What if the Philippines did not digitise its paperwork, but rebuilt the work itself? This plan starts from a blank page: AI agents do the routine work of government, a named person stays accountable for every decision, and the citizen is asked for each fact only once.
Not a government page. AIPhilippines.org is a privately owned, for-profit company and is not affiliated with any government agency. This page is an independent concept proposal written by AI Philippines (Osome Apps Inc.). It is not government policy, it was not requested, reviewed or endorsed by any government body, and nothing here describes a service that exists today.
A hand-drawn explainer of this plan (about two and a half minutes, English narration; the captions are drawn into the video and also offered as a caption track). Every frame is drawn by code: no AI-generated images or video.
Buod sa Filipino
Ito ay isang malayang panukalang konsepto ng AI Philippines (Osome Apps Inc.), isang pribadong kumpanya. Hindi ito patakaran ng pamahalaan at walang kaugnayan sa alinmang ahensya ng gobyerno.
Ang ideya: muling buuin ang paraan ng paggawa ng pamahalaan. Ang mga AI agent ang gagawa ng paulit-ulit na trabaho, habang may taong laging mananagot sa bawat desisyon. Isang beses lang hihingin sa mamamayan ang bawat impormasyon, at kusang darating ang serbisyo sa mahahalagang yugto ng buhay: kapanganakan, pag-aaral, trabaho, negosyo, sakuna at pagreretiro.
Kasama sa plano ang iisang pambansang pundasyon (digital na pagkakakilanlan at pahintulot, palitan ng datos, mga rehistro, plataporma ng mga agent), mga karapatang nakapaloob sa disenyo (paliwanag, apela sa tao, privacy, at karapatang pumili ng serbisyong walang AI), at sunod-sunod na hakbang sa loob ng sampung taon para sa bawat sangay ng pamahalaan hanggang sa barangay.
The ten biggest ideas
Agents do the work; a named person is accountable
Every agent is registered to an accountable officer. The agent prepares, checks and carries out routine steps; the officer owns the outcome and signs every decision that matters.
An agent may say yes; only a person may say no
Agents can grant what the rules clearly allow. Any denial, penalty, cut in benefits or other loss of a right is decided by a person, in writing, with reasons and a way to appeal.
Ask once, then never again
A fact given to one office is never requested by another. Offices fetch it from the registry that owns it, through a logged data exchange, with the citizen's consent where the law requires it.
Services arrive as life events
Birth, school, a first job, a business, a disaster, retirement: one event triggers every service that follows from it, across agencies, without a separate application for each.
One shared foundation, separate keys
Identity, consent, data exchange, registries and the agent platform are built once for the whole state. Each constitutional body holds its own keys and data, so shared technology never erases the separation of powers.
Autonomy is earned in tiers
A five-step ladder ties what an agent may do to the safeguards around it. Agents move up only after public evaluation, and move down automatically when they fail.
Every action leaves a public trace
Each agent action is written to a tamper-evident log. Citizens see who used their data and why; auditors check spending as it happens; the public can follow each peso from law to delivery.
Built for typhoons, not for offices
Barangay nodes work offline, text messages and voice calls are first-class channels, and data lives in several regions. Forecast-based aid moves before the storm arrives.
Every language, and always a person
Services speak Filipino, English, the major regional languages and Filipino Sign Language. Anyone can choose a person instead of AI at any step, with no penalty or delay.
Reskill, do not replace
No civil servant loses their job to an agent. Routine posts become case owners, agent supervisors, field workers and auditors, with paid training and a guaranteed path.
The starting point: Singapore is rebuilding government work around AI agents
This plan was prompted by a Computer Weekly report on how Singapore's Government Technology Agency (GovTech) wants to rebuild public-sector work around AI agents. That article could not be opened by our tools (the site refused automated reading), so the summary below is drawn from GovTech's own published speech at its STACK 2026 conference and from other press reports of the same announcements. The sources are listed at the end of this page.
What is being rebuilt
Singapore's point is that the real prize is not automating today's tasks one by one. It is redesigning how public services are designed and delivered, now that software can act and not only answer. GovTech's first ten years built digital foundations such as a national login; the next ten are framed as rethinking what government can do.
How agents do the work
Agents plan a task, use approved tools and reach into government systems to finish multi-step work. Public officers get a secure personal assistant that is gaining the ability to act. One demonstration showed a voice assistant finding a sports court, filling in the resident's details with their permission, accepting standard declarations and paying. The longer view is that residents may one day have their own agents that tell government what they need.
The role of people
People stay accountable. The speech drew a line between drafting a message and sending it, and between recommending a payment and making it. Higher-risk actions such as payments, formal communications and deleting data need a person's approval. It also warned against letting automation hollow out how junior staff learn judgement; in the speech's words, "automate the task, not the learning".
The platforms
A shared platform to build, connect, run, monitor and evaluate agents across government, so agencies do not each build their own.
A registry of agents that records who owns each agent and what it does, reported for roughly 150,000 public officers.
A managed place for officers to deploy AI-assisted applications with government hosting, single sign-on, logging and security scanning built in.
Shared platforms for identity, case management, workflows and data that agencies combine instead of rebuilding.
The safeguards
A risk and capability framework that matches what an agent can do to the controls it needs, turned into technical requirements for the whole government: human oversight, permissions, tool access, testing and monitoring.
Tracing: teams must be able to see each step an agent took, what it retrieved and where it failed. A trustworthy model does not make a trustworthy system.
Least-privilege access, tested recovery, and automated red-teaming of citizen-facing assistants before launch.
A plain rule that when something goes wrong, a person or an institution answers for it.
What this plan takes from Singapore: shared platforms, a registry, risk-matched autonomy and human accountability. What it adds: a design for every branch of a much larger, multilingual, disaster-prone archipelago, with rights and anti-corruption controls built into the foundation.
What other countries teach
No country has finished this journey. Each of the efforts below solved one part of the problem, and each also shows a limit. The plan borrows the pattern, not the product.
Where
The pattern worth borrowing
The limit to learn from
Estonia
A data exchange layer (X-Road) where each agency keeps its own system and every exchange is signed, encrypted and logged; services built on asking citizens for data only once.
It took decades of steady investment and a strong digital identity first.
India
Digital public infrastructure as open building blocks: identity, payments, documents and consent-based data sharing that public and private services reuse.
Exclusion when identity checks fail for the poorest; consent and grievance paths must be real.
Singapore
Shared government platforms, a national login, life-event services for a newborn, and now an agent platform, an agent registry and a risk framework.
A compact city-state; the model must be adapted for more than 7,000 islands and many languages.
Ukraine
A national app that kept services running through war, and an assistant that completes a service inside a chat, not only answers questions.
The first agent service is narrow; scale and safety are still being proven.
United Kingdom
One design standard and one front door for government, plus a small in-house AI team building open tools such as consultation analysis.
One internal assistant was retired once commercial tools matched it: build only what is truly shared.
United Arab Emirates
A stated goal of running half of federal sectors, services and operations on agentic AI within two years, with a common platform and mass training.
Speed set from the top needs equally strong independent oversight.
European Union
A risk-based AI law: some uses banned, high-risk uses in public services, justice and policing tightly controlled, transparency for the rest.
Compliance alone does not create good services.
OECD, UN and UNESCO
Shared principles for trustworthy AI in government; evidence that most public AI is still stuck in pilots; advice to start with high-benefit, low-risk uses.
Principles are not architecture; someone must build the foundation.
The Philippines already has laws that point in this direction, including a data privacy law, a national identity law, an ease-of-doing-business law and a 2025 e-governance law. This plan learns from them but does not depend on today's systems. It is a rebuild.
Vision and principles
The vision: a government that knows what it already knows, acts on it without being asked, explains itself, and can be checked by anyone. AI is the centre of how the work is done. People remain the centre of who decides and who answers.
1. Agents do the work; humans are accountable
Every agent has a named accountable officer and a published mandate: what it may read, what it may do, and what it may never do.
The asymmetry rule: an agent may approve what the rules clearly allow. Only a person may deny, penalise, suspend, or take anything away.
No decision is ever blamed on "the system". Responsibility cannot be delegated to software.
