Blog/Product Strategy
Product Strategy

Moyasar × HUMAIN Agentic Commerce: Payments When the Buyer Is an AI Agent

i
iFynx StudioSep 20, 2026 · 7 min read
moyasar agentic commerce humain ai payments

Machine-to-machine checkout is leaving the demo stage

On the sidelines of Money20/20 Middle East (14–16 September 2026), Moyasar Chief Strategy Officer Muath Alduhishy told Sharikat Mubasher the Saudi payments company showcased Agentic Commerce, developed with HUMAIN, enabling AI agents to initiate and complete payments. Moyasar plans further AI-powered products while focusing on Saudi Arabia and expanding across GCC markets. Industry coverage around the event also stresses the control plane merchants will need: authentication, authorisation, spending limits, refunds, and disputes when no human sits in the checkout loop — with market launch ambitions discussed toward end-2026.

What product teams must design next

1. Agent identity is the new cardholder. Who is the principal? What mandate did they grant the agent? Encode spend caps, merchant allowlists, and time-boxed credentials in the payment UX for both consumer and B2B flows.

2. Dispute UX without a human click path. Classic chargeback journeys assume a person remembers the purchase. Agentic commerce needs audit timelines the agent (and the user) can both replay.

3. GCC expansion is a compliance product. Cross-border agent payments amplify KYC/AML and local scheme rules. Design country packs early — iFynx’s regional product work treats regulation as a design system token set, not a late legal review.

iFynx takeaway

Moyasar × HUMAIN puts Saudi infrastructure on the agentic commerce map. Build payment experiences where the mandate is visible, limits are enforceable, and disputes are explainable — before your first agent buys something you cannot reverse cleanly.


Agentic checkout acceptance criteria for payment teams

When the buyer is an AI agent, classic “click to pay” assumptions collapse. Moyasar × HUMAIN demos only become production if merchants and issuers can answer who authorized what, with which limits, and how to reverse it.

  • Mandate object: merchant, amount ceiling, currency, expiry, allowed MCC / product classes — all visible to the human principal
  • Human confirmation policy configurable per risk tier: always / above threshold / never for trusted recurring agents
  • Immutable receipt that names the agent ID, model/provider if disclosed, and the human account holder
  • Chargeback / dispute path that does not require the human to “prove they clicked” — they didn’t
  • Idempotency keys on agent-initiated pays so retries do not double-capture

If finance cannot reconcile agent payments in the same ledger as human checkout, agentic commerce stays a conference booth.


Saudi and GCC edge cases agents will hit first

Local rails and habits break generic agent payment designs. Test these before promising volume.

  1. mada and local wallets: agent must present a mandate the issuer already understands — do not invent a parallel auth UX
  2. VAT-inclusive pricing: agent carts must show the same total the human would see on the merchant site
  3. COD-heavy merchants: agent should fail closed if the merchant does not support prepaid agent checkout
  4. Arabic invoice requirements: generated receipts must be locale-correct for ZATCA-adjacent workflows where applicable
  5. Shared corporate cards: agent spend must respect department budgets and dual-control rules, not only personal limits

An agent that “succeeds” while creating an unusable Arabic fiscal invoice is a support incident, not a win.


Ops playbook: piloting agentic payments with a merchant cohort

Run a bounded pilot. Publish the kill criteria before marketing the partnership.

  • Cohort: 10–20 merchants with strong API maturity and bilingual support coverage
  • Daily dashboard: auth rate, reverse rate, human-escalation rate, average mandate age at pay time
  • Joint war-room with Moyasar / HUMAIN / issuer for the first two salary-week peaks
  • Merchant FAQ in Arabic and English explaining how customers revoke an agent’s spending power
  • Exit criteria: fraud rate or dispute rate above agreed threshold pauses new mandates automatically

Agentic commerce is a trust product wearing a payments costume. Design the revoke button as carefully as the pay button.

Treat this as an operating practice, not a one-off redesign workshop. Schedule a monthly review of the metrics and checklists above, assign a single accountable owner, and keep a short changelog of product fixes shipped since the last review. When stakeholders ask for more visual polish, redirect the conversation to the open defects that still block conversion, trust, or operator adoption. Depth compounds when the same checklist is reused across markets with only locale-specific payment, language, and compliance adjustments.

