Do you want smarter ideas in your entrance tray? Register in our weekly newsletters to obtain only what matters to the leaders of AI, data and business security. Subscribe now
In Venturebeat’s Transform 2025 ConferenceOlivier Godement, product manager of the OpenAI API platform, provided a look behind the scene of how business teams are adopting and deploying AI -scale agents.
In a 20 -minute discussion panel, I organized exclusively with the winged, the former Stripe researcher and current head of the OpenAI API unpacked the latest OpenAI developer tools, API responses and SDK agents, while highlighting real world patterns, security considerations and return examples of early adopter costs such as Stripe and Box.
For business leaders who cannot attend the live session, here are the 8 main conclusions:
Agents are quickly moving from the prototype to production
According to Dioso, 2025 marks a real change in how AI is being displayed at a scale. With more than one million monthly active developers who now use the OpenAI API platform worldwide, and the use of tokens increased by 700% year after year, AI is out beyond experimentation.
“Five years have passed since we launched essentially GPT-3 … and the man, the last five years have been quite wild.”
Doynement emphasized that current demand is no longer just about chatbots. “The cases of use of AI are moving from simple questions and answers to use cases in which the application, the agent can do things for you.”
This change led Openai to launch two main orientation tools for developers in March: the API responses and the SDK agents.
When to use individual agent architectures versus subgentes
A main theme was the architectural choice. The highlight pointed out that single agent loops, which encapsulate the full access of the tool and context in a model, are conceptually elegant but often little practical on a scale.
“Building precise and reliable individual agents is difficult. As, it is really difficult.”
As complexity increases, more tools, more possible, more logical users’ inputs, the dishes often move towards modular architectures with specialized subciburants.
“A practice that has emerged is essentially to break down agents in multiple subcibents … you would separate concerns such as software.”
These sub -management work, as roles in a small team: a classification agent classifies the intention, level one agents handle routine problems, and others increase or solve edge cases.
Why API answers is a step change
He positioned the API of the answers as a fundamental evolution in developer tools. Previously, developers manually orchestrated model call sequences. Now, that orchestration is handled internally.
“The API of the answers is probably the new largest abstraction layer we present from almost GPT-3”.
It allows developers to express intention, not only configure model flows. “You care to return a very good customer response … the response API essentially handles that loop.”
It also includes incorporated capabilities for the recovery of knowledge, web search and calls of functions, toolas that companies need for workflows of real world agents.
Observability and security are integrated
Security and compliance were the most important. Diende cited key railings that make OpenAi battery viable for regulated sectors such as finance and medical care:
- Policy -based refusals
- SOC-2 record
- Data residence support
The evaluation is where winged sees the greatest gap between demonstration and production.
“My opinion is that the evaluation of the model is probably the largest bottleneck for the mass adoption of AI.”
Operai now includes tracking and evaluation tools with the API battery to help teams define how success is seen and track how agents work over time.
“Unless investing in evaluation … It is really difficult to develop that confidence, that confidence that the model is being precise, reliable.”
Early ROI is visible in specific functions
Some business use cases are already delivering measurable profits. Giving Aldo examples of:
- Stripewho uses agents to accelerate the management of invoices, inform “35% faster invoice resolution”
- Boxwho launched knowledge attendees that enable “Zero -touch ticket triage”
Other cases of high value use include customer service (including voice), internal governance and knowledge attendees to navigate dense documentation.
What is needed to launch in production
The highlighting emphasized the human factor in successful deployments.
“There is a small fraction of very high -end people who, every time they see a problem and see a technology, run it.”
These internal champions do not always come from engineering. What unites them is persistence.
“Your first reaction is, ok, how can I make it work?”
Operai sees many initial implementations promoted by this group: people who promoted the early use of Chatgpt in the company and are now experiencing with complete agents systems.
He also pointed out a gap that many overlook: the experience in the domain. “Knowledge in a company … does not meet engineers. It meets OPS teams.”
Making agents construction tools accessible to non -developers is a challenge that Operai aims to address.
What follows for business agents?
The altre offered a look at the road map. Operai is actively working on:
- Multimodal agents that can interact through text, voice, images and structured data
- Long -term memory to retain knowledge through sessions
- Orchestration between cloud To admit complex and distributed you environments
These are not radical changes, but iterative layers that expand what is already possible. “Once we have models that can think not only for a few seconds, but for minutes, for hours … that will allow some quite amazing use cases.”
Final word: reasoning models do not have
The altre closed the session reaffirming its belief that models with reasoning capacity, those that can be reflected before responding, will be the true facilitators of the long -term transformation.
“I still have conviction that we are more or less at the level of maturity GPT-2 or GPT-3 of those models … we are still scratching the surface in what the reasoning models can do.”
For business decision makers, the message is clear: the infrastructure for agent automation is here. What matters now is to build a focused case, empower interfunctional equipment and be ready to iterate. The next phase of value creation is not in novel demos, but in durable systems, made up of the needs of the real world and the operational discipline to make them reliable.
#OpenAIs #API #leader #companies #succeed #agents
