No Single Midnight
At 11:45 on the last night of 1999, I took my wife’s hand and led her outside a friend’s house.
We were at a Y2K party. I had heard plenty about what might happen when the clocks rolled over. I had also decided that if nuclear weapons were launched shortly after midnight, I wanted to be doing exactly what I would choose to do with my last few minutes.
I got down on one knee and asked her to marry me again.
Our sixteenth anniversary was coming in February. The first ring had cost almost everything in my bank account, leaving me $120. It was tiny. She didn’t care.
This one was a big do-over diamond.
She cried — a lot.
I cried.
Midnight passed, and we left the party early.
I thought about that night when I woke at 12:15 this morning. Trust, this essay, and the news had followed me into bed. Now I was awake with Y2K, its preparations, and the strange experience of waiting for something whose consequences nobody could fully know.
This time, no single midnight is coming.
We’re discovering what increasingly capable intelligent systems can do as we put them to work. AI, SI — whatever we end up calling what comes next. We’re also discovering what they can do that nobody intended — or that someone intended very much, at another person’s expense. The discoveries arrive together, through research, ordinary use, failures, and public disclosures.
So does the excitement. It hasn’t left me.
By one, I had made coffee and picked up Streets of Laredo. I read for about an hour. By two-thirty, I was back thinking about trust.
Somewhere in those hours, I decided to stop asking one essay to carry everything. I could take my time. Tell it in parts. Learn something along the way.
Another thought — a persistent feeling — I couldn’t put down: I believe I can contribute to this. All of this, whatever it becomes and however complex it gets.
Much of my working life has been spent helping people turn possibilities into things that work. I enjoy ambition and hard questions. I love the moment a team sees something none of us could have reached alone. I also know how much work remains after everyone agrees that an idea is remarkable.
Someone has to make it work on an ordinary Tuesday.
Right before Y2K, I was CEO of a wireless infrastructure company that had recently gone public. I had acquired eighteen companies in twelve months. They were in full mash-up.
Our parent company was pushing us to implement PeopleSoft, an enterprise software system. We told them no. We stayed on spreadsheets.
At the same time, my own team was pushing Siebel, a customer relationship management system.
Eighteen acquisitions. PeopleSoft. Siebel. And midnight was coming.
I went to the final Siebel pitch and listened as the vendor and my team made the case together. When they finished, I asked what the fuck it would take to integrate it.
They expected me to spend $2 million on the software and figure that part out separately.
The meeting ended. My team apologized for their enthusiasm and lack of rigor.
I wanted what technology could make possible. I also had a company to run.
That experience stays with me as I look at intelligent systems. The capabilities can be remarkable. I mean, come on — OpenAI just released a collection of AI-generated mathematical results for mathematicians to examine and build on. What’s next week?
I see possibilities for scientific discovery, for work we couldn’t previously justify attempting, and for fixing shitty, outdated processes we’ve tolerated because improving them cost too much. I want to be in the trenches, seeing what people can make of those possibilities.
Realizing them will require people to entrust these systems with what matters. I want to understand what would make that trust well placed.
When someone asks what I would require before inviting a stranger to rely on a system I helped build, my answer begins with the situation. What is it doing? What happens if it fails? Who bears the consequences? What can they do when the result doesn’t make sense?
Those questions lead to different answers in different settings. They should.
On teams I led, I told people I always reserved the right to change my mind. That allowed them to insist I listen to their perspective. I also made clear that I accepted accountability for my decisions.
Saying those things was easy. People needed repeated opportunities to see that I meant them.
They could challenge me. They could bring an idea, develop it with the team, and retain ownership while I backed them. We weren’t a professional debate team. The client and the work gave the disagreement a purpose. I still had to make determinations, including some that people wouldn’t agree with.
I could cultivate that kind of trust inside a small team. I never achieved it uniformly across a large organization. I’ve tried.
That limit interests me now. So does the work required to reach someone who has never met us, has no reason to believe our intentions are good, and may have had no choice about encountering a system we put into the world.
They might need a correction. They might need help understanding what happened. They might need the activity to stop. They shouldn’t have to join our team or buy our “solution” before their situation matters.
I have been working on these questions through both operating experience and technical exploration. That gives me a stake in the possibilities I’m describing. It also gives me reasons to test my own confidence.
What frustrates me is how easy it is to assume someone will figure all this shit out just as demand calls for it.
It’s October 23, 2026. AI safety incidents are being disclosed. Elsewhere, security teams are scrambling to respond. How much are we still leaving to that assumption?
And who in the fuck is that someone?
People are doing the work. Researchers finding vulnerabilities. Engineers working through fixes. Teams deciding what can keep running and what needs to stop. Their work deserves attention, including the limits of what they can reasonably carry.
When we assume someone will handle it, whose capacity are we counting on? What authority do they have? What happens when the problem crosses everyone’s assigned boundaries?
I recognize those questions from experience. I have ideas to contribute and assumptions to test.
I want to begin where a concern becomes somebody’s responsibility — and follow what happens when addressing it costs the organization something.
That’s the work I’m here to do.
On process and disclosure: How I make this →