Related reading

The building blocks of an agentic checkout

When an AI agent pays on someone’s behalf, the payment flow needs a few components that classic e-commerce never required:

  1. Mandate: a record of what the user authorised: which merchants or categories, maximum amount per transaction and per period, validity dates, and whether the agent may act without asking each time.
  2. Agent credential: a token issued to the agent, linked to the mandate and revocable at any moment, rather than the user’s raw card or account details.
  3. Merchant verification: a way for the merchant to know that a request comes from a recognised agent acting under a valid mandate.
  4. Step-up rules: conditions that force the agent to return to the user, such as a new merchant, a price above the limit, or a changed delivery address.
  5. Receipts and timeline: a human-readable record of every agent action, sent to the user immediately.

Saudi context: open banking and payment initiation

In Saudi Arabia, SAMA’s Open Banking Framework already includes a release for payment initiation services (September 2024), which lets consumers initiate payments from third-party applications in a secure, standardised way. That gives agentic commerce a regulated foundation to build on, alongside card networks and local payment schemes. Product teams should follow the specific consent and authentication requirements of each rail rather than inventing their own.

Designing the user’s control panel

Users will only delegate payments if they feel in control. A good control panel lets them see every active agent and mandate, change limits, pause an agent instantly, and review recent actions. Notifications should be immediate and specific: what was bought, from whom, for how much, under which mandate. In Arabic interfaces, amounts, dates and merchant names must display correctly in right-to-left layouts, and numerals should follow the user’s preference.

Disputes and refunds

Disputes are harder when no person clicked “buy”. The evidence the user, merchant and payment provider need includes the mandate in force at the time, the agent’s request, any step-up confirmation, and the merchant’s response. Designing a shared dispute timeline from these records makes resolution faster and fairer.

Security risks specific to agent payments

Agents read web pages, emails and product descriptions that may contain hidden instructions. The OWASP Top 10 for LLM Applications 2025 lists prompt injection as the top risk and warns about excessive agency. For payments, that means the agent’s payment ability should be limited by the mandate in code, not by the agent’s judgment, and that instructions found in content should never change who gets paid or how much.

Merchant readiness for agent traffic

Merchants will need to prepare too. Product catalogues should expose structured data, such as prices, availability, delivery times and return policies, that an agent can read reliably, rather than relying on marketing copy. Checkout APIs should support agent credentials and return clear, machine-readable errors when a mandate does not allow a purchase. Merchants should also decide how to handle agent-initiated orders in fraud scoring, since traditional signals such as mouse movement and session behaviour disappear when no human is browsing.

B2B agentic payments

For businesses, agentic payments look like automated procurement: reordering supplies, paying recurring invoices or settling with suppliers under approved budgets. Here, mandates map naturally to purchase policies and approval matrices. The same principles apply: limits enforced in code, evidence for every action, and step-up approval when a request falls outside policy. Linking agent payments to e-invoicing records, which are mandatory or being phased in across Saudi Arabia and the UAE, makes reconciliation and audits much easier.

Teams starting now should prototype the mandate and control panel first, test them with real users in Arabic and English, and only then connect live payment rails. The trust model is the product; the payment call is the easy part.

Getting that right early also makes conversations with banks, schemes and regulators far easier.

Frequently asked questions

What is agentic commerce?

Commerce where an AI agent searches, chooses and pays on a user’s or business’s behalf within limits the user sets.

Does the user still need to approve payments?

It depends on the mandate. Many designs let agents act within strict limits and require confirmation for anything outside them.

Who is liable if an agent buys the wrong thing?

That depends on the scheme rules, the provider’s terms and local regulation, which is why clear mandates and evidence trails matter.

Can merchants tell an agent from a human?

They should be able to. Agent credentials and verification are core parts of the emerging designs.

Related: Mastercard Agent Connect · Open finance and AI agents. Building payments or AI products for the Kingdom? See app development for Saudi Arabia and AI development for Saudi Arabia.