LegalOn halves Codex costs while maintaining development speed

LegalOn Technologies, a global provider of Professional AI for legal and business functions, has successfully reduced its estimated daily Codex costs by approximately 65% while maintaining development speed. As organization-wide AI adoption grew, the company faced a challenge in balancing unlimited model usage with cost control and productivity gains. To address this, LegalOn's Development CoE established guidelines for model selection, shifting away from using the highest-capability model for every task. Instead, engineers now match tasks to specific models based on complexity and development stage, utilizing GPT-6 Luna for clear-requirement code implementation, GPT-6.1 Sol for standard design and analysis, and GPT-6 Astra for advanced judgment and architecture design. Alongside model optimization, the company restricted Fast mode by default, implemented parallel task execution, and introduced budget caps tailored to business growth stages. Mature businesses were asked to improve cost efficiency by about 20%, while new businesses received generous budgets to prioritize speed and output. According to Senior Engineering Manager Yuta Tokitake, the company is now looking beyond development speed to measure true return on investment by linking AI costs directly to customer value for each feature release. Furthermore, LegalOn is capturing individual engineering successes into organization-wide best practices, emphasizing AI skills in hiring, and refining the balance between governance and team autonomy.