Interview #102 Alex Salazar, Co-founder & CEO at Arcade.dev

Alex Salazar, Co-founder & CEO at Arcade.dev, breaks down why treating AI agents as teammates is an architectural dead end, and why they should be treated as software that is unpredictable enough to count as adversarial. He explains why security is usually the first team to say no, why an agent's own permissions are only half the problem when an intern could ask it for the CEO's pay, and what has to be true before a bank lets an agent wire money, from payee accounts registered ahead of time to checking the CRM so a refund is not hallucinated. Alex also dives into approval fatigue, research showing people check an agent's work less when they see it as a peer, and why no model should ever be trusted, open-weight or not. Finally, he shares why the vendors behind most agent strategies likely did not exist before 2024, and why agents moving money will feel mundane by the end of next year.

Free AI lessons for leaders, built on these interviews: https://tonyphoang.com/learn

Be a guest on the show: https://tonyphoang.com/guest

Live interviews at your conference: https://tonyphoang.com/live

Music and credits: https://tonyphoang.com/credits