2. Citizen-first, once-only, proactive
Once-only: government asks for each fact one time, then reuses it through the data exchange.
Proactive: when the state already knows someone qualifies, it offers the service. People confirm; they do not apply.
Life events, not agencies: birth, school, a job, a business, illness, disaster, migration, retirement and death are each one journey with one case owner.
A service clock: every request shows a deadline, and missing it escalates automatically to a person with authority.
3. Rights by design
Due process: notice, reasons and a hearing before any adverse decision.
A human right to appeal: every outcome can be reviewed by a person who has the power to change it.
Explainability: each decision comes with the rule applied, the facts used and where they came from, in the person's own language.
Privacy: purpose limits, data minimisation and a personal log of every access to one's data.
Inclusion: no service requires a smartphone, a bank account, literacy in English, or internet at home.
The right to a non-AI channel: a person can always be chosen instead, with the same deadline and no penalty.
4. Integrity and anti-corruption by design
No discretion without a log: any human override is recorded with the officer's name and reasons.
No queue without a clock: applications are handled in order, and every skip is visible.
Follow the peso: each appropriation carries an identifier from the budget law to the last receipt.
Analytics look at public money, contracts and office holders' official acts. They do not profile ordinary citizens.
5. Open, sovereign and replaceable
Open standards and open source by default for everything the state builds.
Model-agnostic: any AI model can be swapped without rebuilding services. No single vendor, model or cloud is a point of failure.
Provider-agnostic payments: the state pays and collects through any licensed bank, e-wallet or cash agent on equal terms.
The shared national AI and data foundation
Every branch builds on the same nine layers. They are built once, run as public infrastructure, and governed so that no single office controls them all. Each constitutional body is its own tenant with its own keys: the courts, Congress, the commissions and each local government keep custody of their own data and agents.
The architecture stack: nine layers, built once and shared by every branch.
Layer 1. Digital identity and consent
One identity for every person from birth, and one for every organisation, usable by card, phone, fingerprint or face at an assisted counter, or through a trusted helper for those who need one.
A consent wallet: people see each request to use their data, approve or refuse it, set standing permissions and withdraw them.
Verifiable credentials: licences, diplomas, titles and clearances are signed digital records that anyone can check in seconds, also on paper with a code.
Delegation: parents, guardians, carers and authorised representatives can act for others, with limits and expiry.
Agent identity: every agent has its own credential, so its actions are never mistaken for a person's.
Layer 2. The data exchange
Data stays where it is created. Offices ask each other through a common exchange, and each request states its legal purpose.
Every exchange is signed, encrypted, time-stamped and logged on both sides.
The catalogue of what each office holds, and who may ask for it, is public.
The same exchange serves agents and people: an agent gets no path to data that a person in its role would not have.
Layer 3. Registries as the single source of truth
One authoritative registry for each kind of fact: people and civil status, businesses, land and property, vehicles, addresses and maps, licences and professions, schools and learners, health facilities, beneficiaries, public assets, laws and rules.
Each registry has one owner that answers for its accuracy, a public data standard, and a way for anyone to correct a record about themselves.
Rules as code: laws, rates and eligibility rules are published in a form that both people and agents can read, with the legal text as the authority.
Layer 4. The government agent platform
An agent registry: every agent's owner, purpose, autonomy tier, tools, data permissions, model, evaluation results and incident history, open to the public.
Identity and permissions for agents: least privilege, short-lived credentials, separation of duties between agents.
A tool and API catalogue: the only way agents can act. Tools enforce the rules, so an agent cannot do what its tool does not allow.
Case management and workflow shared by all agencies, so a case moves between offices without being re-entered.
Model hosting that is sovereign or hybrid: sensitive work runs on models hosted in the country under public control; other work may use commercial models through a gateway that strips personal data and can switch providers.
A kill switch for every agent, held by its accountable officer and by the governance authority.
The autonomy ladder
Tier
What the agent may do
Safeguard
A0 Inform
Answer questions from published sources
Cites sources; says when it does not know
A1 Prepare
Draft, summarise, pre-fill, check completeness
A person reviews before anything leaves the office
A2 Recommend
Propose a decision with reasons
A person decides; samples are audited
A3 Act, reversible
Carry out approvals the rules clearly allow; schedule, notify, pay small fixed amounts
Limits per case and per day; instant undo; full log; random human review
A4 Act, high-stakes
Large payments, formal notices, changes to registries
Two-person approval every time; never for denials or penalties
Layer 5. Evaluation, red-teaming and monitoring
No agent goes live without passing public test suites for accuracy, fairness across regions and languages, safety and resistance to manipulation.
Independent red teams, including universities and civil society, attack agents before and after launch.
Live monitoring of errors, complaints, overrides and drift. Falling below the bar moves the agent down a tier automatically.
Incidents are published, with what went wrong and what changed.
Layer 6. Audit logs anyone can verify
Every agent action and every human override is written to an append-only log whose integrity can be checked from outside.
My Data Log: each person sees which office or agent used their data, when and for what purpose, and can challenge it.
The audit commission has standing, read-only access to financial events as they happen.
A public spending ledger follows each appropriation to contracts, deliveries and payments, with personal data removed.
Layer 7. Security
Zero trust: every request is authenticated and authorised, inside or outside the network.
Defence against agent-specific attacks: instruction injection, poisoned data and tool misuse are tested continuously.
A national security operations centre with AI-assisted detection, and public bug bounties.
Keys and secrets held in hardware, split between institutions for the most sensitive systems.
Layer 8. Resilience for an archipelago
Several data regions across Luzon, Visayas and Mindanao, each able to run alone.
Barangay nodes: small local servers that keep essential services running without internet and synchronise when the link returns.
Text, feature-phone menus and voice calls are full channels, not fallbacks. Radio and satellite links cover the last mile.
Typhoon mode: when a warning is raised, services switch to pre-agreed emergency rules: aid is released on forecast, records are cached locally, identity checks are relaxed and reconciled later.
Paper still works: every digital record can be printed with a code that proves it is genuine.
Layer 9. Languages and accessibility
Filipino and English everywhere; Cebuano, Ilocano, Hiligaynon, Waray, Bikol, Kapampangan, Pangasinan, Maguindanaon, Tausug, Maranao and other major languages by region.
Filipino Sign Language through video interpreters and signing avatars reviewed by the Deaf community.
Voice-first design for people who do not read or type comfortably; plain language at every step.
Public language data and models, built with native-speaker communities and released openly.
How one request flows through agents and people
Take one life event: a child is born. The hospital or the midwife records the birth once. That single entry starts a case with one human owner and a visible deadline.
How one request flows through agents and people.
Intake: the front-door agent confirms who is asking, in their language, on any channel.
Case agent: opens the life event, lists what the family is entitled to, and asks only for what no registry holds.
Specialist agents: the civil registry agent issues the birth record, the health agent enrols the child, the benefits agent checks eligibility. Each uses only its own tools.
Rules check: each proposed action is tested against rules as code and the agent's tier.
Clear yes: the action is carried out and logged, and the family is told what happened and why.
Anything else (missing facts, a conflict between records, a possible denial) goes to the human case owner, who decides and gives reasons.
Appeal: the family can ask another person to review, and can see the whole trail in their data log.
Every branch and body: the map
The same foundation serves every part of the state, and each part keeps its own keys. The sections that follow take each body in turn: the agents it would run, the work humans keep, a before-and-after story, its data and links, the risks and safeguards, and how success would be measured.
A map of branches and agents around the shared foundation.
The executive branch
The executive delivers most services, so it runs the most agents. The same pattern repeats in every department: agents prepare and carry out what the rules clearly allow, a named officer owns each case, and people keep every decision that takes something away.
Departments are named as they are today so readers can find their way. In a true rebuild, several would be reorganised around life events.
Office of the President and the Cabinet
Today: Office of the President, Cabinet clusters, the presidential management and legislative liaison staffs.
Before
A citizen's letter to the Palace is forwarded from office to office, and they never learn what became of it.
After the rebuild
The letter gets a tracking number, an owner and a deadline the same day, and the citizen can watch each step until a named officer answers.
Agents it would run
A national dashboard agent that tracks every programme against its targets and deadlines.
