From Teddy Bears to Voice Agents: Voice AI, Local-First & App Security

Hansen describes local-first software as responsive, usable offline, collaborative across devices, and built to give people control over their data. He says local data also helps developers: synchronous reads and writes avoid much of the complexity of optimistic updates. On-device language models make local access relevant to AI as well.

Hansen says ElevenLabs works on voice AI research, creative tools, and a platform for building agents that can use tools and access business data. A teddy bear project for his children led him to try an agent that could talk and play 20 Questions. He also describes family code words as a safeguard against voice impersonation, alongside company measures such as voice watermarking and consent steps for cloning.

Ola recounts inspecting a banking app to access his transaction data, using a man-in-the-middle proxy and APKTool after encountering certificate pinning. The effort exposed obfuscated JavaScript in the login flow, but he stopped after a weekend. Hansen also recounts finding an unsanitized address field on a Danish government site that allowed him to load JavaScript from his own domain; the site was fixed within weeks.

Hansen describes limiting weekday screen use and setting a shared family screen time of about an hour to an hour and a half on Fridays. He treats children’s disappointment at a boundary as sadness to acknowledge, not a reason to reverse the decision. He also explains Stoicism as an ongoing practice, shaped in part by recognizing that solving one problem often brings others.