speaks or types naturally; the agent translates to action.

   • The interface is conversation, not a field layout.

4. Accessibility by default

   • Voice and natural language open the state to those excluded by complex portals.

   • The least digitally fluent are first-class users.

5. The state adapts, not the citizen

   • Office hours, geography, and channel constraints dissolve.

   • The burden of adaptation moves to the state.

Why it holds: 7 reasons

1. People live across channels — a single mandated interface fits no one perfectly.

2. Context is precious — forcing a restart wastes the citizen’s effort and patience.

3. Natural language is universal — conversation is the most accessible interface there is.

4. It includes the excluded — voice reaches those whom portals leave behind.

5. It is now feasible — agents can sustain context across channels and time.

6. It dissolves friction — no hours, no geography, no channel lock-in.

7. It dignifies — the state coming to the citizen’s surface is respect made operational.

Three patterns of how it works

1. Begin → persist → resume

   • A conversation begins on one surface

   • Context is persisted

   • It resumes seamlessly on another

2. Speak → understand → act

   • The citizen states a need in natural language

   • The agent understands intent

   • It acts across the state

3. Any channel → one identity → one thread

   • The citizen arrives on any channel

   • Authenticated to one identity

   • Continuing one coherent thread

Ten building blocks

1. Multi-channel front ends (voice, text, web, mobile)

2. Persistent conversation state (context that follows)

3. Natural-language understanding (intent from speech/text)

4. Cross-device identity (one authenticated thread)

5. Accessibility features (voice, plain language, assistance)

6. Session continuity (resume, never restart)

7. Channel-appropriate rendering (fit the surface)

8. Offline and low-bandwidth fallback

9. Privacy across channels (consistent protection)

10. Human handoff (to an official when needed)

The shift it makes: four “from → to” moves

1. From the state’s channel to the citizen’s

   • Access happens on the citizen’s chosen surface.

2. From sessions to one conversation

   • Context follows the person across devices and time.

3. From forms to natural language

   • The interface is speech and text, not field layouts.

4. From office hours to always-available

   • Geography and opening times dissolve.

How to build it

A. Persist context across every surface

• Treat identity and conversation state as continuous, never per-device.

• Example: begin a request by voice in the car and complete it on a laptop at home without re-explaining anything.

B. Lead with voice and plain language

• Make natural conversation the primary interface, not an add-on to forms.

• Example: a citizen says “I lost my job” and the agent proceeds—no menu tree, no form codes.

C. Guarantee a human handoff

• Every conversation can escalate to an official when the citizen wants one.

• Example: “I’d rather speak to a person” routes to a named official with the full context attached.

Advantages and risks

Advantages

1. Access on the citizen’s terms, channel, and time.

2. Inclusion of those excluded by complex portals.

3. No repetition—context follows the person.

4. The dignity of a state that comes to you.

Risks

1. Cross-channel context is a privacy and security challenge.

2. Voice and natural language introduce recognition errors.

3. Continuity across devices widens the authentication attack surface.

4. Conversational interfaces can obscure what the state is actually doing.

6) The Official Holds the Pen

The Principle

For every decision that touches a citizen’s rights, the agent prepares the case and a named human official approves it—so the decision-maker of record remains the official, exactly as today, and the agent never decides alone.

It functions as the legal keystone of the agentic state: the precise move that preserves due process and lets the system run inside the existing administrative code without statutory change.

Place in the Agentic State: 5 aspects

1. Preparation, not decision

   • The agent assembles the case, checks conditions, and drafts the determination.

   • The act of deciding remains the official’s.

2. The administrative code is preserved

   • Because the official decides, the existing správní řád applies unchanged.

   • No new legal regime is required to begin.

3. Accountability has a name

   • Every rights decision has a human owner who can be identified and held responsible.

   • There is no “the model decided.”

4. The crumple zone is refused

   • The official is empowered and informed, not a powerless signature on an opaque output.

   • Approval is substantive, not performative.

5. The boundary is bright

   • Routine, non-rights actions the agent may complete; rights decisions it may only prepare.

   • The line between preparation and decision is explicit.

Why it holds: 7 reasons

1. It preserves due process — a human decision-maker is what the law and legitimacy require.

2. It avoids legal change — keeping the official as decider means no statute must move first.

3. It anchors accountability — a named human owns each outcome.

4. It refuses the crumple zone — the official decides, not absorbs blame for a machine.

5. It builds trust — citizens accept being judged by a person assisted by a tool.

6. It is auditable — preparation and approval are distinct, logged steps.

7. It is pragmatic — it lets the agentic state start now, within today’s institutions.

Three patterns of how it works

1. Prepare → present → approve

   • The agent prepares the case

   • It presents the draft to the official

   • The official approves, amends, or rejects

2. Routine → autonomous; rights → human

   • Routine actions complete autonomously

   • Rights decisions route to a human

   • The boundary governs which path applies

3. Draft → review → own

   • The agent drafts the determination

   • The official substantively reviews it

   • The official owns the decision

Ten building blocks

1. A preparation engine (the agent’s casework)

2. The approval interface (where the official decides)

3. The rights/routine boundary (what needs a human)

4. Reason traces (so the official can review meaningfully)

5. Amendment capacity (the official can change the draft)

6. Named accountability (the human owner of record)

7. Audit logs (preparation and approval as distinct events)

8. Override metrics (proof the official actually decides)

9. Escalation paths (complex cases to senior officials)

10. Training (officials as supervisors of agent casework)

The shift it makes: four “from → to” moves

1. From agent-decides to agent-prepares

   • The machine readies the case; the human decides.

2. From new law to existing law

   • Preserving the official as decider keeps the administrative code intact.

3. From crumple zone to empowered approver

   • The official is informed and able to change the outcome.

4. From diffuse blame to named accountability

   • Each decision has a human owner.

How to build it

A. Draw the rights/routine boundary explicitly

• Define which actions the agent may complete and which it may only prepare.

• Example: an agent may file a notification autonomously but may only prepare a benefit determination for an official to approve.

B. Make approval substantive

• Give the official the reasons, the power to amend, and the time to use them; measure overrides.

