OpenAI Just Launched Presence: What It Actually Does (And Why "Free" Isn't the Right Word)

So OpenAI dropped something new called Presence on July 22, and it's already getting a lot of buzz. Real talk though: if you came here looking for a free tool to mess around with this weekend, I've got to stop you right there. This isn't that kind of launch, and I'll explain exactly why.

What Presence Actually Is

Presence is an enterprise product. It's built for big companies that want to deploy AI agents for stuff like customer support, sales outreach, insurance claims, and IT help desk requests. Think less "chatbot you sign up for" and more "system OpenAI's engineers help a company build and wire into their existing tools."

According to OpenAI's own announcement, Presence bundles a few things together: company policies and procedures, permission controls that limit what an agent can touch, escalation rules for when a human needs to step in, and pre-launch simulations that test the agent against tricky or high-risk situations. After launch, it keeps learning from real conversations, and OpenAI's Codex model reviews those sessions to suggest tweaks, though a human still has to approve any changes before they go live.

Why This Matters

Here's the part that's genuinely interesting. OpenAI says it's been running Presence on its own English-language phone support line, and the thing is already resolving about 75% of inbound calls without a human getting involved. That's not a hypothetical demo, that's live production traffic. A Codex-driven feedback loop reportedly cut human handoffs by 15 percentage points in just 10 days of tweaking.

If you've used AI customer support before, you know how bad it usually feels. Rigid scripts, no memory of context, escalations that go nowhere. What makes Presence different, at least on paper, is the ongoing improvement loop. Most support bots get built once and left alone. This one is supposed to keep getting better based on what actually happens in production.

So Who Can Actually Use It?

This is where the "free trial" idea falls apart. Presence launched as a limited general availability program, not a self-serve product. You can't just sign up and start building. Deployments are handled by OpenAI's own Forward Deployed Engineers and select partner integrators, which is basically consulting work dressed up as a product launch.

Companies like BBVA, SoftBank, and IAG are reportedly among the early users testing it out. So if you're an individual developer hoping to poke around with Presence the way you would with ChatGPT, that's not currently on the table. There's no public pricing either, so anything you see floating around about cost is speculation until OpenAI says otherwise.

What's Next

OpenAI has said it wants Presence to eventually cover more channels beyond voice and chat, with email getting mentioned in some of its promotional materials. That part hasn't been confirmed as available yet, so treat it as a stated ambition rather than a shipped feature.

It's also worth watching how this fits into OpenAI's bigger enterprise push. This comes not long after other business-focused tools from the company, and it lines up with the "Deployment Company" subsidiary OpenAI set up earlier this year. Honestly, this looks like OpenAI leaning harder into selling services to big businesses, not just API access to developers.

Bottom Line

Presence is real, it's live, and the early results OpenAI is sharing sound solid. But it's an enterprise sales product, not something you or I can log into this afternoon for free. If a headline promised you a free trial, that's not accurate based on what OpenAI has actually announced.

Curious whether this enterprise-agent approach spreads to the rest of OpenAI's lineup, or stays a niche offering for big companies willing to pay for hand-holding?