Article
Agentic foundations: Addressing the onboarding bottleneck
Why point-to-point connections fail, and what comes next
In our previous posts, we explored the inventory gap, the trust paradox, and why legacy enterprise systems were not built for machine-driven commerce. Each of these constraints is difficult enough to solve on its own.
The challenge becomes significantly harder when businesses need to solve all three simultaneously inside a real enterprise, while launching on an entirely new commerce surface. Let's assume a business has made progress on all three fronts: their product data is accessible, trust frameworks are taking shape, and infrastructure can support autonomous interactions. Even with this critical foundation in place, they still have a problem. The bottleneck now sits squarely at the onboarding layer.
Why point-to-point does not scale
Every integration requires reconciling data readiness, protocol compatibility, risk frameworks, and backend exposure at the same time — and across teams, systems, and timelines that were never designed to move together. It becomes a multi-month, bespoke engineering project for every merchant and platform.
The friction exists on both sides. For merchants, participation today effectively requires a dedicated engineering team willing to work through custom integrations, waiting lists, and largely undocumented processes. The merchants who have made it through are almost exclusively large enterprises with the resources to absorb that friction.
For agent platforms and developers, every new merchant is a fresh project involving different order management systems (OMS), tax engines, product information management (PIM) systems, protocol configurations, and risk postures. Platform teams that should be focused on AI logic and user experience are instead managing a sprawling web of one-off connections.
The instinctive response to this kind of fragmentation is to build more connectors, such as software development kits (SDKs) for specific platforms, automated wizards, and pre-built integrations for the most common commerce stacks. The problem with this is scale. With at least 10 major agent platforms and thousands of enterprise merchants — each running their own combination of legacy PIM, OMS, tax, and fulfillment systems — maintaining point-to-point connections requires an engineering investment that grows with every new participant. Every time a merchant updates a tax engine or a platform changes its query structure, something breaks. This introduces systemic fragility rather than a sustainable resolution.
What the solution looks like
Historically, the only definitive solution to this bottleneck is an abstraction layer: neutral architecture that unifies both sides and eliminates the need for direct, complex integrations.
For agentic commerce in practice, it means three things
Protocol abstraction: a neutral layer that translates between merchant infrastructure and agent platform requirements, so merchants integrate once rather than once per platform, and agent developers do not have to account for thousands of unique backend configurations.
Portable risk and identity: trusted identity, payment tokens, and risk profiles that are portable across the agentic ecosystem from day one, so adding a new agent surface does not mean rewriting fraud rules and re-establishing liability frameworks from scratch.
Single-point distribution: a merchant connects their product data, pricing, inventory, and checkout logic once, and that connection makes them reachable across every surface that builds on the abstraction layer.
The bottom line
The merchants who have successfully integrated into agentic commerce so far are not the ones with the best products or the most sophisticated technology. They are the ones with the most engineering capacity to absorb repeated integration friction.
This creates an uneven playing field, favoring engineering muscle over product quality in an ecosystem that should benefit everyone. The abstraction layer model does not eliminate the complexity of agentic commerce — it relocates it from individual merchant stacks and to shared infrastructure. That shift is what turns a collection of isolated pilots into something that can realistically reach real scale.
Agentic commerce still has a product market fit question to answer. Getting there requires, at minimum, that onboarding stops being a test of how much engineering friction you can absorb.
To learn more about building a scalable foundation for autonomous shopping, visit Adyen’s Agentic Commerce hub.