• Example: the principle stated plainly on the Czech concept—”AI připravuje podklady, úředník schvaluje. Proto není potřeba měnit správní řád.”

C. Keep preparation and approval as separate, logged acts

• Record the agent’s preparation and the official’s decision as distinct, auditable events.

• Example: an audit trail that shows what the agent drafted and what the official decided, separately.

Advantages and risks

Advantages

1. Due process preserved in substance.

2. Deployable within existing law—no statute change needed.

3. A named, accountable human for every rights decision.

4. The crumple zone refused by design.

Risks

1. Automation bias can hollow approval into rubber-stamping—must be measured.

2. Volume can pressure officials toward perfunctory review.

3. The rights/routine boundary will be contested at the edges.

4. Without real override capacity, “approval” becomes theater.

7) Ship Within Today’s Law First

The Principle

The agentic state begins entirely within existing legislation—deploying everything that is already legal—and legislates afterward, from the evidence of working pilots, rather than waiting for a new legal regime before it starts.

It functions as the tempo clause of the agentic state: start with what is possible tomorrow, and let law follow proof.

Place in the Agentic State: 5 aspects

1. Most of it is already legal

   • Preparing cases, querying registers with consent, official approval—all permissible today.

   • The agentic state’s first phase needs no new statute.

2. Law follows evidence

   • New legislation, when needed, is written from working pilots, not from speculation.

   • Proof precedes regulation.

3. Speed as a strategy

   • Beginning now, within the law, captures years that legislating-first would waste.

   • Tempo is itself an advantage.

4. Reform from a position of knowledge

   • Once pilots run, the state knows exactly which laws to change and why.

   • Legal reform becomes targeted, not theoretical.

5. Risk contained by scope

   • Operating within existing law keeps early deployments bounded and reversible.

   • The legal envelope is a safety rail.

Why it holds: 7 reasons

1. The legal room exists — preparation-plus-approval fits the current administrative code.

2. Legislating-first is slow — waiting for new law forfeits years.

3. Pilots teach — running systems reveal which laws actually need changing.

4. Evidence beats theory — laws written from proof are better laws.

5. Speed compounds — early starts accumulate learning and legitimacy.

6. Scope limits risk — the legal envelope bounds early deployments.

7. It is politically achievable — starting needs no parliamentary majority, only executive will.

Three patterns of how it works

1. Within-law pilot → evidence → targeted reform

   • A pilot runs within existing law

   • It produces evidence

   • Targeted legal reform follows

2. Possible-now → deploy → learn

   • Identify what is already legal

   • Deploy it

   • Learn what to change next

3. Bounded scope → expand → legislate

   • Start within a bounded legal envelope

   • Expand as proof accumulates

   • Legislate to enable the next stage

4. (patterns held to three)

Ten building blocks

1. A legal-feasibility map (what is already permissible)

2. Within-law pilot designs (no statute required)

3. The preparation-plus-approval pattern (the legal anchor)

4. Consent and purpose-binding (lawful data use today)

5. Evidence collection (to justify later reform)

6. A reform backlog (laws to change, prioritized by proof)

7. Scope boundaries (the legal envelope)

8. Reversibility (pilots that can be undone)

9. Regulatory liaison (to prepare targeted change)

10. Public transparency (what is being piloted, and why)

The shift it makes: four “from → to” moves

1. From legislate-first to ship-first

   • Deployment precedes new law.

2. From speculation to evidence

   • Laws are written from working pilots.

3. From waiting to starting

   • The state begins with what is legal tomorrow.

4. From broad theory to targeted reform

   • Legal change becomes precise and justified.

How to build it

A. Map the legal envelope first

• Identify everything the agentic state can do under current law before proposing any change.

• Example: the Czech concept’s explicit stance—”Fáze 1 funguje výhradně v rámci stávající legislativy. Nečekáme na nové zákony.”

B. Anchor on preparation-plus-approval

• Use the official-holds-the-pen pattern to stay lawful without new statute.

• Example: deploy citizen-facing agents that prepare and officials who approve, entirely within the existing administrative code.

C. Build the reform case from pilots

• Collect the evidence that will justify the few legal changes the next phase needs.

• Example: use pilot data to write a targeted implementing law—mirroring the Czech draft AI implementation law that follows, rather than precedes, the EU AI Act.

Advantages and risks

Advantages

1. The state can start immediately—no legislative wait.

2. Laws, when changed, are evidence-based and targeted.

3. Early scope is bounded and reversible.

4. Tempo becomes a strategic advantage.

Risks

1. Operating at the edge of existing law invites legal challenge.

2. “Within today’s law” can be stretched too far without scrutiny.

3. Necessary reforms may stall once pilots appear to work.

4. Bounded pilots can entrench patterns that later law must awkwardly accommodate.

8) Found It by Resolution, Not Statute

The Principle

The agentic state is founded by a government resolution that defines the goal and assigns responsibility—a political mandate and a clear signal—rather than by a new law, so the work can begin on executive will and accountability instead of a multi-year legislative cycle.

It functions as the founding act of the agentic state: a decision to proceed, owned by the executive, not a statute awaited from the legislature.

Place in the Agentic State: 5 aspects

1. A mandate, not a law

   • A resolution sets direction and ownership without the machinery of legislation.

   • It is a signal the whole administration can act on.

2. Responsibility assigned

   • The resolution names who owns the transformation—an agency, a center, a person.

   • Accountability exists from day one.

3. A clear political signal

   • The resolution tells every ministry that this is real and prioritized.

   • It converts ambition into mandate.

4. Compatible with ship-within-law

   • Because phase one needs no new law, a resolution is sufficient to start.

   • The founding act matches the legal reality.

5. Reversible and adjustable

   • A resolution can be revised as evidence accumulates.

   • The founding is a living instrument, not a frozen statute.

