a day ago
- AI models have reached an inflection point where their output is worth more than the cost for real development work, leading to increased inference spending.
- The author's annual AI inference spend grew from $100 in 2023 to $5000 in 2025, reflecting growing utility.
- o1-pro was the first model consistently useful for professional coding, capable of debugging large files and saving hours per week.
- AI tools like Claude Code and prompt libraries in Zed Editor dramatically reduce task times (e.g., from 20 minutes to 2 minutes for boilerplate functions).
- The author built multiple features and a full website backend using Claude Code, estimating 100 hours of manual work reduced to under 10 hours.
- Current AI tools save the author about 10% of work time, with potential for 2-3x acceleration by end of 2026.
- Models like o3 are expected to significantly outperform o1-pro, but initial costs will be high (e.g., $10-20 per debugging session via API).
- Senior dev taste (software architecture judgment) remains a challenge for AI, as synthetic data for high-quality design is hard to generate.
- Junior developers may be replaced faster than seniors because their tasks are easier to automate; juniors are advised to develop senior dev taste quickly.
- Inference spend is expected to jump to thousands or tens of thousands of dollars per year per developer, following Jevons paradox: cheaper AI leads to more usage.
- Context size is a current bottleneck; future cost drops will enable expensive but valuable codebase-wide scans.