A coordination agent that spots overlapping or conflicting work between departments.
A drafting agent for orders and briefings that cites law and evidence.
A correspondence agent that routes every letter from a citizen to the right office and follows it to an answer.
Work humans keep
Setting direction, making appointments and taking political decisions.
Signing every order; nothing is issued by an agent.
Data and links
Reads programme, budget and delivery data from every department through the exchange.
Publishes commitments and progress to the public ledger.
Risks and safeguards
Concentration of power: the Office sees aggregates, never individual citizens' records.
Dashboards gamed to look good: figures come from registries, not self-reporting.
Success measures
Share of commitments delivered on time.
Days to answer citizen correspondence.
Health
Today: Department of Health, the national health insurer, the food and drug regulator, public hospitals and rural health units.
Before
A patient from an island town travels for a day, queues for a referral slip, and pays up front because the insurance papers are incomplete.
After the rebuild
The health centre's agent books the hospital bed, arranges the boat, confirms coverage before they leave, and the hospital is paid without a single form from the family.
Agents it would run
A patient navigator that books care, explains benefits and arranges transport from remote islands.
A claims agent that pays hospitals for valid claims in hours and flags odd patterns for review.
A disease surveillance agent that spots outbreaks from clinic and laboratory reports.
A medicine supply agent that forecasts stock for every health centre.
A clinical assistant that drafts notes and checks drug interactions for the clinician.
Work humans keep
Every diagnosis and treatment decision stays with licensed clinicians.
Denied claims, public health orders and ethics reviews.
Data and links
A lifelong health record that the patient controls; clinicians see it with consent or in an emergency, and every view is logged.
Links to civil registry, social welfare and local governments.
Risks and safeguards
Wrong advice: clinical agents stay at the prepare tier and never act alone.
Health data misuse: no access for employers, insurers or police without a court order.
Success measures
Out-of-pocket spending and travel time to care.
Days to pay a claim; stock-outs of essential medicines.
Education
Today: Department of Education, the higher education commission, the technical skills authority, state universities and colleges.
Before
A teacher spends evenings on forms; a child who changes schools waits weeks for a report card to arrive.
After the rebuild
The teacher's paperwork is done by an agent and reviewed in minutes; the child's record arrives before they do, and their tutor already knows where they left off.
Agents it would run
A personal tutor for every learner, in their mother tongue, aligned to the curriculum.
A teacher's assistant that prepares lessons and marks routine work, and takes over forms and reports.
An enrolment agent that moves a learner's record when the family moves.
A school resources agent that tracks classrooms, books and repairs.
A scholarship and skills agent that matches learners to grants, training and jobs.
Work humans keep
Teaching, mentoring, discipline and final grades belong to teachers.
Curriculum, child protection and school leadership.
Data and links
One learner record from kindergarten to work, owned by the learner and their parents.
Links to civil registry, health (nutrition, vaccination), social welfare and labour.
Risks and safeguards
Children's data: the strictest tier, no advertising, no profiling for anything but learning.
Screens replacing teachers: tutors support class time, and schools without devices get shared ones first.
Success measures
Reading and numeracy at grade level.
Hours of teacher time returned to teaching; dropouts recovered.
Social welfare
Today: Department of Social Welfare and Development and local social welfare offices.
Before
A family in need queues at several offices with the same documents and may still be left off a list they never saw.
After the rebuild
The family is told what they qualify for, confirms by text or with a barangay worker, and receives aid the way they choose, with the reasons in writing.
Agents it would run
A benefits agent that finds every programme a household qualifies for and enrols them with consent.
A payment agent that sends aid through any bank, e-wallet or cash agent the family chooses.
A case support agent that prepares social workers' files and reminds them of visits.
A disaster relief agent that pre-positions aid from forecasts.
Work humans keep
Social workers handle protection cases, home visits and any removal from a programme.
Appeals and exceptions.
Data and links
A beneficiary registry built from civil, income and disaster data, visible to the household itself.
Links to health, education, labour, housing and local governments.
Risks and safeguards
Wrongful exclusion: agents may add people; only a social worker may remove them, after notice.
Patronage: lists are rule-based and published in aggregate, with no manual insertion.
Success measures
Share of eligible households reached.
Days from need to payment; leakage and exclusion errors.
Finance and revenue
Today: Department of Finance, the internal revenue and customs bureaus, the national treasury.
Before
A small business owner pays a fixer to file returns they do not understand and waits months for a refund.
After the rebuild
The return arrives already filled in. They check it, confirm on their phone, and a refund that is clearly due is paid within days.
Agents it would run
A tax agent that pre-fills returns from data the state already has and answers questions in plain language.
A refund agent that pays clear refunds automatically.
A customs agent that clears low-risk cargo in minutes and sends risky cargo to inspectors.
A treasury agent that forecasts cash and manages payments.
A compliance agent that selects audits by published risk rules.
Work humans keep
Assessments, penalties, seizures and settlements are decided by officers, with reasons.
Tax policy and debt strategy.
Data and links
Links to the business registry, payroll, banks (with legal basis), customs and local treasuries.
Aggregate revenue published to the spending ledger.
Risks and safeguards
Selective enforcement: audit selection is logged and its rules are reviewed by the audit commission.
Leaks of tax data: access only by role, with every query recorded.
Success measures
Cost and hours to comply.
Refund time; share of cargo cleared the same day; revenue per peso of collection cost.
Budget and procurement
Today: Department of Budget and Management and the procurement service and policy board.
Before
Funds are released late, bids are compared by hand, and the public learns a project failed after the money is spent.
After the rebuild
Every project page shows its budget, bidders, prices, progress photos and payments, and money moves when work is verified.
Agents it would run
A budget agent that links every line to a programme, a result and a place on the map.
A release agent that frees funds when conditions are met.
A procurement agent that drafts tender documents, checks bids against the rules and compares prices nationwide.
A contract agent that tracks delivery with photos, sensors and citizen reports before payment.
Work humans keep
Budget choices and trade-offs.
Bid awards, disqualifications and contract termination, by named committees.
Data and links
The follow-the-peso identifier ties appropriation, allotment, contract, delivery and payment together.
Links to treasury, audit, public works, local governments and the business registry.
Risks and safeguards
Bid rigging: pattern analysis across bidders, prices and ownership, reviewed by people.
Agents favouring incumbents: evaluation rules are public and tested for bias.
Success measures
Days from budget law to release.
Share of tenders with real competition; price versus benchmark; on-time delivery.
Economic planning
Today: Department of Economy, Planning, and Development.
Before
Plans are written for years ahead on data that is already old.
After the rebuild
Planners test options on current, anonymous data and publish the model so anyone can challenge the assumptions.
Agents it would run
A planning agent that models the effect of policies on households, regions and the budget.
A project appraisal agent that checks costs and benefits of large investments.
A monitoring agent that compares plans with what was delivered.
Work humans keep
Choosing national priorities and approving major projects.
Consulting regions and sectors.
Data and links
Anonymised data from the statistics authority and every department.
Links to budget, finance and regional development bodies.
Risks and safeguards
False precision: every forecast shows its uncertainty.
Models hiding value choices: assumptions are published.
Success measures
Forecast accuracy.
Share of approved projects that meet their appraisal.
Trade and industry
Today: Department of Trade and Industry, the corporate registrar, investment promotion agencies.
Before
Opening a small shop means visits to national and local offices, each asking for the others' papers.
After the rebuild
The owner describes the business once, by voice if they like, and receives every registration and permit together, often the same day.
Agents it would run
A business start agent that registers a firm, its tax number, social insurance and local permit in one step.
A permits agent that renews licences before they lapse.
A consumer agent that takes complaints and mediates simple ones.
An export agent that guides small firms through rules of other markets.
Work humans keep
Enforcement, closures and consumer rulings.
Trade negotiations and industrial policy.
Data and links
One business registry shared with revenue, labour, local governments and regulators.
Links to customs and standards bodies.
Risks and safeguards
Fake firms: identity of owners verified, with beneficial ownership recorded.
Small firms left behind: assisted counters in markets and barangays.
Success measures
Hours and visits to start a business.
Complaint resolution time; survival of new firms.