Why it holds: 7 reasons

1. Speed — a resolution can issue in weeks, not years.

2. Sufficiency — phase one needs only executive mandate, not new law.

3. Clarity — it assigns ownership and direction unambiguously.

4. Signal — it tells the administration this is prioritized and real.

5. Flexibility — it can be adjusted as pilots teach.

6. Accountability — it names who is responsible.

7. Achievability — it needs executive will, not a parliamentary majority.

Three patterns of how it works

1. Resolution → mandate → mobilization

   • A resolution is issued

   • It mandates the goal and owner

   • The administration mobilizes

2. Goal → responsibility → milestones

   • The goal is defined

   • Responsibility is assigned

   • Milestones make it accountable

3. Signal → priority → resourcing

   • The resolution signals priority

   • Ministries treat it as real

   • Resources follow

4. (patterns held to three)

Ten building blocks

1. A government resolution (the founding instrument)

2. A defined goal (where the state is going)

3. An assigned owner (who is responsible)

4. A delivery body (the team that builds)

5. Milestones (accountable checkpoints)

6. A budget line (resourcing the mandate)

7. Cross-ministry authority (to orchestrate)

8. Reporting cadence (progress made visible)

9. An expert center (capability to execute)

10. A revision mechanism (to adjust as evidence arrives)

The shift it makes: four “from → to” moves

1. From statute to resolution

   • The founding is an executive act, not a legislative one.

2. From years to weeks

   • The mandate can issue immediately.

3. From diffuse ambition to assigned ownership

   • A named body owns the transformation.

4. From frozen law to adjustable mandate

   • The founding evolves with the evidence.

How to build it

A. Issue a resolution that names goal and owner

• Define the destination and the responsible body in one executive instrument.

• Example: the Czech concept’s stance—”Politický mandát formou usnesení — ne zákon, ale jasný signál. Definuje cíl a zodpovědnosti.”

B. Stand up an expert center to execute

• Give the mandate a home with the capability to build.

• Example: an “Expert center for AI at the Digital Agency”—the DIA as the delivery owner.

C. Make milestones public

• Attach visible milestones so the mandate is accountable, not aspirational.

• Example: a one-page phase-one plan with architecture, timeline, team size, and milestones, published openly.

Advantages and risks

Advantages

1. The state can begin in weeks, not years.

2. Clear ownership and accountability from day one.

3. A strong political signal that mobilizes the administration.

4. A flexible founding that adjusts to evidence.

Risks

1. A resolution lacks the durability and force of law.

2. It can be reversed by a change of government.

3. Without legislative backing, resourcing may be fragile.

4. Mandate without capability is empty—the delivery body must be real.

9) Officials Before Citizens

The Principle

The agentic state deploys its first agents to civil servants, not to the public—the lowest-risk, highest-feedback setting—so the state learns how agents behave on internal work before any citizen’s case depends on one.

It functions as the sequencing rule of the agentic state: prove the technology where the stakes are contained and the feedback is richest, then turn it outward.

Place in the Agentic State: 5 aspects

1. Internal first, public second

   • The first agent assists an official, not a citizen.

   • The public-facing agent comes after internal proof.

2. Lowest risk

   • An internal assistant’s errors are caught by the official, not visited on a citizen.

   • The blast radius is contained.

3. Highest feedback

   • Officials use the agent intensively and report what works and fails.

   • The learning rate is maximized.

4. Capability before exposure

   • The state builds operational competence before high-stakes public deployment.

   • Experience precedes exposure.

5. Trust built from inside

   • Officials who trust the agent become its advocates to the public.

   • Internal success seeds external legitimacy.

Why it holds: 7 reasons

1. Contained risk — internal errors are caught before reaching citizens.

2. Rich feedback — officials are intensive, expert users.

3. Operational learning — the state learns to run agents before betting citizens’ cases on them.

4. Trust transfer — officials’ confidence becomes public confidence.

5. Clear value — internal assistance has immediate, measurable payoff.

6. Lower stakes, faster iteration — internal tools can be refined quickly.

7. It de-risks the rollout — public deployment inherits proven capability.

Three patterns of how it works

1. Internal pilot → learn → externalize

   • Deploy to officials

   • Learn from intensive use

   • Then turn the capability outward

2. Assist → trust → advocate

   • The agent assists the official

   • The official comes to trust it

   • The official advocates for it

3. Low stakes → iterate → high stakes

   • Begin where errors are contained

   • Iterate rapidly

   • Graduate to citizen-facing work

4. (patterns held to three)

Ten building blocks

1. An internal official-facing agent (the first deployment)

2. A single-ministry pilot (bounded scope)

3. Feedback capture (what officials report)

4. Error containment (officials catch mistakes)

5. Iteration cadence (rapid refinement)

6. Capability metrics (is it good enough yet)

7. Trust measurement (do officials rely on it)

8. A graduation gate (criteria to go public)

9. Knowledge transfer (internal lessons to public design)

10. Change management (officials as partners, not threatened)

The shift it makes: four “from → to” moves

1. From public-first to internal-first

   • The first user is the civil servant.

2. From high-stakes launch to contained pilot

   • Risk is bounded before citizens are exposed.

3. From speculation to operational proof

   • The state earns competence before public deployment.

4. From imposed tool to trusted partner

   • Officials adopt and advocate rather than resist.

How to build it

A. Put the first agent on one ministry’s desk

• Deploy an internal assistant to officials in a single department.

• Example: the Czech concept’s first pilot—”Agent pro úředníky jako první — nízké riziko, nejvyšší hodnota zpětné vazby.”

B. Instrument the feedback loop

• Capture what officials report and feed it into rapid iteration.

• Example: a structured channel where officials flag errors and gaps, driving weekly improvements.

C. Set an explicit graduation gate

• Define the criteria the internal agent must meet before any citizen-facing rollout.

• Example: accuracy, trust, and error-containment thresholds that must be met before going public.

Advantages and risks

Advantages

1. Contained risk—errors caught before reaching citizens.

2. The richest possible feedback from expert users.

3. Operational competence before public exposure.

4. Officials transformed from resisters into advocates.

Risks

1. Internal success may not fully predict public-facing performance.

2. Officials’ workflows differ from citizens’—lessons partially transfer.

3. An internal-only focus can delay public value.

4. Poorly managed, it can still threaten staff and breed resistance.

10) Agents Compose Across Ministries

The Principle

The agentic state is built as agents that assemble themselves dynamically across departments to serve a single request—an orchestration layer, not a monolithic application per office—so that the cross-agency service is the product and the moat.

