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

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

·1 hr 4 min·4 clips

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

1. The Cloud Pod episode 324, recorded September 30th, 2025, centers on AI assistants and cloud platform changes across AWS, Microsoft, and Anthropic. 2. Justin, Jonathan, Ryan, and Matthew host the show, and their roles matter because they compare cloud releases from operator and user perspectives. 3. The episode asks whether newer AI and cloud features actually work as advertised, or whether they mostly add cost and complexity. 4. The opening joke about Justin going on vacation turns into a conversation about who will record next week and whether BoltBot will keep the show notes moving. 5. A follow-up segment revisits Google’s formal complaint to the European Commission over Microsoft’s cloud licensing practice and the reported 400% markup for Windows server workloads outside Azure. 6. The hosts cite the UK Competition and Markets Authority estimate that restrictive licensing costs UK cloud customers £500 million annually. 7. They also mention a study saying ending restrictive licensing could unlock £1.2 trillion in additional EU GDP by 2030 and generate £450 billion in annual fiscal savings and productivity gains. 8. The discussion shifts to Microsoft’s Office agent mode, where Excel, Word, and Copilot Chat use OpenAI reasoning models and Anthropic models for multi-step document creation. 9. The hosts describe Excel prompts for finance reports, loan calculators, and budget trackers, with validation steps and spreadsheet benchmarking. 10. They also cover Word’s conversational drafting flow and Office Agent’s three-step process for researching and generating PowerPoint presentations and documents. 11. Anthropic’s Claude Sonnet 4.5 gets the most technical attention, including its 77.2 SWE-bench Verified score and support for over 30 hours of focus on complex tasks. 12. The hosts note that Sonnet 4.5 keeps the same pricing as the previous Sonnet model at $3 per million input tokens and $15 per million output tokens. 13. The segment also mentions the Claude agent SDK, which adds memory management, permission systems, and sub-agent coordination to the same infrastructure behind Claude Code. 14. The hosts call out improved computer use performance at 61.4 on OSWorld, plus the Claude for Chrome extension for Max subscribers. 15. They cite early customer results such as a 44% reduction in vulnerability intake for security agents and an 18% improvement in planning performance for Devon. 16. AWS topics include ECS managed instances, which the hosts describe as a bridge between serverless simplicity and EC2 flexibility with GPU support. 17. The AWS rundown also covers IPv6-only support for ECS and EC2 Auto Scaling, plus a new AMI governance feature with Marketplace Codes, Deprecation Time, Creation Date, and AMI Names. 18. The tone is conversational and skeptical, with the hosts moving between product details, personal workflow examples, and jokes about AI slop, licensing, and cloud bills. 19. Listeners who manage cloud infrastructure, follow AI tooling, or care about AWS and Microsoft product changes will get the most value. 20. Listeners who want a tightly edited, narrative episode without product chatter 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