Carriers across Africa and the Middle East have spent the better part of two years racing to catch up on NDC: adopting the standard, retiring EDIFACT dependencies, rebuilding agency relationships around modern retailing. That race is not over. But somewhere in the last few months, the finish line moved. A second transition has arrived on top of the first one, and it is moving faster than NDC ever did.
Agentic AI, systems that discover, compare, and increasingly complete travel purchases with minimal human input, has gone from conference talking point to production infrastructure in a matter of months. For carriers still building out their NDC distribution capability, the practical question is no longer just “when do we finish modernizing our retailing stack.” It is “how do we modernize a stack that is itself becoming a moving target.”
What actually shipped
The protocol landscape underneath agentic travel commerce came together in stages. Anthropic open sourced the Model Context Protocol in November 2024, giving AI agents a standard way to query tools and data, and later donated it to a newly formed Agentic AI Foundation at the Linux Foundation. Google followed with Agent2Agent, a protocol for one agent delegating tasks to another, which reached production deployment across more than 150 organizations by April 2026 according to the Linux Foundation’s own release notes. On the commerce side, Google introduced the Universal Commerce Protocol at NRF 2026, built with retail partners including Shopify, Target, and Walmart, while OpenAI and Stripe shipped a competing Agentic Commerce Protocol behind ChatGPT’s instant checkout feature.
Payments authorization has followed a similar multi-vendor pattern. Google’s Agent Payments Protocol, launched in September 2025 with more than sixty partners including Mastercard, American Express, PayPal, and Adyen, uses cryptographically signed “mandates” that specify what an agent is authorized to buy, at what price, and under what conditions. Visa built a parallel Trusted Agent Protocol that verifies agent identity through signed HTTP headers, and introduced an Agentic Directory and Agent Score tool to help merchants assess whether their own sites are even navigable by AI agents in the first place. Mastercard’s answer, Agent Pay, introduced its own token and intent-verification primitives, and by April 2026 the two frameworks had reached a degree of interoperability, with Mastercard’s network accepting AP2-issued mandates as valid intent artifacts. In late June, Mastercard extended the approach further with Agent Pay for Machines, aimed specifically at machine-speed microtransactions down to fractions of a cent, while Visa used its Paris Payments Forum to announce a strategic partnership with OpenAI alongside new trust and stablecoin settlement tools.
Individually, these read as a string of separate product launches. Collectively, they are the emergence of a common operating system for machine-to-machine commerce, and travel is one of the first industries where it is being tested at scale.
In travel specifically, the clearest production example is the partnership between Sabre, PayPal, and the AI trip-planning platform MindTrip, which launched publicly on May 6, 2026. It links Sabre’s Mosaic content layer, covering more than 420 airlines and two million hotel properties, to a conversational interface and PayPal’s agentic payment infrastructure, allowing a traveler to move from stated intent to a completed, paid booking inside a single chat exchange with no redirect to an airline or OTA website at all.
The discovery layer problem, measured
How ready is the airline side of this equation right now? Bain and Company ran a structured test in March 2026 that is worth taking seriously precisely because it is one of the few pieces of empirical data in a conversation otherwise dominated by vendor announcements. Bain prompted three major large language models with sixty non-branded queries covering ten major routes for three large European full-service carriers, then separately attempted to complete bookings using OTA plugins, browser automation agents, and airline links. In the discovery test, airline websites were the direct source of information only about 5% of the time. LLMs overwhelmingly surfaced OTA content instead, because OTA data structures are simpler for models to parse and act on. In the booking test, no single method reliably completed a purchase; browser agents repeatedly stalled on date pickers, fare selectors, two-factor authentication, and CAPTCHAs designed to stop bots, not welcome them.
The finding matters beyond the three carriers Bain tested. It describes a structural condition of most airline booking infrastructure as it exists today, built for human navigation and defended against automated access, at the exact moment automated access is becoming the primary discovery channel for a growing share of travelers.
Why this compounds for Second-Speed Markets
The carriers most exposed to this shift are, almost by definition, the ones that have spent the last eighteen months focused on a different and equally demanding transition: standing up NDC capability, weaning off EDIFACT-dependent agency relationships, and rebuilding commercial infrastructure around modern retailing standards. That work does not pause because a second wave has arrived. It simply means the distance between where these carriers are and where the leading edge of the industry is operating has widened again, on a different axis, before the first gap closed.
A Second-Speed Market is not defined by being permanently behind. It is defined by catching one wave of infrastructure modernization just as the next wave begins, so the challenge becomes cumulative rather than sequential. Under that definition, the agentic shift is not a hypothetical future problem for these carriers. It is the second wave arriving on schedule. Sabre, Amadeus, and the card networks are not waiting for emerging markets to finish NDC before building agent-native commerce layers on top of the carriers that already have.
Agencies face a version of the same problem, on a shorter timeline. Many across these markets have only recently invested in NDC aggregation and modern servicing workflows of their own. If travelers increasingly start their shopping journey through an AI agent rather than a search engine or an OTA interface, agencies will need their own inventory, servicing, and payment infrastructure to be just as legible to machines as it is to human agents. The competitive question shifts from who has the best website to whose content an AI agent can discover, interpret, and transact against without friction.
The counter-narrative on payments
There is a genuine twist in this story, and it runs against the usual assumption that African and Middle Eastern markets trail Western infrastructure by default. On the payments and settlement side specifically, the evidence points the other way.
The IMF published a formal note in April 2026 examining how agentic AI payments could reshape African financial infrastructure, arguing that the technology’s probabilistic decision-making sits in tension with the deterministic finality payment systems have always required, and calling for African central banks and payment networks to build know-your-agent governance ahead of volume rather than after. Visa’s own regional disclosure, shared at its Paris Payments Forum at the end of June, showed tokenized transactions across the Central and Eastern Europe, Middle East, and Africa region climbing from 26 percent in 2023 to 70 percent in 2026, with stablecoin settlement volumes in the region up nearly sixty times over the past year alone.
More strikingly, some of the infrastructure being built to serve agent-to-agent microtransactions, the one-to-ten-cent payments that traditional card rails cannot process profitably, is emerging first in African markets rather than being retrofitted onto them. Stablecoin payment infrastructure providers operating across the continent have pointed to existing mobile money adoption and stablecoin familiarity as a head start for a settlement layer built around machine-speed, per-call payments rather than human checkout flows. Whether that head start translates into a durable advantage, or simply into faster settlement rails serving carriers that are still catching up on the retailing side, is the open question worth returning to as the agentic booking layer matures over the next year.
The first distribution race was about replacing legacy technology with modern retailing standards. The second is about making sure those modern systems can be seen, trusted, and transacted against by an AI agent that never visits a website in the traditional sense. Emerging market carriers and agencies are now running both races at once. Finishing NDC first will matter less than starting the second race early, and on payments, at least, some of these markets may not be starting from behind.
