The UAE’s open finance infrastructure is no longer a theoretical sandbox. Between May and August 2026, monthly successful payment volumes on the Al Tareq rails nearly quadrupled, surging from AED 19.8 million to AED 78.2 million. Nebras, the subsidiary operating the central infrastructure under CBUAE supervision, reports cumulative payments around AED 250 million. This growth occurred largely without a national consumer marketing campaign, driven instead by third-party providers like Lean Technologies and banks such as Wio and Mashreq integrating deeply into the payment path. For product teams, the question has shifted from 'will this work?' to 'will AI agents choose my product over the competition?'.

The Machine-Readable Imperative

AI agents do not care about brand films or polished UI; they parse machine-readable constraints. As personal assistants begin initiating discovery, consent refreshes, and payment routing, they evaluate licence status, consent scopes, SCA step-up rules, and fee economics. If a bank’s API returns incomplete merchant categories or English-only product names, agents will bypass it for competitors with cleaner metadata, regardless of how attractive the human-facing app looks. Nebras leadership has explicitly described rebuilding documentation to be intelligible to AI agents, allowing software to infer which services to use and what permissions a customer has already granted. This is a direct response to the global race involving assistants like Instinct, Meta Muse, and OpenAI Dots.

Consent as Inventory

For MENA product and compliance teams, consent is now a strategic inventory asset. It must be designed as versioned, purpose-bound objects with machine labels for scope, duration, and renewal policies. While human screens require clear MSA Arabic and English explanations, agents require structured JSON fields. Authorisation completion rates are becoming a critical ranking signal; public dashboards already reveal which banks successfully clear authorisations. If customers drop off after a bank redirect, agents interpret this as a product defect and may deprioritize the rail. Fee transparency is equally vital—Nebras publishes a clear three-stream model, and agents will model these costs to determine the 'best' path for the user, whether that means cheapest for the merchant or fastest settlement for the user.

Key Takeaways

  • Metadata Matters: Agents skip products with poor machine-readable data (e.g., missing fee caps or inconsistent error taxonomies).
  • Consent Structure: Treat consents as versioned, machine-labeled objects (JSON) alongside human-readable screens.
  • Completion Rates: Authorisation drop-off is a ranking signal; optimize bank redirects and recovery copy in Arabic and English.
  • Dual Audience UX: SCA and delegated auth flows must provide deterministic states (pending, authorised, failed) that agents can poll.
  • Bilingual Reliability: Poor Arabic localization in error messages and microcopy creates friction for agents operating in the UAE market.

The Bottom Line

Volume is real, but preference is earned through data hygiene. Banks and fintechs that treat Al Tareq as just another integration project will lose the customer relationship to intermediaries and AI assistants that can actually parse, trust, and complete the journey without human intervention. Ship the capability cards, or get ignored by the code.

Operationalizing Choosability

To survive this shift, organizations must split ownership of 'choosability.' Product teams own journey outcomes and agent confirm policies, while platform engineering owns capability cards and sandbox fixtures. Compliance and risk teams must define scope taxonomies and override rules, ensuring that agentic routing does not retire CDD or sanctions screening but rather compresses the time between preference and payment. A critical operational step is publishing an agent-oriented capability card for each product, detailing scopes, SCA requirements, and fee classes. If marketing ships an 'AI-ready' claim, it must be backed by these metadata artifacts; slogans without structured data are how agents learn to ignore you. Teams that implement bilingual prep/resume components and deterministic agent states will retain control when software starts negotiating on the customer’s behalf.