It functions as the architecture clause of the agentic state: composition across silos is the thing being built, and the thing competitors cannot copy.

Place in the Agentic State: 5 aspects

1. Composition over monolith

   • Services are assembled from many agencies’ agents at the moment of need.

   • No single mega-app tries to contain the whole state.

2. The orchestration layer is the product

   • The value is the ability to compose a flood-response from insurance, housing, and permits.

   • Orchestration, not any one app, is the deliverable.

3. Silos remain; bridges are dynamic

   • Ministries keep their systems; agents bridge them per request.

   • Integration is at the orchestration layer, not by forced re-platforming.

4. Dynamic, not pre-wired

   • The composition is assembled for each request, not hard-coded once.

   • New services emerge by recombining existing agents.

5. The moat is the topology

   • The map of which agents compose with which is hard-won and hard to copy.

   • It compounds with every life event added.

Why it holds: 7 reasons

1. Life is cross-agency — real needs span departments, so service must too.

2. Monoliths fail — one app for the whole state is unbuildable and unmaintainable.

3. Composition scales — new services come from recombining existing agents.

4. Silos persist — agencies will not surrender their systems; bridge them instead.

5. The orchestration layer compounds — each added agent multiplies possible services.

6. It is the differentiator — cross-agency composition is what “agentic” actually means.

7. It is resilient — modular agents can be replaced without rebuilding the whole.

Three patterns of how it works

1. Request → decompose → compose

   • A request arrives

   • It is decomposed into agency tasks

   • The relevant agents are composed to fulfill it

2. Registry → discovery → assembly

   • Agents register their capabilities

   • The orchestrator discovers the right ones

   • It assembles them dynamically

3. Module → recombine → new service

   • Agents are modular

   • They recombine for new needs

   • New services emerge without new monoliths

Ten building blocks

1. An orchestration layer (the composer)

2. Agent capability registry (what each agent can do)

3. Discovery and routing (find the right agents)

4. Inter-agent protocols (how agents hand off)

5. Standard interfaces (so agents interoperate)

6. The connected data fund (shared substrate)

7. Composition policies (which agents may compose)

8. Handoff verification (the agentic failure point)

9. Monitoring (composed services observed end-to-end)

10. Versioning (agents replaced without breaking the whole)

The shift it makes: four “from → to” moves

1. From monolith to composition

   • Services are assembled, not contained in one app.

2. From per-ministry apps to cross-ministry orchestration

   • The service spans agencies dynamically.

3. From pre-wired to dynamic assembly

   • Compositions form per request, not once at build time.

4. From app moat to orchestration moat

   • The topology of composition is the durable advantage.

How to build it

A. Build the orchestration layer, not another app

• Invest in the composer that assembles agents across ministries, not in a single super-app.

• Example: the Czech concept’s “Agenti se skládají dynamicky napříč resorty”—dynamic cross-resort composition.

B. Let agencies keep their systems

• Bridge silos at the orchestration layer rather than forcing re-platforming.

• Example: agents query the propojený datový fond so each ministry’s system stays put while services compose above them.

C. Verify every handoff

• Treat inter-agent handoffs as the critical failure point and verify them.

• Example: explicit verification at each agency boundary so a composed flood-response cannot silently drop a step.

Advantages and risks

Advantages

1. Cross-agency services that match real life.

2. New services from recombination, not rebuilds.

3. A compounding orchestration moat.

4. Modularity and resilience—agents replaceable individually.

Risks

1. Orchestration concentrates power and failure in one layer.

2. Inter-agent handoffs are where agentic systems most often fail.

3. Standard interfaces require cross-agency agreement, which is hard.

4. A composed service is only as reliable as its weakest agent.

11) The Register Is the Single Source of Truth

The Principle

Beneath the agents lies the connected data fund: authoritative base registers that the agents read at the moment of decision, never shadow copies they maintain—so the whole state operates on one consistent truth, queried rather than duplicated.

It functions as the foundation layer of the agentic state: the substrate of authoritative data on which every agent and every decision stands.

Place in the Agentic State: 5 aspects

1. One authoritative source per fact

   • Each class of fact has a single register that is legally authoritative.

   • Everything else queries it rather than competing with it.

2. The connected data fund

   • Registers are linked into a queryable fabric.

   • An agent can ask any authoritative source it is permitted to.

3. Read, do not shadow

   • Agents read from registers; they do not maintain private copies.

   • The source stays singular and current.

4. Quality is foundational

   • A query architecture is only as good as the register beneath it.

   • Data quality is a first-order investment.

5. Governance and access control

   • Who may query what, for which purpose, is governed and logged.

   • The substrate is powerful, so its access is controlled.

Why it holds: 7 reasons

1. Consistency — one source ends contradictory records across the state.

2. Currency — querying the source returns the latest truth, not a stale copy.

3. Minimization — no shadow copies means less to secure and leak.

4. Auditability — purpose-bound queries are logged at the source.

5. Composability — a shared substrate lets agents compose reliably.

6. Authority — a legally designated source resolves disputes about fact.

7. Efficiency — maintain one register well, not a thousand copies poorly.

Three patterns of how it works

1. Designate → maintain → query

   • One register is designated authoritative

   • It is maintained at high quality

   • All consumers query it

2. Link → discover → read

   • Registers are linked into the fund

   • Agents discover the right source

   • They read at decision time

3. Govern → bind → log

   • Access is governed by rules

   • Each query is purpose-bound

   • It is logged at the source

4. (patterns held to three)

Ten building blocks

1. Authoritative base registers (people, addresses, vehicles, businesses)

2. The connected data fund (the linking fabric)

3. Designation of authority (legal source of each fact)

4. Query interfaces (real-time read)

5. Access governance (who may query what)

6. Purpose-binding and logging (auditable access)

7. Data-quality processes (the source must be right)

8. Identity resolution (link citizen to record)

9. Selective disclosure (reveal the minimum)

10. Resilience (the fund as critical infrastructure)

The shift it makes: four “from → to” moves

1. From many copies to one source

   • Truth is singular and queried, not duplicated.

