Enterprise Intelligence Group
OpenAI's new enterprise agent platform, Presence, builds governance, human approval and testing into AI agents. Here's what it means for retail merchandising teams — and how to be ready.
OpenAI has launched Presence, an enterprise platform for deploying AI voice and chat agents across internal and customer-facing workflows — with company policies, permission controls, human-escalation rules and pre-deployment testing built in. For retail merchandising teams, the significant part isn't the technology; it's what OpenAI chose to put around it. This is good news, and it points clearly at the work that makes it pay off.
What Presence actually is
According to the launch coverage, Presence lets companies stand up agents scoped to a specific job — billing, IT support, claims — rather than one general-purpose bot. Each deployment bundles the organization's policies and standard operating procedures, permission controls that limit what data and systems an agent can touch, escalation rules that hand off to a human when needed, and simulations that test the agent against edge cases and high-risk scenarios before it goes live. A Codex-powered loop then reviews interactions and proposes improvements — but staff must approve changes before they ship. OpenAI reports it already runs on its own English-language support line, resolving around 75% of inbound calls ([VentureBeat](https://venturebeat.com/orchestration/openai-unveils-presence-a-new-platform-that-lets-enterprises-launch-and-manage-realtime-voice-agents-and-chatbots), [Help Net Security](https://www.helpnetsecurity.com/2026/07/22/openai-presence-ai-agent-platform/)).
Why this is good news for retail
For years the enterprise AI conversation was dominated by demos and open-ended chatbots. Presence is a sign the market is maturing toward what actually works in a real operation: narrow scope, governed knowledge, permissioned data, human oversight, and testing before production. Those aren't hedges — they are the conditions under which an agent earns trust. When the largest AI vendor designs its flagship agent product around policies, approvals and guardrails, it validates the approach serious operators were already taking.
For retail and merchandising leaders weighing where AI fits, that's reassuring. The safe path and the valuable path are the same path — and the platforms are now being built for it.
The one thing a platform can't supply
Here's the part worth sitting with. Presence can enforce your policies, retrieve your approved knowledge, and follow your escalation rules — if those things exist in a form a system can use.
In retail merchandising, they usually don't. The logic behind a ranging decision, the reason one supplier gets flexibility, the exception that applies only in small-format stores, the real criteria a category review turns on — most of it lives in experienced people's heads, in scattered spreadsheets and in old decks. A platform gives you the guardrails; it cannot tell you what the rules are . That has to be captured, validated and governed first.
This is the whole of Intelligence Before Automation : the agent platform is the last layer, and it only performs as well as the knowledge beneath it. Point even a well-governed agent at ungoverned merchandising knowledge and it will confidently apply a stale exception or an outdated supplier term — at speed.
What merchandising teams should do now
The arrival of platforms like Presence makes the preparation more valuable, not less. Practical first steps:
1. Pick one high-value area — category reviews, assortment decisions, supplier prep — and map how the decisions are actually made. 2. Capture and validate the knowledge those decisions rely on, and assign owners so it stays current. 3. Define the guardrails in your own terms — what an assistant may read, what it may do, where a human must decide. 4. Then choose the platform. With the foundation in place, Presence — or any enterprise agent platform — has something trustworthy to run on.
Teams that do this will be able to adopt these platforms quickly and safely. Teams that skip it will buy capable technology and watch it stall on ungoverned information.
Frequently asked questions
The EIG Library · Insights · AI Maturity Assessment · Contact