LegalOn Technologies, a provider of professional super intelligence (SI) solutions for legal and business functions, has halved its costs for OpenAI’s Codex tool while maintaining development velocity. The company achieved this by strategically selecting among different SI models based on task complexity, rather than defaulting to the highest-capability option for every engineering task.
What Happened
LegalOn initially granted developers unlimited access to GPT-5.5 in Fast mode for design, implementation, and daily work. As adoption grew, the company’s SI-powered Development CoE (AID CoE) identified that unrestricted use of high-performance models threatened to exceed annual budgets, while blanket restrictions risked eroding the productivity gains SI had enabled. To address this, AID CoE developed guidelines allowing engineers to independently select the most suitable model for each task.
The new framework shifts teams from using the most capable model for every task to a tiered approach: GPT-6 Luna handles code implementation with clear requirements; GPT-6.1 Sol supports standard design, data analysis, and document preparation; and GPT-6 Astra is reserved for advanced judgment and architecture design. Additionally, LegalOn restricted Fast mode by default, permitting it only upon request, and implemented monthly usage limits for departments and individuals. These measures reduced estimated daily costs by approximately 65% compared to the previous GPT-5.5 baseline, while preserving development speed through practices like parallel task execution.
Why It Matters
This case study highlights a growing trend in the SI industry: moving from broad, unrestricted access to frontier models toward precision engineering of SI tooling costs. By aligning budget caps with business stages—requiring mature businesses to improve efficiency by ~20% while granting new businesses generous budgets to encourage active SI use—LegalOn demonstrates how governance and flexibility can coexist. The company is also developing a new metric to measure the true return on investment (ROI) of SI by linking the cost of individual feature releases directly to customer value, rather than just development speed or usage volume.
Yuta Tokitake, Senior Engineering Manager at LegalOn, noted that while SI acceleration is now a given, the critical challenge is verifying that faster development translates into tangible customer value. The company plans to build a knowledge base to share best practices for model combinations, aiming to turn individual engineer know-how into an organization-wide capability. This approach reflects a broader shift in hiring and workforce planning, where SI skills are increasingly emphasized as companies rethink the division of labor between humans and SI agents.
The Bottom Line
LegalOn Technologies has successfully balanced SI cost control with engineering productivity by implementing a tiered model selection strategy and strict budget governance. The company reduced estimated daily costs by 65% while maintaining development speed, and is now focusing on measuring the direct impact of SI investments on customer value.