Spotnana, the New York travel infrastructure company founded by Sarosh Waghmar, released a multi-agent AI architecture on July 20 that automates routine work for corporate travel agents, including refunds, airline schedule changes and the import of bookings made outside its platform.
The architecture Spotnana announced Monday assigns discrete jobs to specialized agents, then maps those agents to Spotnana APIs that execute defined workflows. An orchestration agent routes requests and sequences the work. Spotnana says users must confirm actions involving financial transactions, while cases requiring human judgment are transferred to a live agent with a record of the automated steps already taken.
That structure reflects the bet Waghmar has pursued since founding Spotnana in 2020: rebuild the transaction layer beneath travel before trying to reinvent the interface above it. The arrival of generative AI has given that long infrastructure project a new route into the day-to-day work of travel agencies.
Waghmar's long route to the action layer
Waghmar's interest in travel technology began after his first startup collapsed during the dot-com crash. According to a company account of Spotnana's origins, he started a Yahoo Group called Spotnana to share travel deals and rewards strategies. That project developed into a luxury travel agency and then WTMC, a corporate travel management business.
By 2019, Waghmar had concluded that WTMC could not scale on the industry's existing systems. In an interview with investor Madrona, he described the underlying technology as broken plumbing that had survived beneath successive generations of booking interfaces. Spotnana became his attempt to replace it with cloud and API-based software that travel management companies and other sellers could use under their own brands.
Waghmar now serves as founder, co-chairman and chief product officer. Steve Singh, who built Concur and led it through its $8.3 billion sale to SAP, is Spotnana's executive chairman and CEO. Spotnana reports more than $115 million in funding from investors including Durable Capital Partners, Madrona, ICONIQ Growth, Mubadala Capital, Lufthansa Group and Amex Ventures. Durable led Spotnana's $75 million Series B in July 2022.
The new AI system sits on the infrastructure built with that capital. Spotnana has spent years converting travel servicing actions into API-accessible operations, covering content from airlines and other providers. The architecture gives AI agents permission to call those operations within defined boundaries instead of asking a general-purpose model to generate the transaction itself.
AI agents with transaction boundaries
In Spotnana's technical description of the architecture, a user request enters through a booking tool or another channel, an orchestration agent breaks it into tasks, and specialized agents call the relevant APIs.
Spotnana says a model cannot generate its own refund amount, fare rule or corporate travel policy. Those values come from structured records and governed workflows already maintained by the platform. The distinction matters when an automated system can move money, alter a ticket or apply a travel credit.
The initial production release includes four groups of capabilities:
- A servicing agent handles airline segments canceled after schedule changes, checks unticketed segments, issues eligible residual miscellaneous charge orders and processes eligible refunds.
- An external-booking agent imports trips booked outside Spotnana, bringing those records into itineraries, servicing tools and reports.
- A travel-agent copilot suggests replies and summarizes conversations or call transcripts.
- AI-assisted reporting provides analytics on completed servicing tasks, including the volume of work completed by the AI servicing agent, plus AI-generated trend analysis and suggested actions.
Spotnana says the servicing agent handles high-volume, routine workflows. When the workflow reaches a situation that requires judgment, the agent hands the case to a person. The travel-agent product post says the handoff includes the action history, which should reduce the need for an agent to reconstruct what happened before taking over.
Direct Travel is the first visible proving ground
Direct Travel says it is using the new capabilities to automate routine servicing while leaving complex disruptions and personalized advice to human travel advisors. Direct Travel already uses Spotnana as the infrastructure for Avenir, its travel platform.
The relationship is commercially meaningful and closely aligned. Singh is also executive chairman of Direct Travel, placing Spotnana's first named AI customer inside the same leadership orbit. Direct Travel nevertheless gives Spotnana access to live servicing operations at a travel management company already using its infrastructure.
A Spotnana case study says Direct Travel launched more than 30 customers on Avenir during the first two months after its May 2025 debut, with some deployments completed in fewer than 10 business days. That figure covers Avenir adoption, not use of the newly announced AI agents.
Spotnana and Direct Travel have not disclosed how many agents or corporate customers are using the AI functions. They also have not published accuracy rates, failure rates, changes in average handling time or cost savings from the deployment. Without those figures, the claimed operational gains remain unquantified.
An infrastructure product that partners can extend
Spotnana is allowing customers and partners to build their own agents and connect outside AI systems to its APIs. That approach fits Waghmar's original distribution model: Spotnana supplies the transaction infrastructure while travel sellers control the brand, workflow and customer relationship.
It also raises questions that will become more important as partners build on the architecture. Spotnana has not identified the models powering its agents or detailed whether customer conversations and travel records are used for model training. The announcement describes confirmation requirements for financial transactions, though it does not specify broader rollback, audit or liability procedures for incorrectly executed tasks.
Additional tools for travelers, travel managers and travel agents are expected later in 2026, with no specific release dates. The next stage of Waghmar's infrastructure thesis will be measured by whether outside partners can safely build agents on Spotnana and whether the early servicing deployments produce a public operating record. The architecture gives those agents somewhere controlled to act. Spotnana still has to show how reliably they perform at scale.