The Financial Times ran a piece this year that began, as these things do, with a ruined holiday. A columnist got stranded in New Zealand when their flight home to London, routed through Abu Dhabi on Etihad, was cancelled during the US-Israeli strikes on Iran. It was, by any measure, a bad week for air travel. It also turned into a sharp argument about agentic AI, and my Conscium colleague Ted Lappas, our chief technology officer, supplied the closing line argument.
The booking system, tested by crisis
The columnist's account of the aftermath will be familiar to anyone who has tried to rebook a flight in a hurry: comparison sites that buffered for hours before timing out, prices that surged the moment demand spiked, an eventual retreat to booking directly with an airline rather than trust the middlemen. It's a mundane kind of chaos, but a useful one, because it's exactly the scenario agentic AI keeps being pitched to solve. A personal agent that searches, price-checks, applies for visas and books on your behalf, with your sign-off at each step, is a reasonable answer to a genuinely bad afternoon spent refreshing a loading screen.
Why the big three are already circling
The piece noted, correctly, that OpenAI, Anthropic and Google have all shipped commercial agentic protocols this year that could sit underneath exactly that kind of service, and that Google has already marked out travel as a priority. Its travel product lead told an industry conference the company was building end-to-end agentic travel services with partners, promising capabilities users haven't had before, while conceding that trust and transparency have to come with them. An agent that books your travel needs your location, your bank details, your passwords and your inbox to be any use at all. That's not a small ask.
The conflict of interest nobody denies
The more useful part of the argument was the question underneath it: would a Big Tech agent actually work purely for you, or would it quietly steer you toward the company's own commercial partners? The columnist's experience applying for a Canadian travel authorisation made the case better than any hypothetical could. A sponsored search result offered to handle the paperwork for roughly fourteen times what the official government site charged for the identical service, ranked further down the same page. That's not a hypothetical risk. It's the current state of search, and there's no obvious reason an agentic layer built on top of it behaves any differently unless someone deliberately builds it to.
Ted's answer: don't let the platform own the twin
Ted's contribution was the line I'd want anyone reading this to take away. He argues for a future in which each of us runs our own digital twin, owning our own data, built to work in our interest rather than a platform's. He put it plainly: he wants a future where the consumer builds the twin, not one where they rent it from whichever tech company happens to control the booking layer. That's consistent with the wider case Conscium makes about AI agents in general. The safest version of this technology is the one whose incentives sit with the person it's supposedly working for, and that alignment gets considerably harder to guarantee once the agent doubles as a shop window for its maker's other products.
Read the full piece on the Financial Times.