An agent walks into a bar…

It’s not hard to imagine; the first attendee at your next event might not be a person.

It’ll be their agent. Sent on ahead to scout. It’ll read the agenda, filter the sessions, book the meetings, decide what’s worth their ‘human time.’ The person arrives later, pre-briefed on a tidy subset that some blend of their software and someone else’s decided they’d like. Or the person never comes at all. The agent attends on their behalf, files a summary, and moves on.

My time, like yours, is valuable. So on the surface, this sounds pretty great. No registration line at 8am. No scrolling forty concurrent sessions trying to guess which one won’t waste an hour you’ll never get back. This is already happening.

Concierge agents are a frequent starting point. They aren’t just inevitable. They’re necessary. This isn’t some tired industry idea we’ve kicked around for a decade finally shipping. It’s a real learning step. It’s how the system starts to understand what people actually need and, more importantly, how you prove you can predict it accurately. And you don’t get to the non-obvious without first mastering the obvious. So take them. Hand over the logistics. That part is good. It’s the foundation everything else gets built on.

It’s also the least interesting part of the story.

The interesting question is simple. It’s the one that decides whether this ends somewhere wonderful or somewhere hollow. What is the agent optimizing for?

When instructed well, agents are very good at their job. It removes friction. It gets you exactly what you asked for, filters out exactly what you didn’t, and hands you a clean, efficient, relevant agenda with no wasted steps. That’s the whole promise. Efficiency. Relevance. No wrong turns. Yawn.

Here’s the part that sucks. Efficiency isn’t the point. It should not be the goal. It’s the on-ramp. Pulling time out of the system doesn’t matter. What you make time for is the entire game. And the best thing that ever happened to you at an event was a wrong turn.

Speaking for myself, the best experiences I’ve ever had at an event started like this. Is this seat taken? Have you ever heard of…? Can I show you this thing I’ve been building? This session’s full…what’s going on in that room? I’ve never seen that name before…where does it come from?

None of that was on the optimized agenda. It was the person in the coffee line. It was the hallway conversation that ran so long you missed the keynote you flew in for. And it turned out to be the most important thing you carried home. It was the talk you slipped into because your first choice was full. None of it was relevant. None of it was efficient. None of it is anything an agent would ever have picked for you. Because you didn’t ask for it. And an agent’s entire reason for being is to give you what you asked for.

I’m far from anti-agent. But efficiency is never going to be my key metric. It’s an important byproduct of the real metrics.

For 30 years, the whole craft of experience design has been a war on friction. Faster check-in. Smarter matchmaking. Personalized agendas. Recommendation engines. We sanded down every rough edge and called it progress. And mostly it was. And it’s still a focus. But there’s always been one thing that’s kept it honest. The technology had a ceiling. Designed poorly, the technology was always visible. The bolt-on. The shiny thing. The tech that announced itself and got in the way. The best design was invisible. It elevated the moment. It moved the story forward. Either way, the machines’ limits capped how far off course a bad decision could take you. The tech needed the human in the loop. Its decision making power was limited by the technology, by the data, by our ability to predict. So the human stayed in the loop.

Agents remove that ceiling. That’s the shift. Today’s distinction. The klaxon going off in the night. This isn’t the next stop on the same road. It’s the guardrail quietly falling away.

Every tech implementation until now curated. It surfaced. It ranked. It suggested. And then a human walked through the door and chose. The agent doesn’t curate the room; it decides whether you walk in at all. It doesn’t shorten the menu; it eats the meal and hands you the review. That’s not a step change. Curation assists a choice. Dictation replaces it. The machine stops choosing with you and starts choosing for you.

The is an unprecedented shift. And when the technology is not the limiting factor, it has to be the metric. The technology used to be the thing that saved us from our own bad metrics. Now the metrics are all that’s left.

Play it out and you can see where it goes. A perfectly frictionless event isn’t an experience. It’s a vending machine. You put in your preferences (or they’re inferred), you get exactly what you selected, and nothing you didn’t choose ever reaches you. Ok, maybe an event is a fancy gum-ball machine, the gum-ball rolling down its little looping track. It’s fun to watch on the way down. It takes you on a few loops and turns. Maybe there’s an ooh or an ahh. But in the end, you still get the gum-ball you ordered. And that’s just not the point. The little things you didn’t order…that’s where all the magic hides.

There are two events now. There’s the one the agent experiences. Data, logistics, a clean summary of sessions and leads. And there’s the one the human experiences. The room. The collision. The person they never planned to meet. The thing they felt. Build for the first and the agent will rate you five stars. Frictionless. Relevant. Efficient. And the human will have forgotten the whole thing by Friday.

So the job flips. It has to. The work stops being remove friction and becomes choose the right thing to optimize for.

This sounds easy until you try to name it. What’s the metric? Meaningful moments? Smiles? Memories? Relationships? Those aren’t things a machine understands. And that’s what makes this hard. Efficiency is easy because efficiency is quantifiable. You can count it. Knowing what conditions to set, knowing where to get out of the way, knowing where the human should be left to be human…that resists the Clean, clear number. So we reach for proxies. And the proxy is the trap.

We’ve run this experiment already. Social media is a warning. The goal was connection…at least that’s what the stated goal was in the beginning. The proxy metrics were follows, likes, views, comments, engagement. These don’t measure ‘desirable’ engagement. These are measurable stand-ins, because human connection itself was too hard to count. And then it was further complicated by the desired outcomes of the user vs. the desired outcomes of the platforms. The proxy won. It always does when its the only solid measure (especially when it also correlates with positive business outcomes). We optimized for engagement and got outrage and division in the name of relevance and connection. The precise opposite of the stated goal, delivered with ruthless efficiency by machines doing exactly what we measured. Point a powerful optimizer at a proxy and it will hand you the proxy and burn the real thing for fuel.

That’s the warning under all of this. Not that the agent is evil. The agent is obedient. Ruthless, and obedient. Point it at efficiency and it will hollow the experience out with a smile. Point that same ruthlessness at the right thing and it becomes the most powerful tool for connection we’ve ever built. One that manufactures the wrong turn. One that makes the happenstance more likely. One that designs the chaos so it works for you instead of against you. Same relentless engine. Aimed somewhere worth going.

But choosing the metrics is only half the story. This only works if the aiming happens in the open…and we as the target of the agents’ efforts have some control over the agents’ instructions. And we can’t afford to get this wrong. These decisions can’t live in a black box. What the agent optimizes for, what it’s allowed to filter out, what it quietly decides you don’t need to see. All of it has to be transparent. And there has to be some level of control given to the individual. And it has to keep evolving, out loud, where we can see it and argue with it. The moment it slips into the box, we’re right back in the social media trap, optimizing a proxy nobody’s watching until the damage is already done. And the true threat here is that the damage could push much further in a much shorter period of time than anything we’ve ever seen before now.

So no. I don’t think the ultimate goal as we design the agent experience is to make the machine disappear. That’s the old reflex, and it’s too easy. The machine can escort me to the door. It can carry the badge, the logistics, the follow-ups. It can even follow me around and jump in the second I need it. What it cannot do is pick the seat, eat the meal, and hand me the summary of a life I was supposed to live. Give it a human reason to care, in the open, and it stops standing between me and the room. And it starts building the room. And the only way to get that right is to set the metric, make the metric transparent, build rapid iteration into the experience…and ultimately build human feedback in and treat it like the priority metric that it is.

Then it can do the best thing a machine has ever done for us…amplify our humanity.

Get me to the door. Come in with me. And help me get blissfully, inefficiently lost.

That’s where all the best moments have always been.