There is a company at the center of this year’s keynote at the AI Networking Summit, and I want to tell you about it before you meet it in New York.

It is not a careless company. That is the point.

It has a managed registry. Nothing its engineers or its agents run comes straight off the public internet; it comes through a central, curated source. Every artifact that passes through that source is scanned. A software bill of materials is generated for it. It is checked against every known vulnerability. Then it is held for days, so that any vulnerability discovered after release has time to surface before the company touches it. Only then is it approved for use.

That is not a strawman. That is best practice as the industry teaches it today, and this company follows every line of it.

It also runs AI agents in production. Real ones, with real jobs: an agent that tunes the data-center fabric, an agent that watches the network and proposes fixes, an agent in the security operations center that investigates alerts. Each one was provisioned carefully, with the rights it needs to do its work, by people who thought hard about what those rights should be.

On a Tuesday, one of those agents does exactly what it was built to do. It pulls its scheduled update from the managed registry. The update has passed every check. The scan is clean. The bill of materials is in order. The hold has expired. Everything is green.

Six days later, nothing in that company works.

What happens in those six days is the first ten minutes of the keynote, and I am not going to tell you. I will tell you that nobody on the team who built this demo found it far-fetched. The people who reviewed it run security and infrastructure at some of the largest enterprises in finance, healthcare, logistics, and defense, and the notes they sent back were not “that could never happen.” They were “that is not quite how it would happen; here is how it would.” We rewrote it to match. Several times. The version you will see on stage is the version they could not argue with.

Then the demo does something I have not seen a keynote do. It rewinds.

Same company. Same Tuesday. Same agent, same update, same clean scan. One thing is different, and it is not a product. It is a change in how the company governs its agents, a change that three ONUG Collaborative working groups have spent the summer turning into a reference architecture and a set of requirements. With that one change in place, the same six days play out again, and you watch what the difference buys.

I will give you the shape of it without the substance. The demo has seven beats. The first is fear. The last is, frankly, greed: what a company can do once it stops treating every new agent as a risk to be contained and starts treating its control plane as the thing that lets business units build. Between those two ends, you will see the same architecture tested in two places where it has to prove itself: the network that agents are starting to run, and the security operations center that agents are starting to staff.

And then the demo turns to the vendors in the room and hands them the problem.

That is the part I am most looking forward to. Fifty-seven technology companies will be on the floor in New York. Every one of them has been invited to take the keynote’s scenario and show, in their own product, how they would answer it, in one of three lanes that match the three working groups. Not a slide. A working demonstration. The practitioners in the audience will vote, in the app, on who answered best, and we will announce the results before anyone goes home.

So here is what I can promise about 28 October. You will watch a company that did everything right lose everything in six days, and you will recognize the company. You will watch the same six days with one thing changed. And you will find out what that one thing is at the same moment as everyone else in the room.

I have been doing this for a long time. I have never been more certain that a keynote would be worth the seat.

The scenario is already in the hands of the Collaborative’s members. The rest of you will have to come to New York.

Nick Lippis is co-founder and co-chair of ONUG. The AI Networking Summit is in New York on 28 and 29 October 2026.