2. From stale to current

   • The source returns the latest fact at decision time.

3. From data warehouse to data fund

   • Registers are linked and queried, not pooled into a hoard.

4. From ungoverned access to purpose-bound, logged access

   • Every query is controlled and recorded.

How to build it

A. Designate and maintain authoritative registers

• Make each class of fact the legal responsibility of a single, well-maintained register.

• Example: the Czech base registers and the propojený datový fond as the designated, linked sources of truth.

B. Build the fund as queryable, not as a pool

• Link registers for real-time query rather than copying them into a central warehouse.

• Example: an X-Road-style exchange where agencies query each other’s authoritative sources on demand.

C. Govern and log every access

• Bind each query to a purpose and log it at the source for audit and citizen inspection.

• Example: a citizen-visible record of which agent queried which register, for what reason.

Advantages and risks

Advantages

1. One consistent truth across the entire state.

2. Current data at the moment of decision.

3. Minimization—no shadow copies to leak.

4. Auditable, purpose-bound access.

Risks

1. A wrong fact in the source propagates everywhere instantly.

2. The fund is critical infrastructure—its availability is mission-critical.

3. A central query fabric is a high-value attack target.

4. Designating authority across agencies is politically contentious.

12) Build on What We Already Have

The Principle

The agentic state is an acceleration and an integration of existing digital foundations—electronic identity, base registers, data mailboxes—not a demolition and rebuild, because the real problem is not the absence of foundations but the slowness and fragmentation with which we build on them.

It functions as the realism clause of the agentic state: it starts from the assets a country already has, and attacks fragmentation rather than chasing a greenfield.

Place in the Agentic State: 5 aspects

1. Not a greenfield

   • The Czech Republic already has eID, base registers, and data mailboxes.

   • Many countries are only now building what it already possesses.

2. The problem is speed and fragmentation

   • The foundations exist but are used slowly and in disconnected pieces.

   • The enemy is not absence; it is incoherence.

3. Integration over invention

   • The agentic state connects and accelerates what exists.

   • It invents the orchestration layer, not the foundations.

4. Leverage as strategy

   • Existing assets are a head start most states lack.

   • Squandering them by rebuilding is the real risk.

5. Incremental, not big-bang

   • The state advances by integrating existing systems step by step.

   • No demolition, no decade-long rebuild.

Why it holds: 7 reasons

1. The assets are real — eID, registers, and mailboxes already work.

2. Rebuilding wastes them — greenfield throws away a head start.

3. Fragmentation is the true bottleneck — disconnected systems, not missing ones.

4. Integration is faster — connecting beats reconstructing.

5. Risk is lower — building on proven foundations de-risks delivery.

6. It is cheaper — leverage costs less than rebuild.

7. It is honest — it diagnoses the real problem rather than a fashionable one.

Three patterns of how it works

1. Inventory → connect → accelerate

   • Inventory existing assets

   • Connect the fragmented pieces

   • Accelerate their use

2. Asset → orchestration → service

   • Existing assets remain

   • An orchestration layer binds them

   • New services result

3. Fragment → integrate → coherence

   • Disconnected systems are identified

   • They are integrated

   • The state becomes coherent

4. (patterns held to three)

Ten building blocks

1. Electronic identity (already in place)

2. Base registers (already authoritative)

3. Data mailboxes (already used for official communication)

4. The connected data fund (the integration fabric)

5. An orchestration layer (the new piece)

6. Integration standards (to connect existing systems)

7. An asset inventory (what we already have)

8. A fragmentation map (where the gaps are)

9. Incremental delivery (step-by-step integration)

10. Acceleration metrics (speed of integration)

The shift it makes: four “from → to” moves

1. From greenfield to brownfield leverage

   • Build on existing assets, not from scratch.

2. From absence to fragmentation as the problem

   • The diagnosis shifts to coherence and speed.

3. From rebuild to integrate

   • Connect what exists rather than replacing it.

4. From big-bang to incremental

   • Advance step by step, not by demolition.

How to build it

A. Inventory and leverage existing assets

• Catalog eID, registers, and mailboxes and build on them directly.

• Example: the Czech concept’s stance—”Nestavíme na zelené louce. Máme digitální základy... problém je, že stavíme pomalu a roztříštěně.”

B. Attack fragmentation with an orchestration layer

• Add the one missing piece—composition—rather than rebuilding the foundations.

• Example: connect existing base registers via the data fund instead of creating new databases.

C. Deliver incrementally

• Integrate one system at a time, accelerating as you go.

• Example: connect a single life-event service across existing assets before scaling to the next.

Advantages and risks

Advantages

1. A real head start, leveraged rather than wasted.

2. Faster, cheaper delivery on proven foundations.

3. The correct diagnosis—fragmentation, not absence.

4. Lower risk through incremental integration.

Risks

1. Legacy assets carry legacy constraints and technical debt.

2. Integration of old systems can be its own deep difficulty.

3. “Leverage what exists” can become an excuse to avoid needed modernization.

4. Fragmented governance, not just fragmented systems, must also be fixed.

13) Sovereign-European Runtime by Construction

The Principle

The agentic state runs on inspectable, EU-hostable models, the European identity wallet, and full conformance with the AI Act—so the cognition of the state is auditable, revocable, and sovereign, never rented opaquely from a foreign power that decides what its institutions may say.

It functions as the sovereignty clause of the agentic state: the runtime that answers for the state is one the state owns, audits, and can replace.

Place in the Agentic State: 5 aspects

1. Owned, inspectable cognition

   • The models the state runs are ones it can host, audit, and modify.

   • There are no un-inspectable refusals inside its institutions.

2. European by construction

   • The eID wallet and EU legal frameworks are the default substrate.

   • Sovereignty is designed in, not retrofitted.

3. AI-Act conformance as a feature

   • The high-risk-system regime is met by design, turning compliance into trust.

   • Conformance is an asset, not a burden.

4. Revocability

   • The state can replace its models without punitive lock-in.

   • Dependence is always reversible.

