Why Adyen rebuilds the agent stack.

The Institutional Calculation

Adyen's posture toward the agent-commerce shift reflects something the press notes consistently underread: the company is a Dutch institution first and a technology company second. That ordering is not a criticism. It is an explanation. Founded in Amsterdam in 2006, Adyen built its model on the premise that financial infrastructure should be invisible, reliable, and trusted by the most demanding buyers on earth. Spotify, Booking.com, Microsoft, Uber, McDonald's — these are not accounts you win by being fast. You win them and keep them by being right every time. Adyen's single-platform architecture, which processes authorisations for all of those accounts without routing through a patchwork of regional acquirers, is a structural advantage that took fifteen years to construct. The company did not build it by chasing trends.

The agent-commerce shift posed a different kind of challenge. It did not threaten Adyen's existing book of business — the company's enterprise retention rate held at 97 per cent through Q1 2026. It threatened the marginal decision: the merchant evaluating Adyen against Stripe for a 2027 deployment, designing the system architecture for a procurement workflow where 40 per cent of payment initiation would come from AI agents by year two. In that conversation, Adyen's reliability record was necessary but no longer sufficient. The merchant also wanted to know whether the processor could handle cryptographic agent attestation, agent wallet management, and the reconciliation primitives that let an AI agent report spending to a human principal in a structured format. Adyen's answer, through most of 2025, was that it was evaluating the space. That answer was costing the company deals.

Brouwer commissioned an internal analysis in August 2025, led by chief architect Pieter van den Berg, that quantified the loss. The study examined 34 enterprise merchant evaluations in which Adyen had been shortlisted alongside Stripe in the prior six months. Adyen won 22 of those evaluations, which is consistent with its historical win rate in enterprise payments. Of the 12 it lost, nine cited agent-payments readiness as the primary or secondary decision factor. That was a new category of loss reason. It had not appeared in Adyen's competitive intelligence reports before Q2 2025. Van den Berg's analysis projected that the pattern, if left unaddressed, would erode Adyen's enterprise win rate by roughly eight percentage points by the end of 2026. The projection reached Adyen's board in October 2025. The build decision followed within the month.

What Adyen Is Actually Building

The programme Adyen is executing internally goes by the name Nexus. It is not a rebrand of the existing platform. It is a parallel agent-commerce layer that sits above Adyen's core processing infrastructure and exposes a new API surface purpose-built for non-human transaction initiation. Three people with knowledge of the programme, speaking on condition of anonymity, described its components in enough detail to form a coherent picture. The first component is an agent-identity framework — a credential system that allows enterprise merchants to register authorised agents against specific transaction policies, including spend limits, vendor whitelists, approval hierarchies, and time-of-day restrictions. A registered agent receives a cryptographic identity token that travels with every transaction it initiates. Adyen's authorisation engine reads the token, validates the policy envelope, and processes or declines the transaction without requiring a human intervention step.

The second component is a reconciliation API that produces structured spending reports readable by both enterprise financial software and large language models. The design was shaped by direct feedback from Booking.com, which began piloting Nexus in late January 2026. Booking.com operates AI-assisted procurement agents across its supplier network — hotel contracting, regional marketing spend, infrastructure services. The agents initiate hundreds of thousands of transactions monthly. The reconciliation problem was not whether Adyen could process the payments. Adyen could always process the payments. The problem was that the downstream reporting, previously formatted for human accounts-payable teams, required significant reformatting before an LLM could meaningfully interpret it. Nexus's reconciliation API outputs a transaction summary in a structured format that Booking.com's financial AI can ingest directly, with no transformation layer in between. That product decision, unremarked in any press release, resolved a workflow bottleneck that Booking.com's finance team described internally as one of its top five operational irritants.

The third component is a banking-grade liability framework. This is where Adyen's institutional position becomes a structural advantage rather than a liability. Adyen holds a full banking licence issued by the Dutch central bank, De Nederlandsche Bank. No other major payments processor at its scale holds an equivalent. That licence gives Adyen the legal standing to offer indemnification guarantees on agent-initiated transactions under a framework that is already recognised by European financial regulators. Stripe's Atlas-Agent indemnification is contractual. Adyen's equivalent, built under its banking licence, will sit within a regulatory framework that European enterprise merchants — particularly those operating in financial services, healthcare, and public-sector procurement — regard as categorically more robust. Brouwer made this point explicitly in a closed briefing to twelve enterprise CFOs in Frankfurt in March 2026. Four of those CFOs had been considering migrating to Stripe. Two reversed that evaluation within 30 days.