Agriculture and fisheries
Today: Department of Agriculture and its bureaus, the food authority, irrigation and crop insurance bodies.
Before
A rice farmer hears about aid late, files a claim after a typhoon, and waits a season for the payout.
After the rebuild
Satellite and weather data confirm the damage; the payout reaches the farmer's chosen account in days, with advice on what to plant next.
Agents it would run
A farm adviser that gives each farmer and fisher local weather, pest and price advice by text or voice.
A support agent that delivers seeds, fertiliser aid and insurance payouts automatically after verified loss.
A market agent that links producers to buyers and shows fair prices.
A food supply agent that forecasts shortages and gluts.
Work humans keep
Extension workers in the field.
Import, price and biosecurity decisions.
Data and links
A registry of farmers, fishers, farms and boats, linked to land, weather and markets.
Links to agrarian reform, environment, trade and local governments.
Risks and safeguards
Ghost beneficiaries: registry checked against identity and land records.
Advice that ignores local knowledge: farmers rate advice, and extension workers correct it.
Success measures
Farmer income and yield.
Days to pay insurance; food price stability.
Agrarian reform
Today: Department of Agrarian Reform.
Before
Beneficiaries wait years for a title because records in different offices disagree.
After the rebuild
Conflicts between records are found by an agent and resolved by an adjudicator once, and the title is issued as a verifiable credential.
Agents it would run
A land records agent that reconciles titles, surveys and awards.
A beneficiary support agent that links reform beneficiaries to credit, training and markets.
A dispute intake agent that prepares case files.
Work humans keep
Adjudicators decide every land dispute.
Field validation with communities.
Data and links
Links to the land registry, environment, agriculture and the courts.
Risks and safeguards
Errors in old records: disputed parcels are always decided by a person.
Land grabbing through data: every change to a title is logged and notified to the holder.
Success measures
Time to issue a title.
Backlog of land cases.
Labour and employment
Today: Department of Labor and Employment, the wage boards, the labour relations commission.
Before
A worker who is underpaid must find the right office, fill in forms and take unpaid days off to follow up.
After the rebuild
The worker reports by text; the agent checks the payroll records, proposes a settlement, and an officer steps in if the employer disputes it.
Agents it would run
A job matching agent that connects workers to vacancies and training.
A labour standards agent that checks payroll data for unpaid wages and benefits.
A complaints agent that helps workers file a claim and offers mediation.
A safety agent that schedules inspections by risk.
Work humans keep
Inspectors, mediators and labour arbiters.
Any order against an employer or a worker.
Data and links
Links to social insurance, revenue, the business registry and migrant worker services.
Risks and safeguards
Surveillance of workers: analysis is of employers' compliance, not individual behaviour.
Wrong matches: workers choose; nothing is assigned.
Success measures
Time to resolve a wage claim.
Placement rates; inspection coverage of high-risk sites.
Overseas workers
Today: Department of Migrant Workers and the overseas workers welfare body.
Before
A worker abroad in trouble cannot reach an office in working hours, and the family at home does not know whom to ask.
After the rebuild
A message at any hour opens a case with a named officer at the nearest post, and the family can follow it.
Agents it would run
A deployment agent that verifies contracts and recruiters and issues exit papers digitally.
A welfare agent reachable from any country, in any time zone, in the worker's language.
A distress agent that alerts the nearest post and the family when a worker asks for help.
A reintegration agent that plans the return home: savings, training, business, jobs.
Work humans keep
Welfare officers, labour attaches and case workers abroad.
Decisions on bans, repatriation and legal aid.
Data and links
Links to foreign affairs, labour, social insurance, health and the worker's consent wallet.
Risks and safeguards
Data crossing borders: only the minimum, under agreements.
Fake recruiters posing as the state: every message is verifiable in the app and by a short code.
Success measures
Response time to distress calls.
Illegal recruitment cases; cost of deployment to the worker.
Transport
Today: Department of Transportation, land transport and franchising bodies, maritime, aviation, rail and port authorities.
Before
Renewing a licence costs a day in a queue; ferry schedules to small islands are a matter of asking around.
After the rebuild
The renewal arrives as a reminder with one tap to confirm; ferry, bus and plane times for the whole trip are in one place, also by text.
Agents it would run
A licensing agent that renews licences and registrations without a visit when nothing has changed.
A network agent that plans routes and schedules from real travel data.
A safety agent that tracks vessel, vehicle and aircraft inspections.
A traveller agent that plans door-to-door trips across land, sea and air.
Work humans keep
Franchise awards, safety rulings and accident investigations.
Licence suspensions and revocations.
Data and links
Vehicle, vessel, driver and operator registries.
Links to police, local governments, tourism and weather.
Risks and safeguards
Location tracking: travel data is anonymised and aggregated.
Automated penalties: tickets are issued by an officer, with evidence the driver can see.
Success measures
Time to renew.
Commute times; accidents; islands with a daily published service.
Public works
Today: Department of Public Works and Highways.
Before
A flood control project is reported complete on paper; residents know it is not, and have nowhere to say so.
After the rebuild
Each project has a public page with photos over time. A resident's photo that contradicts the report triggers an inspection before payment.
Agents it would run
A project agent that tracks each road, bridge, school building and flood work from design to handover.
A verification agent that compares progress reports with satellite images, drone surveys and citizen photos.
A maintenance agent that predicts failures and schedules repairs.
A design assistant that checks plans against codes and hazard maps.
Work humans keep
Engineers sign designs and accept works.
Decisions on contractors' performance and penalties.
Data and links
A registry of public assets on one national map.
Links to budget, procurement, audit, local governments and hazard data.
Risks and safeguards
Ghost projects: payment needs independent evidence of delivery.
Falsified photos: images are time-stamped, located and cross-checked.
Success measures
Projects finished on time and on budget.
Share of payments backed by independent verification.
Environment and natural resources
Today: Department of Environment and Natural Resources, mines, lands, forestry and environmental bureaus.
Before
Illegal clearing is found months later; a clearance application disappears into a regional office.
After the rebuild
A change in forest cover raises an alert within days and a ranger is sent; every permit application and comment is visible online.
Agents it would run
A permits agent that handles environmental clearances with public comment built in.
A monitoring agent that watches forests, coasts, air and water from satellites and sensors.
A land agent that keeps surveys and public land records consistent.
A climate agent that tracks emissions and adaptation projects.
Work humans keep
Granting or refusing permits for mines, logging and large projects.
Enforcement and community consultation, including with indigenous peoples.
Data and links
Links to the land registry, agriculture, energy, local governments and science agencies.
Risks and safeguards
Capture by applicants: all contacts on a permit are logged.
Communities without internet: consultation also by radio, text and assemblies.
Success measures
Time to detect and act on violations.
Permit decisions within deadline, with reasons published.
Interior and local government
Today: Department of the Interior and Local Government.
Before
A small municipality has no IT staff, so its residents get paper while the city next door gets apps.
After the rebuild
Every local government runs the same shared services from day one, with its own data and its own choices.
Agents it would run
A local performance agent that compares services across provinces, cities and towns.
A support agent that gives small local governments the same tools as large ones.
A disaster coordination agent that links local responders.
A complaints agent that routes citizens' concerns about local services.
Work humans keep
Oversight decisions and discipline of local office holders, under due process.
Mediation between levels of government.
Data and links
Links to every local government, budget, audit, police and fire services.
Risks and safeguards
Centralisation: local governments own their tenant and data; the centre sees comparisons, not control.
League tables used to punish: figures come with context.
Success measures
Share of local governments on shared services.
Gap in service quality between rich and poor localities.
Justice
Today: Department of Justice, prosecutors, the investigation and immigration bureaus, corrections, parole, and the public attorney.
Before
A detainee stays in jail past their release date because the paperwork is in another office.
After the rebuild
The release date is computed from the court's own record and the corrections officer is alerted weeks ahead.
Agents it would run
A prosecution support agent that organises evidence and checks filings for completeness.
A legal aid agent that explains rights and prepares documents for the public attorney's clients.
An immigration agent that handles routine visas and extensions.
A corrections agent that tracks sentences, good-conduct credits and release dates so no one is held a day too long.
Work humans keep
Every decision to charge, dismiss, detain, deport or release.