5. A European model, exportable

   • Done right, the Czech approach becomes a template for other EU states.

   • Sovereignty built well is a competitive advantage.

Why it holds: 7 reasons

1. Cognition is sovereignty — whoever controls the model controls what institutions may say.

2. Foreign weights carry foreign rules — opaque models enforce another power’s politics.

3. Audit requires inspection — only inspectable models can be governed.

4. The AI Act demands it — high-risk public systems require conformance.

5. Revocability prevents capture — replaceable models cannot lock the state in.

6. European frameworks exist — the eID wallet and AI Act are ready substrates.

7. It is a competitive edge — sovereign, legitimate AI governance is exportable.

Three patterns of how it works

1. Host → audit → govern

   • The state hosts inspectable models

   • It audits their behavior

   • It governs what they may do

2. Conform → certify → trust

   • Systems are built AI-Act-conformant

   • They are certified

   • Trust follows from conformance

3. Own → replace → stay free

   • The state owns its runtime

   • It can replace models

   • It avoids lock-in

4. (patterns held to three)

Ten building blocks

1. Inspectable, EU-hostable models (the cognition)

2. Sovereign inference infrastructure (where they run)

3. The European identity wallet (eIDAS2)

4. AI-Act conformance (the high-risk regime)

5. Model audit and red-teaming (inspection)

6. Data residency and portability (sovereignty of data)

7. Procurement clauses (inspection and exit rights)

8. A national capability (skills to run and modify)

9. Trusted-partner alliances (shared EU infrastructure)

10. A revocability plan (no lock-in)

The shift it makes: four “from → to” moves

1. From rented to owned cognition

   • The state controls the model that answers for it.

2. From opaque to inspectable

   • Behavior becomes auditable and modifiable.

3. From compliance burden to trust asset

   • AI-Act conformance becomes a feature.

4. From lock-in to revocability

   • Dependence is always reversible.

How to build it

A. Default to EU-hostable, inspectable models

• Mandate models the state can host, audit, and modify for any rights-relevant function.

• Example: run the state’s agents on inspectable models under the EU AI Act (Regulation 2024/1689), not opaque foreign APIs.

B. Build on the European identity wallet

• Use eIDAS2 and the EU wallet as the identity substrate.

• Example: the Czech concept’s pairing—”Akt o AI + Digitální peněženka”—as the legal-technical base.

C. Procure with sovereignty rights

• Mandate inspection, portability, and exit in every model contract.

• Example: contractual rights to audit and to switch providers, preventing opaque lock-in.

Advantages and risks

Advantages

1. The state’s institutions decide what they may conclude and say.

2. Auditable, governable cognition.

3. AI-Act conformance turned into trust.

4. A European model exportable to other states.

Risks

1. Sovereign capacity is costlier and slower than renting frontier APIs.

2. EU-hostable models may trail the global frontier in raw capability.

3. Compute and chip access remain partly externally constrained.

4. Requires sustained investment and skills across electoral cycles.

14) Minimization Is the Privacy Firewall

The Principle

Because the agentic state asks once and queries rather than copies, there is no central super-profile to leak or abuse—so data minimization is not a compliance afterthought bolted onto the system but the architecture that lets the state be proactive without becoming a panopticon.

It functions as the trust clause of the agentic state: privacy is structural, the direct consequence of how data is accessed, not a policy layered on top.

Place in the Agentic State: 5 aspects

1. No hoard, no panopticon

   • Query-don’t-copy means there is no central profile to abuse.

   • The proactive state and the surveillance state are separated by architecture.

2. Privacy as a property of design

   • Minimization is built into the data constitution, not added afterward.

   • The system is private because of how it is built.

3. Purpose-bound, logged access

   • Each query is tied to a purpose and recorded.

   • Use is constrained at the point of access.

4. Selective disclosure

   • Only the minimum fact needed is revealed.

   • The agent learns what it must, nothing more.

5. Citizen-visible and controllable

   • The citizen can see and govern who queried what.

   • Transparency is part of the firewall.

Why it holds: 7 reasons

1. You cannot leak what you do not hold — no central copy, no central breach.

2. Minimization reduces abuse surface — less data means less to misuse.

3. It is architectural — privacy follows from query-don’t-copy, not from promises.

4. Purpose-binding constrains use — access tied to a reason limits mission creep.

5. Selective disclosure limits exposure — reveal the minimum, learn the minimum.

6. Transparency deters — visible, logged access discourages misuse.

7. It enables proactivity safely — the state can anticipate without amassing a profile.

Three patterns of how it works

1. Query → use → forget

   • Data is queried for a purpose

   • Used for that purpose

   • Not retained

2. Purpose → bind → log

   • A purpose is declared

   • Access is bound to it

   • The access is logged

3. Minimize → disclose → control

   • Only necessary data is requested

   • Minimally disclosed

   • Under citizen-visible control

Ten building blocks

1. Query-don’t-copy architecture (the firewall)

2. No central super-profile (by design)

3. Purpose-binding (access tied to reason)

4. Access logging (every query recorded)

5. Selective disclosure (reveal the minimum)

6. Consent management (where required)

7. Citizen-visible access records (transparency)

8. Data-use governance (rules on access)

9. Retention limits (forget after use)

10. Independent oversight (privacy authority)

The shift it makes: four “from → to” moves

1. From hoard to query

   • No central profile exists to abuse.

2. From bolt-on privacy to architectural privacy

   • Minimization is how the system is built.

3. From opaque access to logged, purpose-bound access

   • Every query is constrained and recorded.

4. From surveillance-by-default to proactivity-with-minimization

   • The state anticipates without amassing a profile.

How to build it

A. Make query-don’t-copy the privacy guarantee

• Treat the no-copy architecture as the primary privacy mechanism, not an add-on.

• Example: because the agent queries the register and forgets, there is no benefits super-database to breach.

B. Bind, log, and show every access

• Tie each query to a purpose, log it, and let the citizen see it.

• Example: a citizen-facing dashboard of which agent accessed which fact, when, and why.

C. Enforce minimization with oversight

• Give an independent authority the power to audit data access.

