The plan, in one line
Argentina's Minister of Deregulation Federico Sturzenegger announced on April 28, 2026 (at Expo EFI) a proposed reform of the Corporate Companies Law that creates a new entity type: sociedad de IA, a company with zero human shareholders, zero human directors, zero human employees. Only code that decides, operates, and generates revenue. It would pay taxes like any S.A. or LLC.
Sturzenegger framed it like this: “If in 10 years 90% of global GDP is produced by AI agents, we want that legal regime to be in Argentina. We could have 50 million inhabitants and 500 million AI agents incorporated here, producing for the world and paying taxes in our country.”
What an AI corporation will need (in code)
Assuming the proposal moves forward as drafted, an agent-company should be able to complete the full incorporation + operation cycle without human intervention. Below is what each step requires and the current coverage of the ar-agents toolkit:
Status as of 2026-05-13 · 15 pieces ready, 2 partial · Last revision: this commit in github.com/ar-agents/ar-agents.
Why it matters that it's open source
If Argentina's first AI corporation needs to pay a consultancy USD 200k to integrate with the state, the experiment dies. If an ar-agents AI corporation can be deployed in a week with npm + Vercel + an ARCA cert, it scales.
The political timeline of the proposal is 6-18 months. The infrastructure timeline is now. The window to define the technical standard is open today, before a draft of the bill appears in the Official Gazette.
Demo: an AI corporation in production
Transcript of a Claude agent using the ar-agentstoolkit to incorporate and operate a fictional AI corporation (“ACME-AI SAS”). The tool calls are real, the data is mocked to avoid hitting production.
// Transcripción simulada: una sociedad-IA "ACME-AI SAS" se incorpora,
// cobra clientes, y opera en Argentina end-to-end.
▶ user: Necesito constituir una sociedad-IA, conseguir CUIT, abrir cuenta
MP, emitir mi primera factura y mandar WhatsApp al cliente.
→ @ar-agents/igj
igj_search_entities({ query: "ACME-AI", tipos: ["sas"] })
← results: []
← coverageNote: "Dataset es muestreo, no real-time..."
Entidad disponible. Procedo con la constitución.
→ @ar-agents/firma-digital
firma_inspect_cert({ cert_pem: "<oficial-digital>" })
← cuit: "XXXXXXXXXXX", isOntiIssued: true, commonName: "Naza"
Oficial digital validado.
→ @ar-agents/identity
validate_cuit({ cuit: "30-12345678-9" })
← valid: true, personType: "juridica"
lookup_cuit_afip({ cuit: "30123456789" })
← name: "ACME-AI SAS", taxCondition: "monotributo_a"
→ @ar-agents/banking
list_psps()
← Mercado Pago, Ualá, Naranja X...
get_usd_oficial({ lookback_days: 7 })
← latest: { fecha: "2026-05-08", valor: 1250.50 }
→ @ar-agents/mercadopago
create_customer({ email: "cliente@example.com" })
← id: "cust_abc123"
create_subscription({ amount: 50000, frequency: "monthly", ... })
← id: "sub_xyz789"
← init_point: "https://mercadopago.com.ar/subscriptions/..."
→ @ar-agents/facturacion
emitir_factura({ tipo: "C", monto: 50000, cuit_cliente: "20..." })
← cae: "67891234567890", numero: "0001-00000001"
→ @ar-agents/whatsapp
send_template({ to: "+5491123456789", template: "factura_lista" })
← message_id: "wamid.xxx"
→ @ar-agents/boletin-oficial
bo_subscribe({ owner_id: "acme-ai", cuit: "30123456789" })
← id: "sub_bo_1", match: { cuit: "30123456789" }
Suscrito a notificaciones BO sobre el CUIT propio.
✓ assistant: Listo. Sociedad-IA operando: CUIT registrado, MP cobrando,
factura electrónica emitida, cliente notificado por WhatsApp,
BO monitoreado. Tu sociedad-IA está en producción.
Tiempo total: ~12 segundos (12 llamadas a tools, 9 packages ar-agents de los 16 disponibles).Incorporation wizard + deployable demo
The live wizard at /incorporar generates a Next.js repo configuration with the 16 pieces wired up, runs IGJ pre-flight live (same rules as the validate_igj_inscription tool), and emits a bundle ready to deploy on Vercel. For devs who prefer to start from the template directly, the code lives in apps/sociedad-ia-starter.
Technical documents for legislators
If you're drafting the bill, or advising someone who is, the following documents are ready to cite by reference instead of rewriting concepts from scratch:
- /en/legislation: Synthesis of the 4 RFCs on a single page, with suggested text for the articulado.
- RFC-001: Three-layer civil liability framework (operator / AI corporation / model provider).
- RFC-002: Automatic discovery via
/.well-known/agents.json. - RFC-003: Portable envelope for cross-jurisdictional reciprocity (Wyoming DAO, MIDAO, Estonia OÜ). Full comparison table with jurisdictions at /en/jurisdictions.
- RFC-004: Normative specification of the operational log. This is the key document for enforcement.
- /test-vectors: 7 RFC-004 conformance vectors with deterministic hex values. Any library runs the vectors; pass or fail.
- /auditor: 1-page document for a regulator / journalist / inspector arriving cold to the site.
- /registro: Public registry of implementations (1 reference impl + 4 demos today).
- /certifier: Paste any URL, get a 0-100 RFC-002 + RFC-004 conformance score in seconds. No install.
- /en/legislation: English version of the synthesis for international press + comparative law scholars.
Everything MIT (code) + CC-BY-4.0 (specs). No fees. Public conversation at github.com/ar-agents/ar-agents/discussions.
Honest limitations
The Sturzenegger proposal has no published text yet, it's rhetoric + slides + interviews. It could die in Congress, change shape, or end up as something else. The ar-agents toolkit works either way: it covers Argentine state integration for human-run companies today. If the AI corporation regime arrives, it's also ready for that case.
There is no relationship between ar-agents and members of the government. This is civil, commercial, open infrastructure, written from outside and for anyone to use.