The EU Regulatory Tailwind

The European Banking Authority's open consultation on PSD3 classification of AI-agent-initiated transactions closed on 30 June 2026. Adyen filed a 62-page response. Stripe filed 47 pages. The divergence in those submissions tracks the companies' respective strategic positions with precision. Stripe argued for a tiered authentication framework — high-frequency, low-value agent transactions should face lighter-touch verification requirements, allowing the agent-commerce layer to scale without friction. Adyen argued for a liability-tiered framework — the authentication requirement should track not the transaction size but the institutional standing of the certifying entity. Under Adyen's proposed framework, a transaction initiated by an agent registered with a licensed banking institution would receive expedited processing. A transaction initiated by an agent registered only with a non-bank processor would face the full PSD3 strong customer authentication requirement.

The proposal is self-interested, but self-interest and sound policy are not mutually exclusive. The EBA's preliminary working papers, circulated to registered stakeholders in April 2026, indicated that the authority found Adyen's liability-tiering argument "substantively compelling" and was modelling its implications. If the final PSD3 technical standards adopt something close to Adyen's proposed framework — an outcome that several Brussels-based regulatory analysts placed at roughly even odds as of May 2026 — Adyen's banking licence transforms from a background credential into a front-line competitive weapon. Every enterprise merchant in the EU market building an agent-commerce workflow would face a structural incentive to route through a licensed banking institution rather than a non-bank processor. The market Adyen would then be competing in would be one it designed.

The EU tailwind runs deeper than PSD3. The European AI Act's provisions on high-risk AI system oversight, which came into force in phased implementation from August 2024, require documented human oversight mechanisms for AI systems operating in financial contexts. Agent-initiated payments sit in a contested classification zone — some interpretations place them in the high-risk category; others do not. Adyen's legal team filed a position paper with the AI Act supervisory authority in February 2026 arguing that agent-attestation systems operated by licensed banking institutions should be treated as a recognised oversight mechanism under the Act's Article 14 provisions. The paper is under review. Its acceptance would give Adyen's Nexus platform a compliance certificate that no non-bank competitor could replicate without a fundamental change to its regulatory standing.

The Merchant Retention Equation

Adyen's enterprise merchant base represents the most concentrated test case for agent-commerce adoption in the European market. Uber processed $48.7B in gross bookings through Adyen in 2025, spanning 72 countries. The company's procurement AI — deployed internally in Q3 2025 — initiated 14 per cent of Uber's supplier payments by year end. That figure was rising at roughly two percentage points per month. Uber's treasury team, working with Adyen's enterprise solutions group, ran a 90-day pilot of early Nexus components beginning in February 2026. The pilot covered Uber's Netherlands-headquartered European supplier network, approximately 4,200 vendors. Agent-initiated payments in the pilot processed at Adyen's standard authorisation rate — 99.3 per cent — with zero attestation failures in 340,000 transactions. Uber extended the pilot to its full European supplier network in May 2026.

Microsoft's relationship with Adyen is structured differently. Microsoft uses Adyen primarily for its consumer-facing commerce — Surface hardware sales, Xbox store transactions, Microsoft 365 subscription renewals in markets where Microsoft's preferred regional processors don't operate. The agent-commerce case for Microsoft is not procurement but consumer-side: as Microsoft embeds Copilot agents deeper into its Windows and Office ecosystem, those agents will increasingly encounter checkout moments. A Copilot agent helping a user renew a software licence or purchase a hardware accessory needs to execute a payment. The question is whether it executes that payment through an infrastructure that recognises agent identity and applies appropriate authorisation policies, or through a generic checkout flow that treats the agent as a human user and applies none of the relevant controls. Adyen and Microsoft began technical discussions on Copilot-agent payment integration in November 2025. Those discussions had not produced a signed agreement as of the date of this analysis, but both companies confirmed the partnership was active.

The Booking.com pilot, already mentioned in the context of the reconciliation API, carried a data point that Adyen's enterprise sales team has used in every subsequent merchant briefing. In the six weeks before Nexus integration, Booking.com's agent-initiated payment failure rate — transactions that failed due to authentication errors, policy mismatches, or reconciliation format incompatibilities — ran at 3.8 per cent. In the six weeks after integration, it ran at 0.4 per cent. That 3.4-percentage-point improvement represents real money at Booking.com's transaction volumes. More importantly, it represents the argument Adyen's account managers bring to every enterprise renewal conversation: agent-commerce is not a new product category you evaluate separately. It is the next revision of payments infrastructure you already depend on. Adyen is building that revision. You should stay and see it.

What to Watch

Adyen's Nexus programme will not be finished in a quarter. The markers below will determine whether the institutional bet lands on schedule or gets overtaken by faster-moving competitors.

