OpenAI Presence: Inside the AI Agent Platform That Just Declared War on Salesforce and ServiceNow

On July 22, 2026, OpenAI did something it has spent two years hinting at but never quite doing: it shipped a product aimed directly at the enterprise software stack. The product is called OpenAI Presence, and it is, in effect, an AI agent platform designed to run customer-facing and back-office voice and chat workflows for large organisations — the kind of work that today lives inside Salesforce Service Cloud, ServiceNow, Genesys, and a dozen niche contact-centre tools.

OpenAI is selling Presence as a turnkey platform, not a model API. Companies can deploy agents that answer questions, resolve support tickets, take sales calls, file internal requests, and use company systems to actually get things done. The platform combines agent simulations, policy checks, escalation rules, and what OpenAI describes as a "Codex-powered improvement loop" that continuously refines each agent based on real interactions. It is, in other words, the missing layer between GPT-5-class models and the messy reality of running a 10,000-seat contact centre.

Why this launch is different

OpenAI has flirted with enterprise software before — through ChatGPT Enterprise, through custom GPTs, through a long string of API partnerships with incumbents. What Presence signals is a shift from "we'll power your stack" to "we are your stack." That is a meaningfully different posture, and it lands directly on the turf of companies whose valuations depend on workflow lock-in rather than model quality.

The competitive map is now explicit. Salesforce spent 2025 turning Agentforce into the centrepiece of its growth story. ServiceNow has been pitching "AI agents that do work" to every CIO in the Fortune 500. Microsoft has been weaving Copilot Studio into every line of business that runs on Azure. AWS and Google have their own agent platforms in Bedrock and Vertex. Presence is OpenAI's answer to all of them, with a credibility argument incumbents cannot easily replicate: the agent runtime is built on the same frontier models that define the state of the art, and the improvement loop is grounded in OpenAI's own coding and reasoning models rather than a third-party dependency.

What is actually in the platform

The product has three layers worth paying attention to. The first is a deployment layer — companies can configure agents in days, not the quarters a typical CRM rollout demands. The second is a governance layer — policy checks, escalation rules, and audit trails designed to satisfy the compliance teams that have been the single biggest blocker to agent adoption in regulated industries. The third is an evaluation loop that runs simulated customer interactions against each agent before it goes live and uses real production data to keep improving performance. OpenAI is positioning this as the answer to the most common objection to AI agents: that they are not safe enough to let near customers.

The Codex improvement loop is the most novel piece. By using OpenAI's coding model to read transcripts, identify failure modes, and propose agent configuration changes, Presence effectively gives each customer a continuous optimisation pipeline that no incumbent can match with human-tuning alone. It is the same playbook that turned foundation models into infrastructure: ship the model, then make the loop that improves it.

What it means for growing markets

For companies in emerging markets — and for founders thinking about building on top of agent platforms — Presence is a more complicated story. On one hand, a low-friction way to deploy multilingual, voice-capable agents could collapse the cost of running customer operations in markets where talent is expensive and turnover is brutal. A Zimbabwean fintech or an Indonesian logistics company could stand up a competent English-, Shona-, and Bahasa-capable support operation in a week. On the other hand, centralising that capability inside an OpenAI-controlled platform concentrates risk: pricing power, data sovereignty, and the politics of which languages and use cases get prioritised.

The realistic view

OpenAI Presence is not going to displace Salesforce overnight. Incumbents have data, distribution, and switching costs that no model vendor can wave away. But the launch makes one thing clear: the enterprise software layer of the next decade is being reshaped around agents that can actually execute work, and OpenAI intends to own the platform those agents run on. The companies that thrive will be the ones that treat Presence as infrastructure to build on — not as a feature, and not as a threat.