Enterprises expect AI agents to handle messy workflows reliably, honor policies by default, operate under heavy load, and show their work.
Netomi(opens in a new window) builds systems that meet that high bar, serving Fortune 500 customers like United Airlines and DraftKings. Their platform pairs GPT‑4.1 for low-latency, reliable tool use with GPT‑5.2 for deeper, multi-step planning, running both inside a governed execution layer designed to keep model-driven actions predictable under real production conditions.
Running agentic systems at this scale has given Netomi a blueprint for what makes these deployments work inside the enterprise.
## Lesson 1: Build for real-world complexity, not idealized flows
A single enterprise request rarely maps to a single API. Real workflows span booking engines, loyalty databases, CRM systems, policy logic, payments, and knowledge sources. The data is often incomplete, conflicting, or time-sensitive. Systems that depend on brittle flows collapse under this variability.
Netomi designed its Agentic OS so OpenAI models sit at the center of a governed orchestration pipeline built for this level of ambiguity. The platform uses GPT‑4.1 for fast, reliable reasoning and tool-calling—critical for real-time workflows—and GPT‑5.2 when multi-step planning or deeper reasoning is required.
> “Our goal was to orchestrate the many systems a human agent would normally juggle and do it safely at machine speed.”
—Puneet Mehta, CEO, Netomi
To ensure consistent agent behavior across long, complex tasks, Netomi follows the agentic prompting patterns recommended by OpenAI:
Together, these patterns help the model reliably map unstructured requests to multi-step workflows and maintain state across discontinuous interactions.
Few industries expose the need for multi-step reasoning as clearly as airlines, where one interaction routinely spans multiple systems and policy layers. A single question may require checking fare rules, recalculating loyalty benefits, initiating ticket changes, and coordinating with flight operations.
“In airlines, context changes by the minute. AI has to reason about the scene the customer is in—not just execute a siloed task,” said Mehta. “That’s why situational awareness matters way more than just workflows, and why a context-led ensemble architecture is essential.”
With GPT‑4.1 and GPT‑5.2, Netomi can keep extending these patterns into richer multi-step automations—using the models not just to answer questions, but to plan tasks, sequence actions, and coordinate the backend systems a major airline depends on.
## Lesson 2: Parallelize everything to meet enterprise latency expectations
In high-pressure moments—rebooking during a storm, resolving a billing issue, or handling sudden spikes in demand—users will abandon any system that hesitates. Latency defines trust.
Most AI systems fail because they execute tasks sequentially: classify → retrieve → validate → call tools → generate output. Netomi instead designed for concurrency, taking advantage of low-latency streaming and tool-calling stability of GPT‑4.1.
GPT‑4.1 provides fast time-to-first-token and predictable tool-calling behavior, which make this architecture viable at scale; while GPT‑5.2 provides deeper multi-step reasoning paths when needed. Netomi’s concurrency framework ensures the _total system_, not just the model, stays under critical latency thresholds.
These concurrency demands aren’t unique to airlines. Any system exposed to sudden, extreme traffic surges needs the same architectural discipline. DraftKings, for instance, regularly stress-tests this model, with traffic during major sporting events spiking above 40,000 concurrent customer requests per second.
During such events, Netomi has sustained sub-three-second responses with 98% intent classification accuracy, even as workflows touch accounts, payments, knowledge lookups, and regulatory checks.
“AI is central and critical to how we support customers in the moments that matter most,” said Paul Liberman, Co-Founder and President of Operations at DraftKings. “Netomi’s platform helps us handle massive spikes in activity with agility and precision.”
At scale, Netomi’s concurrency model depends on the fast, predictable tool-calling of GPT‑4.1, which keeps multi-step workflows responsive under extreme load.
## Lesson 3: Make governance an intrinsic part of the runtime
Enterprise AI must be trustworthy by design, with governance woven directly into the runtime—not added as an external layer.
When intent confidence drops below threshold, or when a request cannot be classified with high certainty, Netomi’s governance mechanisms kick in to determine how the request is handled, ensuring the system backs off from free-form generation in favor of controlled execution paths.
At a technical level, the governance layer handles:
In highly regulated domains like dental insurance, this kind of governance is non-negotiable. A Netomi customer in the insurance industry processes close to two million provider requests each year across all 50 states, including eligibility checks, benefits lookups, and claim status inquiries where a single incorrect response can create downstream regulatory or service risk.
During open enrollment, when scrutiny and volume peaked, the company needed AI that enforced policy as part of the runtime itself. Netomi’s architecture was up to that complex requirement.
“We built the system so that if the agent ever reaches uncertainty, it knows exactly how to back off safely,” said Mehta. “The governance is not bolted on—it’s part of the runtime.”
## A blueprint for building agentic systems that work for the enterprise
Netomi’s path shows what it takes to earn enterprise trust: build for complexity, parallelize to meet latency demands, and bake governance into every workflow. OpenAI models form the reasoning backbone, while Netomi’s systems engineering ensures that intelligence is operationally safe, auditable, and ready for Fortune 500 environments.
These principles helped Netomi scale across some of the world’s most demanding industries—and offer a blueprint for any startup looking to turn agentic AI into production-grade infrastructure.
Rakuten fixes issues twice as fast with Codex API Mar 11, 2026
Wayfair boosts catalog accuracy and support speed with OpenAI ChatGPT Mar 11, 2026
How Descript enables multilingual video dubbing at scale API Mar 6, 2026
Our Research * Research Index * Research Overview * Research Residency * OpenAI for Science * Economic Research
Latest Advancements * GPT-5.3 Instant * GPT-5.3-Codex * GPT-5 * Codex
Safety * Safety Approach * Security & Privacy * Trust & Transparency
ChatGPT * Explore ChatGPT(opens in a new window) * Business * Enterprise * Education * Pricing(opens in a new window) * Download(opens in a new window)
Sora * Sora Overview * Features * Pricing * Sora log in(opens in a new window)
API Platform * Platform Overview * Pricing * API log in(opens in a new window) * Documentation(opens in a new window) * Developer Forum(opens in a new window)
For Business * Business Overview * Solutions * Contact Sales
Company * About Us * Our Charter * Foundation * Careers * Brand
Support * Help Center(opens in a new window)
More * News * Stories * Livestreams * Podcast * RSS
Terms & Policies * Terms of Use * Privacy Policy * Other Policies
(opens in a new window)(opens in a new window)(opens in a new window)(opens in a new window)(opens in a new window)(opens in a new window)(opens in a new window)
OpenAI © 2015–2026 Manage Cookies
English United States
