712: Mildly Technologically Traumatized

The episode opens with Marco sharing his first ride in a Waymo driverless car. He describes the experience as initially odd but quickly becoming mundane, comparing it to a regular taxi and noting that he felt safer in a Waymo than in a human-driven car, at least in the specific operating area. Marco explains that the Waymo couldn't pick him up at the hotel's driveway, only at the street, highlighting the complex logistical issues surrounding autonomous vehicle deployment. He contrasts Waymo's safety and driving behavior favorably against Tesla, citing Waymo's excellent safety record and its ability to handle typical driving situations like busy left turns smoothly. Another host agrees that Tesla's safety, although generally better than human drivers per mile, is less trusted due to past incidents and company handling. Marco admits and others concur that for city roads in areas Waymo operates, the technology seems remarkably good and natural, making him more comfortable with Waymo than with human drivers. He emphasizes that Waymo is not available everywhere and isn't fully autonomous on all roads, but when it is, it performs very well.

Marco discusses his experience integrating AI into software development, emphasizing that AI excels at executing tasks, especially repetitive or tricky coding chores, but not at designing or planning features. The narrative asserts that experienced developers bring critical skills in asking the right questions and shaping AI output for quality products. Marco pushes back on criticism that AI code is useless, comparing it to abstract art that, while seemingly simple, involves nuanced craftsmanship. He argues that AI augments developer capabilities, enabling faster and more refined software delivery, and dismisses concerns about dependency or devaluation of skill. The conversation also covers security and permissions on macOS, notably the complications around full disk access, app permissions, and AI agents managing headless Macs. There’s skepticism about Apple’s security model, which relies on frequent user prompts and limited granularity, making autonomous management challenging. The hosts lament that Apple’s policies often hinder automation, forcing developers and advanced users to work around restrictions with hacks or side channels, and express hope that future improvements in privacy controls will better balance security with automation needs.

The hosts explore rumors about upcoming Apple products, particularly the lighter and possibly thinner 16-inch OLED MacBook Pro, expressing concern that increased portability might come at the expense of battery life and cooling capacity. They discuss that shrinking chips with newer manufacturing processes (like 2nm) might offset battery reductions, but uncertainties remain about thermal management and sustained power. The episode also briefly covers Apple's expanding smart home ambitions, noting a partnership with LG to develop branded smart home devices like doorbells, locks, and cameras, which will be under LG branding and support, not Apple. The hosts critique Apple’s history of outsourcing these products, which often results in mediocre quality, and speculate on how Apple might feature such products in upcoming presentations. They observe that Apple likely prioritizes privacy and ecosystem control but prefers to partner rather than build these devices themselves, leading to mixed quality and limited integration. Overall, there's skepticism about whether these third-party gadgets will meet user expectations or merely serve Apple's marketing imagery.

The episode takes a serious turn as the hosts analyze Apple's recent announcements regarding increased security controls, notably the introduction of additional permissions and prompts for full disk access — an especially critical concern as AI agents become more capable and autonomous. They discuss a recent Apple developer post emphasizing the need for explicit user actions before granting apps full disk access to mitigate risks of data exposure and privacy breaches. The hosts raise alarm that Apple’s policy of constant prompts might be counterproductive, leading users to disengage or blindly approve, thus undermining security. They explore the technical and security implications, such as unencrypted data at rest in databases (like Messages), and argue that better security might involve encrypting data at rest or more granular permissions rather than burdensome prompts. The conversation laments that Apple’s current security model hampers automation and autonomous AI management, making headless or AI-controlled Macs hard to operate reliably, and suggests that future policy changes could impact AI’s functionality but might also incrementally increase user safety.