AI Exchange - SkyScanner, AWS, Snyk

Building on the success of January’s inaugural event, the AI Exchange returned on Wednesday for its second event, drawing yet another full house at the AWS HQ in Shoreditch. 

We listened to talks from 3 household names who are who are driving innovation in the world of Artificial Intelligence; SkyScanner, Snyk and our hosts at AWS.

Marcelo Vivela, Principal Data Product Manager at Skyscanner, explored the challenges and opportunities of leveraging user data in generative applications.

He shared his journey from startup founder to AI advocate, highlighting that engagement hinges on superior user experience, not just innovative technology. "The technology alone won't drive retention; we need to think, we need to design the experience. User experience is what keeps people coming back," he said, underscoring the critical role of thoughtful design in sustaining user engagement.



Next up, we welcomed to the stage Andrea Bureca, who leads the AI/ML team at AWS UK and Ireland. Andrea explored the Gen Paradox in AI adoption, where companies eager to implement AI get stuck in endless proof of concept cycles, with 95% never reaching production and up to 80% failing.

Andrea identified key barriers, including weak business cases, poor data quality, talent shortages, and difficulty estimating ROI. He stressed that "AI is a mean to an end. Start with a problem". Rather than chasing AI for its own sake, businesses must focus on real challenges AI can solve.

By building solid hypotheses, leveraging quality data, and adopting user centered design, organisations can break free from “PoC purgatory” and turn AI pilots into scalable, value driven solutions.



Finally, Joe Bollen, Staff Product Security Engineer at Snyk joined us to explore the intersection of AI security and social engineering, highlighting how LLMs may exhibit vulnerabilities similar to human psychology. Unlike traditional software flaws, which follow predictable patterns, LLMs’ non deterministic nature makes them susceptible to manipulation in unexpected ways.

Through experiments, Joe demonstrated how tactics like authority, urgency, and emotional manipulation could influence AI behavior, particularly when combined with anomalous prompts. While LLMs resist direct emotional appeals, they can still be coerced into producing unpredictable outputs under the right conditions.

He concluded that "the key to exploiting AI isn't just about technical vulnerabilities; it's about layering social engineering tactics on top of those inherent weaknesses." Understanding these risks requires a versatile approach, merging cybersecurity with behavioral science to secure AI defenses.



As always, a big thank you to everyone involved in running this event, to our hosts AWS for keeping us all watered and fed throughout the evening and to everyone who attended.

Please keep an eye on our Meetup page for event announcements.

If you are interested in speaking, sponsoring or hosting a future event with us, please submit your details here.

We look forward to seeing you all soon!