Key Takeaways
- At TravelTech Show 2026, Trainline CTO Mike Hyde and Anthropic enterprise GTM lead Harry Herbert discussed the "fundamentals" companies need as they scale agentic AI
- Hyde's core idea is a mental model: 600 engineers suddenly become 3,000, and the extra 2,400 are less trustworthy. Process, design, governance, security and financial control change more slowly than models do--which is exactly why they are the foundation that lets you keep adopting the latest model
- For travel and booking businesses, the session offers a concrete reference for moving beyond the 10-20% productivity gains of tool rollouts toward automating entire tasks and jobs, with security design, token budgets and evaluation processes in place
The "Fundamentals" Discussed at TravelTech Show 2026
Trainline has partnered with Anthropic to implement AI internally and in customer-facing applications. At the recent TravelTech Show, Trainline CTO Mike Hyde shared his experience so far.
www.phocuswire.comAt a TravelTech Show 2026 session in London, Mike Hyde, CTO of Trainline--one of Europe's largest rail and coach booking platforms--and Harry Herbert, enterprise go-to-market lead at Anthropic, discussed what travel companies need as they scale agentic AI. Trainline has partnered with Anthropic and is implementing AI both in internal operations and in customer-facing applications, which makes this a practitioner's account rather than a vendor pitch.
The keyword running through the session was fundamentals: security, data and financial control. According to Hyde, these unglamorous elements are the first things put to the test when AI scales. Hyde joined Trainline in 2021 after leading data organizations at Meta, Skype and Microsoft. In April 2025 he was promoted from Chief Data Officer to CTO, overseeing both the technology and data teams with a mandate to embed AI and machine learning into the core of the platform. His remarks carry weight precisely because they come from someone 18 months into the journey, not someone about to start it.
The Mental Model: 3,000 Engineers You Cannot Trust
Hyde opened with a thought experiment. Trainline currently has roughly 600 engineers. Now imagine the workforce effectively grows to 3,000 through agentic AI--with most of the additions being AI agents.
The difference is we suddenly have 3,000 engineers, and all of our systems are designed to support 600 engineers, and the 3,000 engineers are less trustworthy than the 600. You can't assume they're going to do the right thing all day, every day. Then, what kind of a security system would you need to run a 3,000+ company where you don't trust all the people?
Source: Mike Hyde (Trainline CTO)
Hyde called this the "mental model" Trainline is trying to figure out, and suggested it is a good starting point for many businesses: "What would you do if you were 5x as big as you are, with a whole series of very low-trust employees?"
The shift in perspective matters. Conversations about agent adoption tend to drift toward which model is smartest or which tool is most convenient. Hyde reframes the problem as one of organizational security design. For human engineers, companies have spent decades building layered controls--hiring screens, access management, code review, audits. When thousands of agent "new hires" join at once, the same level of control has to be rebuilt in a form suited to non-human workers. Keeping systems designed for 600 people while quintupling the workforce is an obvious recipe for failure.
Agents also mix moments of brilliance with moments of unreliability. That is why designs that do not assume correct behavior--something close to zero trust--are likely to become the standard for managing an agentic workforce. That is the deeper implication of the model.
Five Fundamentals and Token Budgets as a New Financial Discipline
So what are the fundamentals, concretely? Hyde listed five: process, design, governance, security and financial control. Focusing on them also helps companies navigate a rapidly evolving landscape. As Hyde put it, these things are not seen as sexy or headline-grabbing, but "they're the foundation that lets you adopt whatever the latest model is. The fundamentals don't change as fast as the models do."
Models turn over in cycles of months. Investments in optimizing for a specific model depreciate with the next generation, while investments in governance and security design carry over to whatever model comes next. It is the classic principle of investing in the slow-changing layer rather than the fast-changing one, applied to the agent era.
The most concrete of the five was financial control--managing AI cost and tokens. Hyde noted that the industry still needs to figure out the cost of compute going forward, and shared how Trainline handles it today: the company is starting to use a system of individual token budgets for engineers, plus special budgets for projects known to be token-expensive. "But it might be worth it because we're building something new."