Lawyers advise clients; agents do not give final legal advice.
Data and links
Links to the courts, police, prisons, the public attorney and human rights bodies, each with strict role limits.
Risks and safeguards
Predictive profiling: no risk scores for individuals in charging or detention.
Unequal arms: the public attorney gets the same tools as the prosecution.
Success measures
People held beyond lawful release (target: none).
Time from complaint to resolution; legal aid coverage.
Foreign affairs
Today: Department of Foreign Affairs, embassies and consulates.
Before
A passport renewal needs an appointment weeks away and a trip to a city.
After the rebuild
A renewal with nothing changed is confirmed on a phone or at a barangay counter, and the passport is delivered.
Agents it would run
A consular agent that renews passports and authenticates documents without a visit when identity is already verified.
A citizen assistance agent for Filipinos abroad, around the clock.
A briefing agent that summarises treaties, reports and news for diplomats.
A translation agent for working documents.
Work humans keep
Diplomacy, negotiations and policy.
Consular decisions on refusals and on citizens in distress.
Data and links
Links to civil registry, immigration, migrant worker services and the identity system.
Risks and safeguards
Identity fraud: biometric check with a human fallback.
Sensitive cables: classified work runs only on sovereign, isolated systems.
Success measures
Days to issue a passport.
Response time to citizens abroad.
Defence and civil defence
Today: Department of National Defense, the armed forces, the civil defence office and the disaster council.
Before
After a typhoon, responders learn which villages are cut off from social media posts.
After the rebuild
Before landfall, each barangay has a plan, stocks are already in place, and missing reports from an area raise an alarm by themselves.
Agents it would run
A logistics agent for supplies, maintenance and personnel records.
A disaster agent that fuses forecasts, hazard maps and population data into evacuation and relief plans.
A veterans agent that delivers pensions and benefits proactively.
A cyber defence agent that helps protect military networks.
Work humans keep
Command decisions. No agent selects or engages a target; lethal force is always a human decision under law.
Declaring emergencies and ordering evacuations.
Data and links
Disaster data shared with science agencies, local governments, social welfare and health.
Military systems kept apart from civilian platforms.
Risks and safeguards
Mission creep into domestic surveillance: barred by law and by architecture.
Automation of force: prohibited.
Success measures
Warning time and evacuation completeness.
Days to pay veterans' benefits.
Information and communications technology
Today: Department of Information and Communications Technology and the privacy and cybercrime bodies.
Before
Each agency buys its own system, and none of them connect.
After the rebuild
Agencies assemble services from shared parts in weeks, and an officer with an idea can test it safely the same day.
Agents it would run
Platform agents that operate the shared foundation: deployment, scaling, patching.
A security agent that tests government systems continuously.
A connectivity agent that maps coverage gaps and plans public access points.
A developer assistant that helps public servants build small tools safely.
Work humans keep
Platform owners accountable for uptime and security.
Standards decisions, made in the open.
Data and links
Operates the exchange and platform but cannot read the content that passes through them.
Risks and safeguards
A single point of failure or control: the platform operator is separate from the governance authority and from data owners.
Vendor lock-in: open standards and exit plans in every contract.
Success measures
Platform uptime, also during disasters.
Time to launch a new service; share of the population with affordable access.
Science and technology
Today: Department of Science and Technology, the weather and volcanology agencies, research councils.
Before
A warning is issued for a province; a family cannot tell whether it means their street.
After the rebuild
Each household gets a warning for its own location, in its own language, with what to do and where to go.
Agents it would run
Forecast agents that turn weather, flood, volcano and earthquake data into warnings for each barangay.
A research assistant for public scientists.
A grants agent that matches researchers to funding and cuts paperwork.
A language agent that builds open models for Philippine languages.
Work humans keep
Scientists own the warnings and the science.
Research ethics and priorities.
Data and links
Open hazard and climate data for every agency and the public.
Risks and safeguards
False alarms eroding trust: warnings show confidence and are reviewed after each event.
Sovereign capability: public compute and models so the country is not only a customer.
Success measures
Warning lead time and accuracy.
Open datasets and models released.
Energy
Today: Department of Energy, the energy regulator, the grid and electrification bodies.
Before
A solar project waits on signatures from many offices in sequence.
After the rebuild
All offices work on one shared file in parallel, and the developer sees every step and deadline.
Agents it would run
A grid agent that forecasts demand and outages.
A permits agent for power projects, with one timeline across agencies.
A consumer agent that explains bills and handles complaints.
An electrification agent that plans microgrids for remote islands.
Work humans keep
Tariff rulings and licensing decisions.
Emergency load decisions.
Data and links
Links to environment, local governments, trade and the grid operators.
Risks and safeguards
Attacks on the grid: control systems are isolated from agent platforms.
Opaque pricing: tariff models are published.
Success measures
Time to permit a project.
Outage hours; households with reliable power.
Tourism
Today: Department of Tourism.
Before
A small homestay owner cannot get accredited without travelling to a regional office.
After the rebuild
Accreditation is done from a phone with a video inspection, and the homestay appears in the national listing.
Agents it would run
A visitor agent that answers questions and handles entry formalities in many languages.
An accreditation agent for hotels, guides and operators.
A destination agent that manages crowding and protects fragile sites.
Work humans keep
Standards, promotion strategy and enforcement.
Data and links
Links to immigration, transport, environment and local governments.
Risks and safeguards
Tracking visitors: only aggregate flows.
Small operators excluded: assisted onboarding.
Success measures
Time to accredit.
Visitor satisfaction; income reaching local communities.
Housing and settlements
Today: Department of Human Settlements and Urban Development and the housing agencies.
Before
A family in a danger zone is on several lists and none of them leads to a home.
After the rebuild
One case owner follows the family from the list to a key, and they can see where they stand.
Agents it would run
A housing agent that matches families to programmes and financing.
A permits agent for building and subdivision approvals.
A planning agent that tests land-use plans against hazards.
Work humans keep
Allocation decisions and relocation, always with consultation.
Approval of plans.
Data and links
Links to land, social welfare, the housing fund, local governments and hazard maps.
Risks and safeguards
Forced relocation by algorithm: prohibited; relocation needs human process and consent.
Queue jumping: the list order is visible to those on it.
Success measures
Families housed out of danger zones.
Time to approve a building permit.
Government corporations and agencies
Today: Social insurance and pension funds, the housing fund, the land registry, the professional regulator, water, ports, lottery, development banks and other attached agencies.
Before
A retiree submits proof of life and years of records to claim a pension the fund already owes.
After the rebuild
The fund tells the retiree their pension is ready, confirms the account, and pays on the first day they are entitled.
Agents it would run
A member agent for pensions and insurance that keeps contributions right and pays claims proactively.
A registry agent for land titles and professional licences issued as verifiable credentials.
A utility agent for billing, leaks and service complaints.
A governance agent that reports each corporation's performance and finances to its owners: the public.
Work humans keep
Boards and managers are accountable for results.
Claim denials, disciplinary cases and investment decisions.
Data and links
Links to civil registry, employers, revenue, banks and the courts.
Risks and safeguards
Funds used as political tools: rules as code and public reporting.
Share of claims paid without an application; cost per transaction.
The legislative branch
Congress makes the rules that every agent follows. Its own agents help members understand what a bill would do, hear more citizens, and see whether money was spent as the law said. Each chamber controls its own agents and data.
The Senate and the House of Representatives
Today: The Senate, the House of Representatives, their committees, secretariats and budget and research offices.
Before
A bill's effect on a fishing town is unknown until after it passes; citizens' comments reach a hearing only if someone can travel to the capital.
After the rebuild
Before the vote, every member and every citizen can see whom the bill affects and what it costs, and a fisher's voice message is part of the record.
Agents it would run
A drafting agent that turns a policy idea into bill text, finds every law it would touch and flags conflicts.
An impact agent that estimates costs, who gains and who loses, by region and income group, with the model published.
A consultation agent that gathers comments from citizens in any language, by text, voice or assembly, and summarises them fairly with every comment still readable.
A spending tracker that follows each appropriation against the law that authorised it.
A records agent that publishes votes, attendance, hearings and amendments in searchable form the same day.
