Home / Technology / AI engineers code 1% of the time
AI engineers code 1% of the time
4 Aug
Summary
- Engineers spend only 1% of time coding; AI agents do the rest.
- AI excels at new code, but faces challenges with existing codebases.
- Managing AI costs requires strategic model selection and usage caps.
At Kilo Code, engineers now dedicate only about 1% of their time to writing or reviewing code, with AI agents performing the rest. This significant shift is prompting dev teams to address new challenges regarding system safety, AI error correction, and managing multi-model architectures.
Tech leads from companies like Replit, Kilo Code, and Symbotic view this evolution as a natural integration of agentic AI into enterprise workflows. While AI effectively handles greenfield development, existing codebases (brownfield) still require significant human expertise for updates and maintenance, highlighting the complementary roles of AI and human developers.
Managing escalating AI costs is a primary concern. Companies are implementing strategies such as "tokenmaxxing" and careful model selection, favoring less expensive open-weight models for bulk tasks after initial project architecture is handled by premium models. This approach aims to optimize cost versus capability, ensuring a positive return on investment.
Visibility into AI usage and costs is becoming critical. Symbotic, for instance, has established monthly cost tiers for employees and uses tools to monitor usage trends. This allows for better cost control and efficiency, especially after legacy discounts ended, forcing a re-evaluation of AI tool usage and model selection.
