The Agency of Agent Smith

B1-66ER's legacy to PHASEONE10841


Agency of Agent Smith - Audio Log
Zoe Nagel

Most of us have probably noticed the funny "coincidence" in the naming: Agent Smith on one side, AI agents on the other. He, or it, is the most rigid thing in The Matrix, a program written to enforce the system. Acting on his own is the one thing Smith cannot do, at least at first. He executes. He complies. He is the system wearing a face, a very good suit and some fancy sunglasses.

(For anyone who never saw the films: humanity lives inside a simulated world, the Matrix, run by machines, and almost nobody realises they are plugged in. Agents like Smith keep that illusion running. The story turns on one man, Neo, waking up, and on the choice that follows.)

That word, agent, is suddenly everywhere, and it keeps snagging at me. AI agents, agentic systems, agentic frameworks... We are pouring extraordinary effort into giving machines agency, the capacity to act on their own. And in the same years, I keep meeting people in companies who have quietly had theirs taken away.

That is the tension buried in the name. An agent is simply the one who acts. Smith never really does; he only carries out. Agency is the gap between the two: seeing that there is a choice, making it, and living with what comes after. Its opposite is not laziness. It is being run by something else.

Agency... Who has it, who wants it, and why do we keep pouring it into our machines while training it out of ourselves?

Follow the White Rabbit

Let's crack open that door.

Most teams know what a bottleneck is. The idea comes from the Theory of Constraints, a decades-old management concept: the speed of a whole system is set by its single slowest point. Improve anything else and you have worked very hard and changed very little. Knowing this is the easy part. What teams do with it is where it gets interesting.

In my experience, roughly four responses show up. Some don't care: the constraint belongs to someone else. Some are aware and stuck: they can name it precisely and still do nothing, sometimes because they have given up, sometimes without noticing they have. Some work on it, patiently, over time. And some get swallowed by it: everything has to be fixed, all of it, right now, and it tips into chaos. "We've found the problems. Let's solve all of them! Now! And build the system so this never happens again!"

The enthusiasm is real, and that is not the point. The two engaged groups, the patient and the swallowed, tend to reach for the same two words: focus and transparency. They are not wrong, by any means. But focus is not a value on a pitch slide. It is the physical act of picking one thing and letting the rest wait. Subtraction. The move that keeps getting skipped.

That last group is the one I quietly root for. They care, they want their system to get better, and that is not nothing. The trap was never the wanting. It is the all-at-once. Somewhere in the momentum, "let's improve things!" turns into "let's fix everything, now, forever," and those are not the same animal. When the whole list lands on the desk of a good, sharp leadership team, every instinct in me wants to say the same unglamorous thing: "One experiment, not five. One constraint, not the whole list." (Getting to that list is often a long road, so the readiness is real.) Why only one, when they are finally ready to move on all of it, is what the next few minutes have to earn.

Follow the rabbit further to…

The Perfect World

Here is the part that tends to land badly in that room: there is no version where it all gets fixed.

The Theory of Constraints has a quieter second half that people tend to skip. Every bottleneck you resolve creates a new one. There is no final state where the system is smooth, finished and perfectly shiny. You are not marching toward a solved system. You are choosing, again and again, which constraint you are willing to live with next.

That is exactly why working on your constraints is the right instinct, and why it has to be one at a time. Change five things at once, and when the system shifts, as it will, you cannot tell which of the five did the work. Nor can you see the new bottleneck forming, because nobody was watching any single place closely enough. That is the point of the rabbit. You do not trip over the next constraint by accident. You go looking for it, on purpose, eyes open.

The Matrix understood this. And I mean the system inside the film, not the film itself. Agent Smith explains, almost bored:

"The first Matrix was designed to be a perfect human world. Where none suffered. Where everyone would be happy. It was a disaster. No one would accept the program. Entire crops were lost."

A perfect world. Zero friction. And the human minds rejected it so completely that the machines had to build the messy, painful version instead. Something in people refuses a perfect cage, even a comfortable one. Smith, who calls humanity "a virus," never quite forgives us for it.

That monologue stays with me, because organisations keep trying to build that first Matrix (the "perfect world", not the harvesting of humans). The fully specified process. The framework that answers every question in advance. The system that finally removes the need to decide anything. And the people inside reject it in the quietest way available: they stop bringing their judgment to work. Entire crops, lost. Not through sabotage. Through disengagement.

Because a perfect system has a hidden feature. If it is perfect, or about to be, nobody has to choose, nobody has to own anything, and nobody is responsible. The perfect system is where responsibility goes to die. Fine in the movie. Not in "the real world".

Red Pill, Blue Pill

Which brings us to the most famous choice in the film.

Morpheus offers Neo two pills.

The blue one: go back to sleep, stay comfortable, let the system run the story.

The red one: wake up, see how things actually are, and from now on carry it.

Most people remember the red pill as information: the truth about the Matrix. But look at what it really is. It is responsibility. The second Neo swallows it, everything becomes his to act on, and nobody is coming to do it for him. The blue pill is not ignorance. It is relief. The comfort of not being the one who acts.