Just as cloud cost management matured into the discipline of FinOps, agent token consumption calls for the same rigor. Trainline's two-tier setup--personal budgets plus exceptional project budgets--is one of the earliest working examples worth studying.
From Tool Rollouts to Top-Down Automation: Trainline's 18 Months
Trainline has been implementing internal AI capabilities for about 18 months, but according to Hyde the company hit an inflection point in the past three to four months. The initial approach--roll out tools and see what happens--is where most companies start their AI journey. That phase delivered productivity gains of 10% to 20%.
The turning point came after that. "We've come to the point of view that it's not enough to just roll AI out; you get to a point where you step back and realize we need to rethink things more fundamentally, like how will we work in the future," Hyde said. Trainline has now switched to a top-down view, building automations that do not merely save someone an hour or two but take on an entire task or job using a series of agentic steps.
| Phase 1 (until 3-4 months ago) | Phase 2 (after the inflection point) | |
|---|---|---|
| Approach | Bottom-up: roll out tools and see what happens | Top-down automation driven by leadership |
| Unit of automation | Saving individuals an hour or two | Replacing entire tasks and jobs with chains of agentic steps |
| Outcome | 10-20% productivity gains | Rethinking how the company will work (in progress) |
| Guiding question | 'Which tools should we use?' | 'How will we work in the future?' |
Hyde identified three areas where AI's impact shows up: building more intelligent customer products on top of the company's data, understanding and adapting to changes in how people search for and discover travel, and transforming internal operations. On the customer side, Trainline has already rolled out the agentic Trainline Assistant in its UK app, which autonomously handles actions such as processing refunds. "The assistant is learning all the time and we're continuing to scale up the agentic AI system which supports it," Hyde has said of it.
What Anthropic Sees in Successful Adopters
Herbert added the vendor-side view of where customers stand. Eight months ago, the typical picture was companies experimenting with scattered tools while employees used personal ChatGPT licenses at work. That has shifted: organizations now ask how AI needs to be built into workflows and which tools are actually best suited to deliver ROI.
His summary of what successful companies share is crisp: "The businesses that are most successful have top-down vision and mandate, and then from the bottom-up is where you find the use cases. So there needs to be leadership around how and why you're adopting it, while from the bottom up, lots of use cases are being surfaced." Not one or the other--a combination of top-down intent and bottom-up discovery. It maps closely onto the arc Trainline itself has followed.
Herbert also stressed putting processes and evaluation practices (evals) in place to judge whether a new feature is actually relevant to the use cases and customers being served. Without a systematic way to evaluate model output, there is no yardstick for judging whether adoption is working at all. AI sophistication varies widely among travel clients--some are pushing the boundaries of the technology while others are still figuring out how to get started. Even so, "the 'Eureka' moment for a lot of clients is when they see AI working with their own data and their own systems," Herbert said. It is the demonstration in one's own context, not a generic demo, that moves organizations.
Implications for Travel and Booking Businesses
At the same TravelTech Show, Travelport CTO Andrew Jordan pointed out that as travel distribution shifts from deterministic to probabilistic, there is a firm line between the realm of discovery and inspiration and the realm where money changes hands--and on the far side of that line, "you can't get it almost right." Travel and booking businesses handle reservations and payments, where an agent's mistake becomes a financial incident. Hyde's low-trust workforce model and Jordan's precision requirement illuminate the same problem from the internal and customer-facing sides.
Three practical implications follow. First, treat agent adoption as a problem of organizational security and governance design, not tool selection. The real work is rebuilding permissions, audit and controls on the assumption of a large, low-trust new workforce. Second, introduce the discipline of managing token cost as budget early. Trainline's two-tier structure of individual budgets and special project budgets can be copied tomorrow. Third, do not settle for 10-20% productivity gains; prepare evaluation processes with task- and job-level automation in view. Trainline's experience suggests the inflection point arrives only after a running start with broadly distributed tools.
Closing Thoughts
Trainline's 18 months show the agentic AI conversation moving from "what can it do" to "how do you run it." The question of 3,000 untrustworthy engineers is one that every company deploying agents at scale will eventually face, in travel or elsewhere. Models will keep changing. What does not change is that process, design, governance, security and financial control are the foundation everything else rests on.