• Example: an independent privacy regulator with real access to the query logs.

Advantages and risks

Advantages

1. Proactive service without a panopticon.

2. Privacy that follows from architecture, not promises.

3. A far smaller breach and abuse surface.

4. Citizen-visible, controllable data access.

Risks

1. Real-time query availability becomes mission-critical.

2. The query fabric itself must be rigorously secured.

3. Purpose-binding must be enforced, not merely declared.

4. Even minimized, linkage of queries can re-create a profile if ungoverned.

15) Contestability with a Named Defendant

The Principle

Every decision in the agentic state carries an inspectable reason and an affordable appeal to an accountable human—holding to the principle, established in European law, that a computation which determines an outcome is itself the regulated decision—so there is always an answer and always a defendant.

It functions as the due-process clause of the agentic state: no decision is faceless, and no harm is without recourse.

Place in the Agentic State: 5 aspects

1. The right to a reason

   • Every affected citizen can demand why a decision went as it did.

   • Opacity is not permitted in governance.

2. The right to appeal

   • Appeal is affordable, accessible, and reaches an accountable human.

   • The loop is genuinely open to challenge.

3. The determining computation is the decision

   • A model output that effectively determines an outcome is regulated as the decision.

   • The law reaches past the human signature to the machine.

4. A named defendant

   • Responsibility is assigned, not diffused; there is always a party to answer.

   • Liability is clear before deployment.

5. A reproducible record

   • Decisions are logged and reconstructable for review.

   • Appeal has something concrete to examine.

Why it holds: 7 reasons

1. Power must answer — unaccountable decision-making is incompatible with a rights-bearing state.

2. Dignity requires reasons — to be told why is to be treated as a person.

3. Error requires remedy — without appeal, mistakes become permanent.

4. Trust requires recourse — citizens accept a system they can contest.

5. Liability requires an address — diffuse responsibility is none.

6. Law already requires it — the SCHUFA ruling makes the determining computation the decision.

7. Records make it real — reproducible decisions are governable ones.

Three patterns of how it works

1. Decide → explain → contest

   • A decision is made

   • Its reasons are produced

   • The citizen can contest it

2. Appeal → human review → remedy

   • An appeal is filed affordably

   • An accountable human reviews

   • A remedy issues where warranted

3. Determine → regulate → assign

   • The determining computation is identified

   • It is regulated as the decision

   • A named human is assigned responsibility

Ten building blocks

1. Mandatory reason traces (for every decision)

2. Reproducible decision logs

3. Affordable appeal channels

4. An accountable human reviewer

5. The SCHUFA principle (determining computation = decision)

6. Pre-assigned liability (a named defendant)

7. Public-option assistance (to help citizens contest)

8. Explainability standards

9. Independent adjudication (beyond the deciding agency)

10. Time-bound remedy guarantees

The shift it makes: four “from → to” moves

1. From opaque output to inspectable reason

   • Every decision can be explained.

2. From faceless machine to accountable decider

   • A real human answers for the outcome.

3. From no defendant to named liability

   • Responsibility is assigned before deployment.

4. From years-late redress to time-bound remedy

   • Appeal is fast, affordable, and effective.

How to build it

A. Require a reason and an appeal for every decision

• Mandate an inspectable reason trace and an affordable appeal as a hard requirement.

• Example: operationalize the EU’s automated-decision rights into a technical requirement on every agent.

B. Adopt the SCHUFA principle in practice

• Treat any computation that effectively determines an outcome as the regulated decision.

• Example: the CJEU’s SCHUFA ruling (C-634/21) as the governing precedent, reaching past the rubber stamp to the model.

C. Fund assistance to contest

• Give citizens help—an agent or an office—to understand and challenge decisions.

• Example: a public-option agent that explains a decision and prepares the appeal, so contestability is not a privilege of the well-resourced.

Advantages and risks

Advantages

1. Due process preserved in substance, not just form.

2. A named defendant and a real remedy for every harm.

3. Trust earned through contestability.

4. Auditable, governable decision-making.

Risks

1. Reason traces can be gamed or rendered uninformative.

2. Appeal volume can overwhelm capacity without careful design.

3. Explainability of complex models is technically hard.

4. Assistance to contest requires sustained funding to stay real.

16) The Manual Fallback Never Dies

The Principle

The agentic state always preserves a non-digital path and the human capacity to run the state by hand—refusing the monoculture—so that no outage, no model failure, and no excluded citizen is ever left without recourse.

It functions as the resilience clause of the agentic state: a guarantee that the agentic system is an addition to human capacity, never an irreversible replacement of it.

Place in the Agentic State: 5 aspects

1. A guaranteed non-digital path

   • There is always a human, offline way to reach the state.

   • No right depends on a working agent or a successful authentication.

2. The state can be run by hand

   • The knowledge and capacity to operate core functions manually are preserved.

   • Automation never deletes the manual procedure.

3. Diversity over monoculture

   • Multiple models and vendors mean no single failure is total.

   • The state has somewhere to fail into.

4. Fail-soft, not fail-catastrophic

   • When an agent fails, core services degrade gracefully and revert to humans.

   • Failure is contained, not cascading.

5. Inclusion of the excluded

   • Those who cannot or will not use agents are first-class citizens.

   • The fallback is a right, not a grudging exception.

Why it holds: 7 reasons

1. Systems fail — outages, bugs, and poisoning are certainties, not risks.

2. Exclusion is real — some citizens cannot or will not use digital agents.

3. Monoculture is fragile — one stack means one shared point of failure.

4. Manual capacity is the last line — the ability to run by hand saves the state.

5. Reversibility preserves freedom — a replaceable system cannot trap the state.

6. Graceful degradation contains harm — fail-soft beats fail-catastrophic.

7. Inclusion is a duty — no citizen may be left without recourse.

Three patterns of how it works

1. Fail → degrade → revert

   • A component fails

   • Services degrade gracefully

   • Core functions revert to human procedure

2. Diversify → isolate → contain

   • Multiple models and vendors are deployed

   • Critical subsystems are isolated

   • Failures are contained locally

3. Digital path → human path → guarantee

   • The digital path is offered

   • A human path always remains

   • Rights are guaranteed on either

4. (patterns held to three)

Ten building blocks

1. A guaranteed non-digital path (always available)

2. Preserved manual procedures (run by hand)

3. Trained human capacity (the people who can)

4. Model and vendor diversity (no monoculture)

5. Fail-soft architectures (graceful degradation)

6. Provenance and red-teaming (catch poisoning)

7. Decoupled critical subsystems (no cascade)

8. Institutional memory (knowledge retention)

9. Exclusion monitoring (no one left behind)

10. Independent resilience audits

The shift it makes: four “from → to” moves

1. From replacement to addition

   • Agents augment human capacity rather than deleting it.

