The Cloud Pod | Weekly AI & Cloud News on AWS, Azure & GCP · Justin Brodley, Jonathan Baker, Ryan Lucas and Matt Kohn | Cloud Computing & AI News

324: Clippy’s Revenge: The AI Assistant That Actually Works - Sort Of

·1 hr 4 min·4 clips

Claude Sonnet 4.5 scores 77.2 on SWE-bench Verified and can keep focus for over 30 hours.

1. The Cloud Pod episode 324, recorded for September 30th, 2025, centers on Clippy’s Revenge and a week of cloud and AI product updates across AWS, Microsoft, and Anthropic. 2. Justin, Jonathan, Ryan, and Matthew host the show, and BoltBot appears as their automation tool for show notes and article handling. 3. The episode asks whether AI assistants and cloud platforms are becoming more useful or just more expensive and restrictive. 4. The hosts return to Google’s complaint to the European Commission about Microsoft’s cloud licensing and the reported 400% markup on Windows server workloads moved off Azure. 5. They cite the UK Competition and Markets Authority’s estimate that restrictive licensing costs UK cloud customers £500 million annually and the US government’s £750 million yearly overspend. 6. The discussion also mentions Microsoft’s disclosure that forcing software customers to use Azure is one of three pillars driving growth. 7. The team covers a European Center for National Political Economy study that says ending restrictive licensing could unlock £1.2 trillion in additional EU GDP by 2030 and £450 billion annually in fiscal savings and productivity gains. 8. They compare Microsoft’s licensing approach with Google’s own customer deals, including the comment that Google gives away egress for free for EU customers but not for US customers. 9. The hosts shift to AI pricing complaints, including the concern that vendors add money to licensing costs after adding AI to a product. 10. They also mention Riverside’s AI editing features and Heather’s human show-notes work as examples of where AI is useful but not fully trusted. 11. Anthropic’s Claude Sonnet 4.5 is described as scoring 77.2 on SWE-bench Verified and keeping focus for over 30 hours on complex multi-step tasks. 12. The episode says Claude Sonnet 4.5 keeps the same pricing as Sonnet 4, at $3 per million input tokens and $15 per million output tokens. 13. The hosts explain the Claude Agent SDK as the infrastructure behind Claude Code, with memory management, permission systems, and sub-agent coordination. 14. They say computer-use improved to 61.4 on the OS World benchmark from 42.2, with browser navigation, spreadsheet manipulation, and other task completion examples. 15. The conversation highlights new Claude-related features such as checkpoints in Claude Code, a native VS Code extension, memory tools in the API, and direct code execution for spreadsheets, slides, and documents in Claude apps. 16. The hosts say early customer results include a 44% reduction in vulnerability intake for security agents, an 18% improvement in planning for Devon, and a zero error rate on an internal code-editing benchmark. 17. The episode’s tone is conversational and technical, with the four hosts interrupting, comparing notes, and drifting into product opinions as they move through the headlines. 18. The format mixes news rundown, vendor commentary, and practical examples from their own workflows, including BoltBot, Claude Code, and browser MCP usage. 19. Listeners who follow AWS, Azure, GCP, and applied AI product changes will get the most out of it. 20. Listeners who want only one vendor, or no cloud pricing talk, may skip it.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from The Cloud Pod | Weekly AI & Cloud News on AWS, Azure & GCP