• Nexus general availability date. Adyen's internal roadmap targets Q4 2026 for general availability of the full Nexus API surface. A slip past Q1 2027 would open an 18-month window in which Stripe is the only processor with a production-grade agent-commerce stack serving enterprise accounts above $1B in annual payment volume. Watch the developer preview announcement — it will come 90 days before GA, and its scope will telegraph whether the build is on track.
• PSD3 technical standards publication. The EBA is expected to publish final technical standards in Q3 2026. If the liability-tiering argument Adyen advanced in its consultation response is reflected in the final text, Adyen's banking licence becomes a front-line competitive moat in the European market. If the standards adopt Stripe's tiered-authentication model instead, the regulatory tailwind disappears and Adyen competes on product and price alone.
• Acquisition activity. Adyen held $3.4B in cash and equivalents as of March 2026. The three agent-authentication startups most frequently cited in Adyen's internal competitive tracking — Attestiv, a Munich-based agent-identity platform; Kura Payments, a London-based B2B agent-invoicing network; and Formo, a Berlin-based reconciliation API company — have all been in conversations with strategic investors. A move before September 2026 would signal that Adyen has decided to accelerate Nexus through acquisition rather than building alone. The price at which it moves will indicate how much urgency the board has assigned the programme.
• The Microsoft Copilot integration outcome. If Adyen closes a signed agreement with Microsoft on Copilot-agent payment integration, it gains both the revenue and the reference architecture. Microsoft Copilot is the highest-volume AI agent deployment in enterprise software by daily active users. A Copilot-native payment rail built on Adyen would make Adyen the default infrastructure for a substantial share of enterprise agent-initiated transactions in Europe and the United States before Nexus reaches general availability.
• Stripe's European banking licence application. Stripe applied for a European banking licence in Ireland in Q2 2025. The application is under review by the Central Bank of Ireland. If it is approved — likely in late 2026 at the earliest — Stripe gains the regulatory standing that currently gives Adyen its EU liability-framework advantage. The window in which Adyen's banking licence is a unique competitive asset may be 18 months, not five years. The speed of Adyen's Nexus build matters accordingly.

Frequently Asked

What is Adyen's Nexus programme, and when will it ship?

Nexus is Adyen's agent-commerce infrastructure layer — a new API surface built above its existing processing engine that handles agent-identity credentials, transaction-policy enforcement, and LLM-readable reconciliation reporting. Adyen has not publicly announced the programme by name. Based on descriptions from three people with knowledge of the build, the company is targeting a developer preview in Q3 2026 and general availability in Q4 2026. Those dates are subject to the outcome of the PSD3 technical standards publication, which will affect the design of the liability framework component.

Why does Adyen's banking licence matter for agent payments?

Adyen holds a full banking licence from De Nederlandsche Bank, the Dutch central bank. That licence gives it the legal standing to offer indemnification on agent-initiated transactions under a regulatory framework already recognised across the EU, and to apply for compliance certification under the European AI Act's Article 14 human-oversight provisions. Stripe's equivalent indemnification is contractual — enforceable, but not backed by banking-sector regulatory recognition. For enterprise merchants in regulated industries — financial services, healthcare, public-sector procurement — that distinction is material at the point of vendor selection.

Is Adyen at risk of losing its major enterprise accounts to Stripe?

Not in the near term. Enterprise payment contracts at the scale of Uber, Booking.com, and Microsoft run two to four years and are renegotiated on a relationship basis, not on the strength of a single product announcement. Adyen's enterprise retention rate held at 97 per cent through Q1 2026. The risk is not churn of existing accounts. It is losing the marginal new account evaluation — the merchant designing a system for 2027 that weights agent-commerce readiness as a primary criterion. That is the cohort Adyen lost 75 per cent of in the H2 2025 competitive evaluations its internal analysis reviewed. Nexus is built to address that cohort specifically.

How does the EU regulatory environment affect Stripe's competitive position in Europe?

Stripe's Atlas-Agent API was designed to PSD3 compliance standards as Stripe's legal team understood them at the time of the November 2025 GA launch. If the EBA's final technical standards adopt a liability-tiering framework that privileges licensed banking institutions, Stripe's non-bank-processor status would subject its agent-initiated transactions to more stringent authentication requirements than Adyen's equivalent transactions. Stripe has filed a 47-page response to the EBA consultation arguing against this outcome, and it has an active banking-licence application in Ireland. Neither response guarantees a favourable outcome. European enterprise merchants building agent-commerce workflows in 2026 are monitoring the PSD3 process closely before committing architecture decisions.

Should enterprise merchants wait for Nexus before making a decision on agent-payments infrastructure?

For merchants already on Adyen, the calculus is straightforward: the Nexus developer preview is targeted for Q3 2026, and existing Adyen integration work transfers directly to the new API surface. Waiting costs little if the merchant's agent-commerce deployment is not scheduled before Q1 2027. For merchants on no platform, or actively evaluating both options, the decision should factor in the PSD3 outcome, the geography of the merchant's transaction volume, and the regulatory standing requirements of their industry. Stripe's Atlas-Agent is shipping now, with a mature SDK and 47 per cent new-merchant adoption. Adyen's equivalent is eight months away. That gap is real. So is the regulatory and institutional advantage Adyen brings to the EU market specifically.