In October 2025, the author reviewed their AI usage. One year later, significant changes have emerged. The author transitioned from an engineer on Microsoft's Windows AI team to a research engineer at Salesforce AI Research.
Monthly AI token spending now stands at approximately $3,000. While AI usage patterns remain similar -- coding, research, search, summarization, transcription, writing, and art -- the role has shifted toward supporting scientists and building rapid prototypes. A new personal AI policy has been drafted to provide a framework for daily tool use.
The author observes AI leaking into daily life, noting teammates use Claude for Slack messages and researchers delegate thinking to machines. The goal is to adapt the policy over time while focusing on making cool things with other people. Coding practices have evolved dramatically.
The author no longer writes much code personally, instead managing 5-6 Claude Code agents. These agents prototype features in about 20 minutes, write and test code, and create pull requests. This approach allows for rapid experimentation and removes drudgery.
However, the author cautions that left unchecked, agents produce poorly functioning, over-architected slop. Human taste and experience remain essential for quality work. The author uses agents to test audio code across operating systems and implement significant changes, like noise cancellation for Mumble, daily.
The narrative concludes that while AI can help or hurt, a personal policy helps distinguish between the two, ultimately aiming to make cool things with other people.
Source: Hacker News · Summarized by HeadlinesBriefing