OpenAI publishes how it's rebuilding its support so that every interaction improves everything. Can you imagine that when you ask for help you're not opening a ticket but triggering a system that learns and adjusts instantly? That's the bet they describe in their September 29, 2025 post. (openai.com)
Más que tickets, un nuevo modelo operativo
Support is no longer just answering queues and measuring throughput. OpenAI proposes that, faced with hypergrowth and massive scale, the solution was to redesign the whole operating model so every conversation helps the next one. Instead of treating tickets as isolated endpoints, they treat them as data that feeds continuous improvements. (openai.com)
Why does that matter for you or your business? Because it reduces response time, improves consistency, and turns people who handle support into system builders, not just executors.
Tres bloques que hacen girar el sistema
OpenAI frames support around three practical pillars:
- Surfaces: the places where interaction happens, from chat and email to in-product help.
- Knowledge: not just static documents; it's living knowledge that updates with real conversations and context.
- Evals y clasificadores: shared definitions of quality built by people and software that let you measure and improve. (openai.com)
That loop between surface, knowledge, and evaluations lets an improvement in one channel replicate across all others. Sounds like magic? It's applied engineering to experience.
Soporte: de operadores a pensadores de sistemas
One repeated idea is changing the agent role. Rather than focusing only on resolving tickets, agents identify patterns, design tests, and propose classifiers. They become the product's eyes and hands, providing feedback that feeds the system. This also changes training: it's no longer memorizing policies, but evaluating interactions and closing structural gaps. (openai.com)
"Agents aren't just responding to tickets. They're informing our knowledge base and our policies. They have an ear to the ground that we don't." (openai.com)
De primitivas a producción: las herramientas que activan esto
OpenAI mentions several technical pieces that make the approach viable in production:
Agents SDKfor step-by-step traces and observability.Responses APIto build classifiers for tone, correctness, and policy compliance.Realtime APIfor voice support.- The
Evalsdashboard to measure quality over time. (openai.com)
That lets you move quickly from a Q&A prototype to dynamic actions like refunds, incident lookups, or specific automations. If you work in product, think of this as the difference between integrating just a chatbot and having a platform that lets you iterate with metrics and traceability.
Aprendizaje que se compone con el tiempo
The post emphasizes that evaluations turn everyday conversations into production tests. It's not just about closing the ticket; it's measured whether the answer was clear, kind, and correct. Agents flag good and bad examples that become evaluations and, in turn, guide the live model's behavior. The result: faster responses and a system that improves with each interaction. (openai.com)
A key point: they also know when the model shouldn't answer, and that's when a person steps in. That coordination reduces risk and keeps human control where it matters. (openai.com)
¿Qué significa esto para empresas y usuarios?
- For small teams: the possibility of integrating contextual help without spending years on data pipelines.
- For high-demand products: scalability with traceability and metrics that show whether improvements work.
- For end users: less friction, faster and more consistent answers where you're using the service.
If you run support, the invitation is clear: turn your team into makers of the system. It's not just optimizing KPIs; it's designing a machine that learns.
Lectura y recursos
You can review the full post in OpenAI's note to dive into technical details and team quotes. (openai.com)
The model they describe isn't a marketing trick. It's a practical recipe to bring AI into the heart of support operations and make every interaction count.
And you? Are you ready for your support to stop being a destination and become an action integrated into your product?
