Netomi's Lessons for Scaling Agentic Systems in Enterprise Environments

Netomi builds agentic systems combining GPT-4.1 and GPT-5.2 for Fortune 500 companies like United Airlines and DraftKings. GPT-4.1 is used for low-latency tool use, while GPT-5.2 is used for deep multi-step planning, operating within a controlled execution layer to ensure predictable outcomes.

Lesson One: Prepare for real-world complexity, not ideal flows Enterprise requests aren't solved by a single API; they involve complex workflows that span diverse sources such as booking engines, CRM, and payment systems. Netomi implements the following patterns through its Agentic OS to handle ambiguous situations.

• Persistence reminders: Helps GPT-5.2 maintain reasoning throughout long workflows
• Explicit tool-use expectations: Guides GPT-4.1 to call tools for authoritative information without hallucination
• Structured planning: Designs and executes multi-step tasks using GPT-5.2's reasoning capabilities
• Agent-driven rich media decisions: Determines when to return multimodal elements (images, video, etc.)

Lesson Two: Parallelize everything to meet enterprise latency requirements Users are highly sensitive to latency. Instead of the traditional sequential execution approach (classification → retrieval → verification → tool call → generation), Netomi designed a concurrency architecture leveraging GPT-4.1's low-latency streaming and stable tool-calling capabilities.

This design proves its worth during sudden traffic spikes. DraftKings experiences over 40,000 concurrent requests per second during major sporting events, and through the Netomi platform, it maintains 98% intent classification accuracy and response times under 3 seconds, demonstrating the system's reliability.