A rules-as-code agent that publishes each new law in a form agencies' agents can apply.
Work humans keep
Legislators debate, amend and vote. Agents never vote, sponsor or decide what becomes law.
Committee staff judge evidence and write the reports.
Data and links
Each chamber is its own tenant with its own keys: the executive cannot read its drafts.
Reads anonymised statistics, budget and audit data; publishes laws to the rules registry.
Risks and safeguards
Agents shaping law quietly: all agent-drafted text is marked, and its prompts and sources are kept.
Fake public comment: one verified person, one voice, with privacy.
Unequal tools: majority and minority members and both chambers get the same agents.
Success measures
Share of bills with a published impact analysis before the vote.
Citizens taking part in consultations, by region.
Days from enactment to rules as code.
The judiciary
The courts gain the most from agents doing clerical work, and have the most to lose if agents drift into judging. The line is absolute: agents manage, research and draft; judges decide. The judiciary runs its own tenant under its own rules.
The Supreme Court and court administration
Today: The Supreme Court, the Court of Appeals, the anti-graft and tax courts, the court administrator, the judicial and bar council.
Before
A litigant travels to court to learn that the hearing is postponed again.
After the rebuild
Hearings are scheduled when everyone is confirmed, notices arrive by text, and the litigant can follow the case from home.
Agents it would run
A case management agent that files, schedules, serves notices and tracks every deadline.
A research agent that finds statutes and past rulings, with every citation checked against the official text.
A drafting assistant for orders and summaries that the judge edits and signs.
A translation and transcription agent for hearings in any Philippine language.
A consistency agent that shows a judge how similar cases were decided, for information only.
Work humans keep
Judges always decide. No agent rules on a case, weighs credibility or sets a sentence.
The Court alone governs its agents, data and rules.
Data and links
The judiciary is its own tenant with its own keys and hosting.
Receives filings from prosecutors, the public attorney and lawyers; sends orders to prisons, registries and police through the exchange.
Risks and safeguards
Invented citations: the research agent may cite only what it can open in the official reporter.
Automation bias: judges see agent output as drafts, and courts publish how agents are used.
Sealed and family cases: stricter access and no reuse.
Success measures
Time to disposition.
Age of the oldest pending cases.
Hearings postponed.
Trial courts and online dispute resolution
Today: Regional, metropolitan and municipal trial courts, Shari'ah courts, family courts, small claims, barangay justice.
Before
A market vendor owed a small sum gives up because the claim costs more days than it is worth.
After the rebuild
They file from a phone, the other side answers online, an agent helps them settle or a judge rules on the papers within weeks.
Agents it would run
A backlog triage agent that sorts cases by age, urgency and readiness, and proposes which can be resolved quickly.
An online dispute resolution agent for small claims that helps both sides state their case, exchange evidence and try to settle.
A self-help agent that explains procedure to people without lawyers.
A detention agent that alerts the judge when someone has been held longer than the likely sentence.
Work humans keep
A judge decides every case, including every small claim that does not settle.
Mediators and barangay conciliators lead settlement.
Data and links
Links to barangay justice records, prosecutors, the public attorney, prisons and the land and civil registries.
Risks and safeguards
Pressure to settle: either side can ask for a judge at any time.
Digital divide: court help desks and barangay counters.
Success measures
Small claims resolved within the deadline.
Detainees awaiting trial; backlog cleared per year.
The constitutional commissions
The three commissions are independent by design, and the architecture keeps them so: each holds its own keys, and none depends on the offices it checks.
Elections
Today: Commission on Elections.
Before
A voter finds on election day that their name is at another precinct.
After the rebuild
Their registration follows their address, and they receive their precinct and a sample ballot weeks before.
Agents it would run
A voter services agent for registration, transfer and precinct finding.
A logistics agent for materials, staff and polling places.
A campaign finance agent that publishes contributions and spending.
A results transparency agent that publishes every precinct return and lets anyone recount the totals.
An information integrity agent that answers voters' questions from published election rules only.
Work humans keep
People run polling places, count under observation and proclaim winners.
The Commission decides every dispute.
Data and links
Its own tenant and keys, apart from every other branch.
Reads civil registry and address data; publishes returns as open data.
Risks and safeguards
Ballot secrecy is absolute: no agent ever sees how a person voted.
No agent counts votes or decides results; a paper trail and random manual audits check the count.
Political targeting: voter data is never shared with candidates' agents.
Success measures
Registration accuracy.
Time to publish precinct returns; audit match rate.
Audit
Today: Commission on Audit.
Before
An audit report appears a year or more after the money is gone.
After the rebuild
A doubtful payment is flagged before it is released, and the audit team is on site the same week.
Agents it would run
A continuous audit agent that checks every public payment against budget, contract and delivery as it happens.
An anomaly agent that flags unusual prices, split contracts and repeated winners.
A physical verification agent that compares paid works with images and citizen reports.
A reporting agent that drafts audit findings with the evidence attached.
Work humans keep
Auditors judge findings, hear the agency's side and issue disallowances.
The Commission sets audit rules and priorities.
Data and links
Standing read-only access to the financial events of every public body.
Its own tenant: no audited body can see or alter audit work.
Risks and safeguards
False flags burdening honest offices: flags are leads, not findings.
Auditors captured: flags and their handling are themselves logged and sampled.
Success measures
Share of spending audited in real time.
Losses prevented, not only found.
The civil service
Today: Civil Service Commission.
Before
A clerk fears that an agent will take their job, and nobody tells them otherwise.
After the rebuild
They are told in writing that their post is secure, choose a new role as a case owner or agent supervisor, and train for it on paid time.
Agents it would run
A hiring agent that runs open, merit-based recruitment and checks qualifications against registries.
A learning agent that gives each civil servant a personal reskilling path.
A workforce agent that shows where people are needed as routine work shrinks.
An integrity agent that manages declarations of assets and conflicts of interest.
Work humans keep
Appointments, promotions and discipline, by accountable officers under due process.
Unions and staff councils help design the change.
Data and links
One personnel record for every civil servant, which they can see and correct.
Risks and safeguards
Algorithmic hiring bias: agents screen for completeness only, and people assess merit.
Monitoring staff: no keystroke or sentiment surveillance.
Success measures
Civil servants reskilled and redeployed.
Time to fill a vacancy; staff trust in the change.
Accountability bodies
The same logs that let agents work also let watchdogs see. These bodies get standing access to evidence and their own agents to read it.
The Ombudsman
Today: Office of the Ombudsman.
Before
A complaint about a bribe needs documents the complainant cannot get.
After the rebuild
The complaint points to a transaction; the records are already in the log and the investigator starts with evidence.
Agents it would run
An intake agent that takes complaints in any language, protects the complainant's identity and gathers the records.
An evidence agent that assembles timelines from logs, contracts and payments.
A case tracking agent that keeps investigations within their deadlines.
Work humans keep
Investigators and prosecutors decide whether to charge.
Protection of whistleblowers.
Data and links
Access to audit logs and financial events by legal process; its own tenant.
Risks and safeguards
Weaponised complaints: agents gather facts, and people assess them.
Leaks: complainant identity is sealed by design.
Success measures
Time from complaint to resolution.
Conviction and recovery rates.
Human rights
Today: Commission on Human Rights.
Before
A person harmed by an automated decision has no idea whom to ask or what happened.
After the rebuild
They can see the decision's trail, and the Commission can examine the agent that made the proposal.
Agents it would run
A monitoring agent that reviews how every public agent affects rights, with access to the registry and evaluation results.
A complaints agent for rights violations, including harm caused by AI.
A detention watch agent that checks custody records for unlawful or overlong detention.
Work humans keep
Investigators, and Commissioners who issue findings.
Visits to places of detention.
Data and links
Standing access to the agent registry, incident reports and custody logs.
Risks and safeguards
An oversight body without teeth: its findings on agents trigger mandatory review or suspension.
Success measures
Time to act on a complaint.
Rights impact reviews completed before launch.
Anti-corruption analytics
Today: A shared function of the audit commission, the Ombudsman, the anti-money-laundering council and the civil service.
Before
A pattern of overpriced contracts is visible only to someone who reads thousands of pages.
After the rebuild
The pattern appears on an investigator's desk with the documents attached.
