001: What is Anthropic Doing?!

The hosts discuss Anthropic's recent decision to clarify and change their pricing model for Claude. They explain that interactive usage remains subscription-based while non-interactive, programmatic usage now incurs costs per token, supported by a credit system. This change effectively prices out tools that rely on programmatic access, like Cloud P CLI and the Agent SDK, which are no longer covered under the subscription plan. The hosts analyze the strategic reasons behind Anthropic's move, suggesting it aims to shift enterprise usage onto higher-paying plans and push users towards their desktop clients, which are more profitable for them. They also debate the fairness and potential backlash of these moves, with some viewing it as pure capitalism and others questioning the impact on smaller users.

Anthropic has temporarily increased weekly limits for their services by 50% until July 13th, viewed as a damage control measure and market maneuver. The hosts note the frequent use of such tactics by AI companies—offering limited-time perks or extra tokens—to manage user experience and PR. They discuss whether these are genuine incentives or strategic ploys to encourage enterprise adoption, emphasizing Anthropic's focus on steering large clients onto higher-tier plans, essentially a strategy to shed smaller users and grow enterprise revenue.

The conversation shifts to the heightened capabilities of AI models in cybersecurity, specifically mentioning Mythos and GPT 5.5 Cyber. Mythos, capable of executing complex network attacks rapidly, exemplifies how AI is transforming security threats. The hosts discuss the importance for developers and companies to proactively test and fortify their systems using AI tools, as malicious actors could leverage AI to exploit vulnerabilities at scale. They recommend rigorous dependency management, timely updates, and automated security testing as key defenses, acknowledging that AI-assisted attacks and defenses will be a persistent, evolving challenge.

Jared from Cloudflare introduces a significant technical achievement: rewriting an entire complex codebase from Zig to Rust using AI assistance, completed in about a week. The hosts praise the value of end-to-end testing to ensure functional continuity during such extensive rewrites. They explore how AI-powered testing, combined with a focus on preserving current behavior, can enable rapid, large-scale code conversions. The discussion underscores the potential for AI to revolutionize software maintenance and refactoring, especially for long-standing projects.