Agentic DPI: The Moment Interfaces Become Infrastructure

The argument

Digital public infrastructure has solved access. Identity, payments and credential portability work at population scale. The remaining barrier is cognitive: people struggle to find schemes, check eligibility and sequence applications. An agentic layer, set out in the Doot whitepaper as Agent One, would let a citizen state a need and have an agent carry out the steps across services, leaving the underlying systems unchanged.

Once software acts instead of advising, authority becomes a data structure. Each agent is cryptographically bound to the citizen’s identity or a sector credential, and carries permissions that are scoped, time-limited, signed and revocable. The system can then answer the question public infrastructure has to answer: what is this software allowed to do, and for whom.

Three design problems follow. The first is non-determinism. Resilience measures handle failures that are visible; confident, plausible and wrong execution passes through the audit trail recorded as success. The second is consent. Fine-grained, purpose-specific consent works for capable users and becomes a formality for first-time and low-literacy users, who accept defaults. A capability-restricted mode or an independent advocate role would close part of that gap. The third is audit. Tamper-evident logs visible to the citizen make disputes symmetric and expose systemic errors such as misclassified eligibility.

Behind all three sits the governing institution. Aadhaar and UPI earned trust through bodies with defined mandates and predictable, auditable outcomes. An agentic layer needs an equivalent, with its legal form, leadership, judicial review and handling of conflicts with large private platforms settled in advance. Deploying at very large scale before that institution exists reverses the sequence that made earlier public infrastructure work.

Built in that order, identity-bound delegation, enforced consent and citizen-visible audit become accountability infrastructure that other countries could adopt.

Key points

• Digital public infrastructure has solved access; the next problem is delegation.
• Agents bound to a citizen’s identity carry scoped, time-limited and revocable permissions.
• Plausible but wrong execution can pass through audit trails as success, which makes non-determinism a governance design problem.
• Consent needs a mode that protects users who accept defaults.
• The governing institution’s legal form, leadership and review must be settled before large-scale deployment.

Evidence and sources

• The Agent One architecture set out in the Doot whitepaper
• India’s digital public infrastructure: Aadhaar, UPI, DigiLocker, ABHA, UMANG and e-Sanjeevani

Limits

• The essay assesses a proposed architecture that has not been deployed.
• Scale figures in the original, such as a billion users and a billion interactions a day, are illustrative.