Agents it would run
A network agent that maps links between contractors, owners and the officers who approve their contracts.
A price agent that compares what different offices pay for the same thing.
A declarations agent that compares office holders' declared assets with registries of land, vehicles and companies.
A pattern agent for split purchases, rushed year-end spending and repeated emergency buying.
Work humans keep
Every lead is assessed by an investigator before anyone is named.
Decisions to investigate follow the law, not the score.
Data and links
Uses public money, contract and registry data. It does not use private messages, location or social media.
Risks and safeguards
Becoming a tool against opponents: no single office controls it, queries are logged, and results are checked by a second body.
Surveillance creep: scope is fixed in law to public funds and office holders' declared interests.
Success measures
Leads that become cases.
Price gaps closed between offices.
Local government
Most people meet the state at the barangay hall and the city hall. Every local government receives the full foundation as a shared service, keeps its own data and choices, and is no longer limited by its own IT budget.
Regions
Today: Regional offices of national agencies and regional development councils.
Before
Each regional office repeats what the central office already checked.
After the rebuild
Regional staff spend their time in the field, because checking is done once.
Agents it would run
A regional planning agent that combines provincial plans and spots gaps.
A shared services agent that lets small provinces pool specialists.
A regional disaster agent.
Work humans keep
Regional councils set priorities.
Data and links
Regional views of national and local data.
Risks and safeguards
A redundant layer: regions exist to coordinate, not to add a queue.
Success measures
Approvals that needed no regional step.
The Bangsamoro Autonomous Region
Today: The Bangsamoro government, its parliament and ministries, Shari'ah courts.
Before
National systems do not fit the region's laws and languages, so services arrive last.
After the rebuild
The region configures shared services to its own laws and languages from the first day.
Agents it would run
The same foundation run as the region's own tenant, with its own agents for its ministries.
Agents that work in Maguindanaon, Tausug, Maranao and other languages of the region.
A halal, endowment and pilgrimage services agent.
A normalisation and reconstruction tracking agent.
Work humans keep
The region's own institutions govern its agents, data and rules, within its autonomy.
Shari'ah judges and traditional leaders in their roles.
Data and links
The region holds its own keys. Data sharing with national bodies follows the organic law and agreements.
Risks and safeguards
Autonomy overridden by national platforms: tenancy and keys are the region's.
Uneven connectivity: priority for offline nodes.
Success measures
Service parity with the national average.
Services available in the region's languages.
Provinces
Today: Provincial governments, provincial hospitals and engineering offices.
Before
A patient is referred from one hospital to another with a paper slip and no bed.
After the rebuild
The bed is confirmed before the ambulance leaves.
Agents it would run
A hospital network agent for referrals between towns.
A roads and assets agent.
A provincial disaster agent.
A support agent for small municipalities.
Work humans keep
Provincial boards and governors' offices decide policy and budgets.
Data and links
Links to towns, national health, public works and social welfare.
Risks and safeguards
Duplicate systems: provinces configure shared services and do not buy their own.
Success measures
Referral time.
Provincial road condition.
Cities and municipalities
Today: City and municipal governments: business permits, local taxes, civil registry, health, social welfare, planning.
Before
Renewing a business permit every January means a week of queues at several windows.
After the rebuild
The renewal is offered in December with the fees computed; the owner pays the way they prefer and the permit arrives.
Agents it would run
A permits agent for business, building and occupancy, with fire and health checks in parallel.
A local revenue agent for property tax and fees, payable through any channel.
A civil registry agent for births, marriages and deaths.
A city services agent for waste, water, streetlights and potholes, fed by residents' reports.
A budget agent that publishes the local budget as people can read it.
Work humans keep
Local councils and mayors' offices set policy.
Inspectors, health workers and social workers in the field.
Data and links
Each local government is its own tenant.
Links to national registries, revenue, fire, health and the audit commission.
Risks and safeguards
Local capture: queue order and reasons are logged.
Poor towns falling behind: shared services are free to them, with central support.
Success measures
Hours to get a permit.
Local revenue collected at lower cost; complaints resolved.
Barangays
Today: Barangay governments, health workers, day care, peace and order, barangay justice.
Before
An elderly resident needs a certificate to get another certificate, and travels to town for both.
After the rebuild
They ask at the barangay hall in their own language; the barangay worker confirms with them, and everything is issued there.
Agents it would run
A barangay service agent: a voice in the local language, at the hall, by text and by call, that can start any government service.
A certificates agent that issues clearances and residency records instantly as verifiable credentials.
A community health agent that supports health workers with home visit lists.
A disaster agent that keeps the list of who needs help to evacuate.
A mediation agent that prepares barangay justice cases.
Work humans keep
Barangay officers, health workers and volunteers: the human face of the state.
Assisted service for anyone who prefers a person.
Data and links
An offline node in each barangay holds what it needs and synchronises when it can.
Links to the city or municipality, health, social welfare and disaster bodies.
Risks and safeguards
Gatekeeping by local figures: any resident can also reach services directly by phone.
Privacy among neighbours: barangay staff see only what the task needs.
Success measures
Share of services completed at the barangay.
Residents served in their own language; evacuation lists up to date.
The central bank, statistics and public safety
These bodies need special care: the central bank's independence, the neutrality of statistics, and firm civil-liberty limits on police powers.
The central bank and financial regulators
Today: Bangko Sentral ng Pilipinas, the securities, insurance and deposit insurance bodies, the anti-money-laundering council.
Before
A scam victim reports to a bank, a wallet and the police, and the money is gone before anyone acts.
After the rebuild
One report freezes the trail across providers within minutes, under rules the regulators set and people review.
Agents it would run
A supervision agent that reads banks' reports and flags risks early.
A payments agent that keeps the state's payments open to every licensed provider.
A consumer protection agent for complaints about banks, lenders and e-wallets.
A market surveillance agent for fraud and scams.
A financial inclusion agent that helps people open a basic account with their national identity.
Work humans keep
Monetary policy and every supervisory sanction.
The independence of the central bank is untouched.
Data and links
Its own tenant.
Financial data moves only with consent or a clear legal basis.
Risks and safeguards
Financial surveillance: no general access to people's transactions.
Dependence on one payment provider: neutrality is a rule of the platform.
Success measures
Adults with an account.
Time to freeze a reported scam; cost of a payment.
Statistics and civil registration
Today: Philippine Statistics Authority.
Before
Poverty figures arrive years apart; a clerk's typing error in a birth record takes a court case to fix.
After the rebuild
Key figures are current, and a clear clerical error is corrected at the barangay with evidence and a log.
Agents it would run
A registry agent that keeps civil registration complete and consistent.
A statistics agent that produces figures from administrative data in near real time.
A privacy agent that anonymises data before any use for analysis.
A data quality agent that finds gaps and duplicates.
Work humans keep
Statisticians own methods and releases.
Independence from the offices that are measured.
Data and links
Steward of the civil registry and the identity system; provider of anonymous data to planners and researchers.
Risks and safeguards
Re-identification: strict disclosure control and tested anonymisation.
The identity system as a tracking tool: it confirms who someone is and does not record where they go.
Success measures
Births registered within the month.
Timeliness of key statistics.
Police
Today: Philippine National Police.
Before
A victim reports a theft and never hears back.
After the rebuild
They receive a case number, the officer's name and updates until the case closes.
Agents it would run
A reporting agent that lets anyone report a crime and follow the case.
A records agent that writes reports from officers' dictation and keeps custody logs.
A dispatch agent that sends the nearest unit.
An evidence agent that indexes lawfully gathered evidence.
A body camera agent that stores footage that neither officers nor agents can alter.
Work humans keep
Every stop, search, arrest and use of force is a human act under law.
Warrants come from judges.
Data and links
Links to courts, prosecutors and prisons through the exchange, with warrants checked by machine.
Risks and safeguards
Mass surveillance: no live face recognition in public places, no social scoring, no predictive policing of individuals.
Data fishing: each query needs a case number and is logged for the human rights commission and the courts.
Abuse in custody: custody clocks alert supervisors and the public attorney.
Success measures
Response time.
Cases solved; complaints against officers; unlawful detention (target: none).