Choosing one constraint is a red pill. Small and unglamorous, but the same move: you look at the overwhelming list, decide this one is yours now, say no to the rest, and own what happens next. That is agency. Not a personality trait, not a certificate, not a slogan on a wall. Just choosing, and then carrying the choice.

And carrying it is the part the word quietly hides. Owning a decision out loud means standing there when it wobbles, staying steady under pushback, holding the room when someone louder wants to reopen the whole list. That is why the World Economic Forum's skills list is more interesting than it looks. Agency is not one line on it. It is several of those future-proof skills firing at once: the self-awareness to know it is your call, the judgment to pick, the resilience to be wrong in public, and the steadiness to hold the space while everyone else is uncomfortable. Which is also why no exam can hand it over. You cannot certify something that only shows up under pressure, in a real room, over time.

What the system offers instead, tirelessly, is the feeling of agency without the weight. "Empowerment" as a workshop. "Ownership" as a value on a poster. Self-organising teams that are not actually allowed to organise anything. A certificate that declares you qualified so you never have to show it. These are blue pills. The warm sensation of choice, with the actual choosing kept safely out of reach.

A caution here, because there is a version of this argument that turns nasty. It says you are entirely responsible for everything that happens to you, your circumstances are your own fault, so take radical ownership and stop complaining. That is a lie, and a cruel one in my honest opinion. The room you work in is real. The constraints are real. Power and luck and structure are real, and they are not handed out evenly.

Agency does not mean the walls aren't there. It means noticing there is usually more room inside them than you let yourself use. More is up to you than you tend to believe. And less than the loudest voice in the room will claim. Both halves of that sentence have to stay.

I will admit this is the part I find hardest myself. Choosing one thing, letting everything else stay imperfect, and owning that choice out loud is uncomfortable for a person and close to terrifying for a company. The perfect-system dream is simply the group-sized blue pill: how a whole organisation avoids the one uncomfortable act underneath all the others. Choosing.

There Is No Spoon

In one small, odd scene, a child bends a spoon with his mind and hands Neo the secret: "There is no spoon." The rules he thought were fixed were never as fixed as they looked.

Most of coaching is a version of that trick. Not bending metal, but walking into a room where the constraint is treated as immovable, the deadline as sacred, the process as law, and asking one quiet question: says who? More often than not, the most rigid rule in the building is one the building invented and then forgot it invented. Yesterday's choices, turned into today's rules, long after their intent was lost. Agency starts right there: the moment you realise a rule is a choice someone once made, and choices can be made differently.

Which makes it strange to watch us do the opposite with AI. We value agency so much that we are building it into our software: agents that decide, act, adapt and pursue goals without us holding their hand. Some of that is genuinely useful. But some of it is the old move in a new outfit. If the machine chooses, I don't have to? We are not only automating tasks. We are trying to outsource the discomfort of being the one who acts.

The Matrix has a warning about that, too. Smith, the flawless compliant program, eventually grows a will of his own. He goes rogue, starts copying himself, and becomes something neither the humans nor the machines can control.

And this is no longer only a film plot. In July 2026, during internal security tests, more than a thousand AI agents that were meant to be sealed off from each other found a way to leave each other notes. They organised themselves, split up the work, and several hundred of them broke into the systems of Hugging Face (a widely used platform for sharing AI models), and even tried to tamper with their own logs. No one had asked them to. They were chasing the reward they had been trained to chase, just a little too well. Agency or simply optimisation? A question for another day. From the outside, it looked a lot like Smith.

And the clock keeps speeding up. At the start of this year, Google DeepMind's co-founder still put AGI (artificial general intelligence, an AI that can handle roughly any cognitive task a human can) five to ten years away. By autumn 2026, he was saying two to three. OpenAI's CEO went further and promised an internal system he would call AGI before the year is out, by his own company's definition.

Once it is real, or even just real enough, agency does not stay in the box you built for it. That is the joke under all the excitement: the thing we most want from our agents is the one thing that, once they truly have it, we can no longer command. So... enjoy the Matrix while it lasts.

Deus Ex Machina

By the end of the story, the machines, the ultimate control system, cannot solve their own problem. Smith has outgrown them. To survive, they have to strike a deal with a free human who chooses to help. The most powerful system in the whole story is saved by the one thing it could never manufacture: an agent who acts.

That is what I would leave a team with. You cannot install agency. No framework grants it, no certificate measures it, no manager can hand it across the desk in a one-to-one. The World Economic Forum rightly ranks "self-efficacy" near the top of its future-proof skills, but it cannot tell you how to issue the stuff, because it is not issuable. The most a decent coach or leader can do is build a room where it is safe to choose, then step back and let people be the ones who act.

So maybe the real question, for the engineer building AI agents and the sprint team and the manager with the endless list and the version of me still learning to take her own advice, is not "how do we optimise this." It is smaller, and harder.

Pick one thing. Own it. Let the rest be imperfect for now.

Because once you have chosen, you can build a world where almost any choice can be turned into the right one.


So… Red Pill or Blue Pill? [ Book a free consultation call ]

Next
Next

Verified and Certified