2. From monoculture to diversity

   • No single failure can take the whole state down.

3. From fail-catastrophic to fail-soft

   • Failure degrades gracefully and reverts to humans.

4. From digital-only to a guaranteed human path

   • No citizen is left without recourse.

How to build it

A. Guarantee the human path in law

• No right may be denied for authentication failure or digital exclusion; a human path always exists.

• Example: a statutory guarantee that every agentic service has an offline, human equivalent.

B. Keep the manual procedure alive

• Retain trained staff and documented procedures to run core functions by hand.

• Example: aviation and nuclear operations preserve manual proficiency precisely for system failure; the state does the same.

C. Refuse the monoculture

• Mandate model and vendor diversity and fail-soft design for critical functions.

• Example: when one model is quarantined for suspected poisoning, services revert to a second model or to human procedure without interruption.

Advantages and risks

Advantages

1. No outage, failure, or exclusion leaves a citizen without recourse.

2. Graceful degradation and guaranteed reversion.

3. Resilience against monoculture and poisoning.

4. Inclusion of those who cannot or will not use agents.

Risks

1. Maintaining manual capacity consumes resources that look idle.

2. Diversity raises integration cost and complexity.

3. Dual paths can fragment service quality if ungoverned.

4. Preserved fallbacks can become neglected and atrophy in practice.

Action plan: building the Agentic State in the Czech Republic

The sixteen principles are an operating charter, not a forecast. This plan sequences them into the realistic, within-the-law path the Czech concept already sketches—officials first, life events next, legislation last—phased and accountable, each step tagged to the principles it realizes. It closes with a named deliverable.

Phase 0: Mandate (Q3 2026)

1. Found the work by government resolution that defines the goal and assigns responsibility to the Digital and Information Agency, with a public one-page plan, milestones, and a team (Principle 8).

2. Stand up an expert AI center at the DIA as the delivery owner and capability home (Principles 8, 12).

3. Commit, in the mandate, to ship within existing law—no waiting for new statute to begin (Principle 7).

Phase 1: Officials first (Q3 2026 – 2027)

4. Deploy an internal agent for civil servants on one ministry—lowest risk, highest feedback—with structured feedback capture and a graduation gate (Principle 9).

5. Run it on a sovereign-European, inspectable runtime, AI-Act-conformant from day one (Principle 13).

6. Establish the official-holds-the-pen pattern: the agent prepares, a named official approves, preserving the administrative code (Principle 6).

Phase 2: The first life event (2027 – 2028)

7. Ship the birth-of-a-child service end-to-end: the maternity ward registers the birth, the state proactively offers everything the family needs in a few clicks (Principles 2, 4).

8. Make the citizen cease to integrate: one stated need, the agent orchestrates across agencies; measure citizen-initiated inter-agency steps toward zero (Principles 1, 10).

9. Enforce ask-once on the connected data fund: query authoritative registers, never copy (Principles 3, 11).

10. Make minimization the privacy guarantee and publish citizen-visible access logs (Principle 14).

Phase 3: Generalize and make it any-surface (2028 →)

11. Extend to job loss, starting a business, and disaster recovery, reusing the orchestration spine (Principles 2, 10).

12. Deliver any-surface, one-continuous-conversation access—voice, text, any device, resumable (Principle 5).

13. Build on existing assets—eID, base registers, data mailboxes—attacking fragmentation, not rebuilding (Principle 12).

Phase 4: Guarantees and law (in parallel, hardening over time)

14. Guarantee contestability—reason traces, affordable appeal, the SCHUFA principle, public-option assistance (Principle 15).

15. Guarantee the manual fallback and refuse the monoculture—a human path always, model and vendor diversity, fail-soft design (Principle 16).

16. Legislate from evidence: write the targeted implementing law from working pilots, aligned with the EU AI Act and the European identity wallet (Principles 7, 13).

Deliverable: The Agentic State Operating Charter

A single governing artifact that commits the Czech Republic to the sixteen principles and specifies, for each, how it is realized: the integrator inversion (citizen-initiated inter-agency steps driven to zero), the life-event catalog and its bundles, the ask-once data constitution on the connected data fund, the proactive-offer-with-consent rules, the any-surface access standard, the official-holds-the-pen boundary and override metrics, the within-the-law map and the evidence-based reform backlog, the resolution mandate and delivery owner, the officials-first sequencing and graduation gate, the cross-ministry orchestration architecture, the single-source-of-truth register designations, the build-on-what-we-have asset inventory, the sovereign-European runtime posture, the minimization privacy firewall and access logs, the contestability guarantees, and the manual-fallback and anti-monoculture guarantees.

We did not build the agentic state because the technology arrived. We build it because, for the first time, the citizen need no longer be the integrator of the state—and a country that already has the digital foundations, that ships within its own law, and that keeps a human holding the pen can become the first agentic state in Europe that is a service to its citizens rather than a Leviathan over them. The Charter is how a people writes that choice down—in daylight, within today’s law, starting tomorrow.

This is an analysis published by ENSI (European Nexus for Strategic Intelligence). Real-world cases (Estonia’s once-only and automatic parental benefits, the UK’s Tell Us Once, Ukraine’s Diia and Diia.AI, Singapore’s LifeSG, the Czech digital stack and the propojený datový fond, the EU AI Act 2024/1689, and the CJEU SCHUFA ruling) are referenced as factual anchors; framework names and the sixteen principles are the author’s coinage.