Fire and emergency response
Today: Bureau of Fire Protection, emergency lines, jail management.
Before
A caller in a panic cannot explain where they are, and minutes are lost.
After the rebuild
The call is located and understood at once, and the nearest crew is moving while a person stays on the line.
Agents it would run
An emergency call agent that understands any Philippine language and locates the caller.
A dispatch agent across fire, medical and rescue.
A fire safety agent that schedules inspections by risk and issues clear certificates.
A hydrant and equipment agent.
Work humans keep
Firefighters and medics respond; inspectors decide on closures.
Data and links
Links to city permits, buildings, hospitals and disaster bodies.
Risks and safeguards
Inspection as a source of bribes: schedules are set by published risk rules, and results are logged.
Location data kept only for the emergency.
Success measures
Minutes from call to arrival.
Fire deaths; inspections done on schedule.
Governance and law
An AI governance authority
An independent AI and Data Authority sets the autonomy tiers, certifies agents, runs the registry and can suspend any agent in any branch's executive agencies. Within the judiciary, Congress and the commissions, each body certifies its own agents to the same public standard.
It is separate from the platform operator, so the builder never marks its own work.
Members serve fixed, staggered terms and are chosen through an open process that includes the three branches, the professions and civil society.
It reports to Congress and the public every year, and its decisions can be challenged in court.
The laws needed
An AI in government act: the accountable officer, the asymmetry rule, autonomy tiers, the registry, evaluation before launch, and the right to a human.
A data exchange and once-only act: offices must reuse data and may not ask again; purposes are defined in law.
A stronger privacy law and regulator, with the data log as a right.
An administrative procedure law for automated decisions: notice, reasons, appeal and remedies.
A digital identity and credentials law with limits on tracking.
A public records and logs law: what is logged, for how long, and who may see it.
Limits on policing technology, written into law.
Protection for civil servants during the transition.
Standards
Open standards for identity, credentials, data exchange, agent permissions, logs and rules as code.
Alignment with international practice where it exists: the OECD principles, the UNESCO recommendation and risk-based approaches such as the European Union's.
A procurement model
Buy outcomes and shared parts, not one system per agency.
Small, modular contracts that local firms can win; pay for working software.
Every contract includes open standards, data portability, the right to switch models and an exit plan.
Payment services are provider-agnostic: any licensed provider can connect on the same terms.
Public test suites that any supplier's agent must pass.
Open source
Code paid for by the public is public, unless security requires otherwise.
Local governments, universities and firms can reuse and improve it.
Shared with other countries as digital public goods.
Workforce change
A written guarantee: no civil servant is dismissed because an agent took over their tasks.
New roles: case owner, agent supervisor, service designer, data steward, evaluator, field worker, community helper.
Paid learning time and a national public service academy for AI skills.
Staff and unions co-design each service's change.
Citizen oversight
A citizens' assembly, chosen by lot from every region, reviews the system each year and can require a response.
A public registry, public evaluations, public incidents and a public spending ledger.
Open channels for anyone to report harm, with protection from retaliation.
Civil society, the press and universities get research access to anonymised data and to test environments.
A phased roadmap
This plan sets budget aside on purpose, so nothing here is priced. It is sequenced: each phase unlocks the next, and no phase begins until the measures of the one before are met and published.
Year 0 to 1
Foundations, law and first pilots
Build
Pass the enabling laws: an AI in government act, a data exchange and once-only act, and a stronger privacy regulator.
Create the independent AI governance authority and the platform operator.
Launch identity and consent, the data exchange, the first five registries and the agent platform with tiers A0 to A2.
Open the agent registry and the public evaluation suites.
Pilots and scale-up
Birth life event in three provinces, including one island province.
Business start in ten cities and municipalities.
Typhoon mode in one region.
Barangay service agents in two hundred barangays, in four languages.
Measures
Pilot services completed with no form.
Pilot users who chose a human channel, and their satisfaction.
Zero rights-affecting decisions made without a person.
Years 1 to 3
Life events at scale
Build
All core registries connected; rules as code for the fifty most used laws and programmes.
Tier A3 opened for agents that pass evaluation.
Continuous audit live for national spending; the public spending ledger opens.
Offline barangay nodes in every disaster-prone province; ten languages and sign language supported.
Pilots and scale-up
Ten life events nationwide: birth, school, first job, business, illness, disaster, moving home, working abroad, retirement, death.
Online dispute resolution for small claims.
Legislative impact and consultation agents in both chambers.
Measures
Share of the top hundred services that ask for no document the state already holds.
Days from disaster to aid.
Civil servants in new roles.
Years 3 to 5
Every branch, every local government
Build
All departments, courts, commissions and local governments on the shared foundation as their own tenants.
Tier A4 for selected high-stakes actions under two-person approval.
Sovereign model hosting at scale; open models for the major Philippine languages.
All barangays served, online or offline.
Pilots and scale-up
Proactive benefits with no application.
Personal citizen agents that deal with government on a person's behalf, under their control.
Measures
Court backlog and time to disposition.
Spending audited in real time.
Service gap between the richest and poorest localities.
Years 5 to 10
Government that acts before you ask
Build
Most services delivered proactively; applications become the exception.
Agencies reorganised around life events.
Regional cooperation on cross-border identity and credentials.
An independent ten-year review, with the power to recommend switching anything off.
Pilots and scale-up
New uses proposed and tested by citizens, schools and local governments on the open platform.
Measures
Trust in government services.
Hours citizens spend on government per year.
Appeals upheld, as a check that people still correct the machine.
Risks and how each is handled
Bias and unfair outcomes
Evaluation by region, language, sex, age and disability before launch and continuously.
The asymmetry rule: agents cannot deny or penalise.
Public results and independent audits; automatic demotion on failure.
Agent error and failure
Autonomy tiers and limits per case and per day.
Instant undo for reversible actions; two-person approval for the rest.
Traces of every step; published incident reports; a kill switch.
Cyberattack and manipulation of agents
Zero trust, least privilege and short-lived credentials for agents.
Tools, not prompts, enforce the rules, so a manipulated agent still cannot exceed its permissions.
Continuous red-teaming, bug bounties, isolated systems for defence, the grid and elections.
Capture by vendors or by those in power
Open standards, open source and model-agnostic design with exit plans.
Separate tenants and keys for each constitutional body.
An independent governance authority with fixed terms, and oversight by Congress, the courts, the audit and human rights commissions and a citizens' assembly.
Surveillance creep
Purpose limits in law and in the exchange: a request without a lawful purpose fails.
No live face recognition in public places, no social scoring, no predictive policing of individuals.
Every access appears in the person's data log; misuse is a crime.
Digital exclusion
Text, voice, paper and in-person channels with the same rights and deadlines.
Barangay counters and trusted helpers; no smartphone or bank account required.
Languages and accessibility are launch conditions, not later additions.
Loss of human skill and judgement
Automate the task, not the learning: juniors still do reviewed, real work.
Regular drills where services run without agents.
Case owners rotate through manual handling.
Disasters and outages
Several regions, offline nodes and typhoon mode.
Paper with verifiable codes.
Recovery tested every year, in public.
Sources
Each source is marked verified (a primary source we read for this plan), reported (press or secondary reports, or a page seen only through search summaries) or background (general knowledge, not re-read). The plan itself is a design, not a description of anything that exists.
e-Estonia: X-Road interoperability servicesverified Each system connects through the exchange; outgoing data is signed and encrypted, incoming data is authenticated and logged. The once-only principle is background knowledge, not stated on that page.
India Stackreported Identity, payments, documents and consent-based data sharing as open building blocks. The exclusion concern is background knowledge.
AIPhilippines.org is a privately owned, for-profit company and is not affiliated with any government agency. This page is an independent concept proposal written by AI Philippines (Osome Apps Inc.). It is not government policy, it was not requested, reviewed or endorsed by any government body, and nothing here describes a service that exists today.
It names today's institutions only so that readers can find their way; it does not speak for them. It names no person and takes no side in party politics. It was written with AI assistance and may contain errors: check the sources before relying on any fact. Last updated 2026-11-11.
Comments and corrections: [email protected] · Osome Apps Inc., Santa, Pateros, Metro Manila
