Episode Summary
Executive Summary: This Vergecast centers on how AI is changing software creation, privacy, and everyday computing. Boris Cherney explains Claude Code’s evolution from a dev tool to a broadly useful automation layer, while Hayden Field warns that connecting AI tools to personal data creates real, shifting privacy and security risks. The episode closes with practical buying advice for people worried about rising hardware costs.
Main Topics: Claude Code’s evolution and product-market fit (Priority: 5/5): Boris Cherney describes how Claude Code shifted from writing a small share of his code to doing nearly all of it, driven by model improvements and a sudden jump in reliability around Opus 4.5. He frames the product as both a coding tool and a broader automation system. Vibe coding and the changing role of developers (Priority: 5/5): The conversation explores how prompting models to write code has become a real workflow, with developers moving from manual editing to supervising, testing, and directing AI-generated work rather than producing every line themselves. Non-developers adopting coding tools (Priority: 4/5): Boris explains that people outside engineering—data scientists, sales teams, product managers, and others—are using Claude Code and related tools for tasks like analysis, workflow automation, and personal organization, prompting new form factors like Cowork. AI privacy, security, and data-sharing trade-offs (Priority: 5/5): Hayden Field breaks down how users should think about giving AI tools access to Gmail, calendars, files, and computers. Her core message is that the risk is real, policies can change, and users should understand exactly what data they are trading for convenience. Consumer AI versus enterprise safeguards (Priority: 4/5): The discussion contrasts consumer AI products with enterprise offerings, noting that enterprise use cases generally provide clearer controls and stronger boundaries, while consumer products often have more ambiguous terms and higher uncertainty. Phone upgrade timing amid hardware cost pressure (Priority: 3/5): A hotline question about upgrading a 15 Pro Max becomes a broader discussion of whether rising RAM costs, tariffs, and supply pressures should push users to upgrade earlier. The advice: only accelerate if you were already near upgrade time.
Key Arguments: AI coding tools have crossed from novelty to durable utility; Claude Code is now good enough that Boris no longer needs to manually write code for many tasks. The biggest leap was not just capability but autonomy: the model can now test, inspect, and fix its own work, reducing the need for human hand-editing. Non-engineers are finding value in coding agents for everyday automation, which suggests the market is much broader than terminal-first developer tooling. Cowork exists because users were already forcing a developer product into non-developer jobs; Anthropic responded by making a safer, more approachable interface. AI tools should be treated like any other data-hungry service, but with more caution because policies are newer, less regulated, and more likely to change unexpectedly. People should double-check important outputs like taxes, financial data, or health information, even if the tool is very capable. Using AI inside an ecosystem that already holds your data, like Google/Gemini, may be a simpler privacy trade-off than spreading sensitive data across new vendors. Hardware buyers should not panic-buy unless they were already close to replacing their device; if you need a new phone now, buy it, but a good existing phone is still a good bet for a couple more years.
Data Points: Claude Code timeline: 1 year old - Boris says Claude Code came out a year ago and the episode revisits its impact. Boris’s code-writing share: 100% AI-written - He says Claude Code went from writing about 10% of his code at launch to 100% after Opus 4.5. Claude Code early share: ~10% - Approximate amount of Boris’s code Claude Code wrote when it first launched. Claude Code mid-2025 share: ~30% - Boris says model improvements around Sonnet 4 and Opus 4 raised AI-written code to around 30%. Claude Code later share: ~50% before jump - He describes a progression to roughly half of his code before the final jump to full AI generation. Cloud Code impact on global code: ~4% of all commits - Boris cites a study claiming Claude Code writes around 4% of commits globally, noting the number is likely higher now. Cowork usage pattern: ~1/3 of his code on iOS app - Boris says he does about a third of his coding on the iOS app now. Workforce adoption: Half of sales team weekly - He says roughly half of Anthropic’s sales team uses Claude Code every week. Product growth pattern: Exponential - Boris says Cowork has been a hit immediately and has grown exponentially since launch. Supported companies: Spotify, Shopify, Ramp, Netflix, Novartis, Nvidia, Snowflake, Salesforce - Boris lists major companies using Claude Code. Security boundary: Virtual machine included - Cowork includes a VM so it cannot access more of the computer than the user allows. Prompt injection defense: Opus 4.6 most aligned yet - Boris says this model is the best Anthropic has built for prompt injection resistance. Phone example: 15 Pro Max - The hotline listener Lucas says he currently uses an iPhone 15 Pro Max. Other phone example: Pixel 7A - A Verge staff example of someone considering an upgrade because of RAM concerns. Attachment scale: 13,000 unread emails - David Pierce cites his own email backlog as a reason he wants AI help.
Pivotal Quotes: "Claude code does 100% of my coding." — Boris Cherney: Boris describes how far his own workflow has shifted after model improvements. "The UI of the future has not been discovered yet." — Boris Cherney: He explains that Anthropic is still experimenting with how AI tools should be presented to users. "You should assume that anything you send to someone or create digitally will eventually be public." — Hayden Field: Hayden uses this as a practical privacy framework for deciding what data to share with AI tools.
Implications: AI coding is now a real workflow, but trust, security, and privacy are still unsettled. Users should adopt these tools carefully, understand the trade-offs, and avoid over-committing sensitive data. For hardware buyers, the safest move is to upgrade based on need, not panic.
About The Vergecast
The Vergecast is the flagship podcast from The Verge about small gadgets, Big Tech, and everything in between. Every Friday, hosts Nilay Patel and David Pierce hang out and make sense of the week’s most important technology news. And every Tuesday, David leads a selection of The Verge’s expert staffers in an exploration of how gadgets and software affect our lives – and which ones you should bring into yours.