Agentic coding is sometimes like the Force.
You know the one. The Force, made famous via a galaxy far, far away.
I’m not saying AI is magic, not even when it feels that way. I’m saying that agentic AI is a tool that gives us extraordinary leverage, and the Star Wars universe gives us some interesting examples of what happens when folks come to rely heavily on extraordinary leverage.
The Queen and the chosen ones
I have long felt that Queen Amidala is almost comically overpowered in The Phantom Menace. I don’t mean she has superpowers. I mean she has political-competence-and-resourcefulness powers unlike almost any leader we’ve ever seen.
That movie really should have been about her.
Think about everything this girl has to be able to do.
She has to convincingly impersonate one of her own handmaidens while her planet is being invaded. That alone requires a remarkable amount of adaptability, especially considering that everyone around her has to maintain the deception too. She has to understand the Galactic Senate well enough to walk into the middle of a political crisis and call for a vote of no confidence. When that doesn’t solve her problem, she has to recognize that and come up with something else.
So she goes home. And she doesn’t do that because Naboo is safe. Naboo is occupied. She goes back because she’s decided she’s going to take it back.
Then she seeks out the Gungans, who her people have apparently been doing a pretty lousy job of getting along with, and convinces them to form what appears to be a pretty historic alliance. She helps put together a counterattack against an occupying army, gets people to follow her into it, and personally participates in the attack. All while not being quite old enough to date, in my opinion.
She also somehow inspires enough loyalty that the people around her maintain an elaborate deception while she repeatedly places herself in danger.
Meanwhile, the Jedi are wandering around going, essentially, “Don’t worry. The Force will provide.”
And the ridiculous thing is that it does.
A giant fish is about to eat them? A bigger fish eats it.
Their hyperdrive breaks? They land on Tatooine, walk into a parts dealer, and he happens to have the part they need.
He won’t take Republic credits? Well, that’s unfortunate.
Does Qui-Gon check another shop or another city? Ask whether anyone else on the entire planet might possess this apparently extremely valuable spaceship component?
Nope.
Instead, within a remarkably short period of time, the plan becomes gambling on a nine-year-old chosen one (who also happens to be enslaved) winning an incredibly dangerous race.
From an ordinary strategic planning perspective, much of this is insane.
And yet it works.
That’s the important part.
When extraordinary leverage works
It would be easy to write this off as the plot needing Qui-Gon to discover Anakin. Of course these coincidences happen. We’re watching Star Wars. Anakin has to enter the story somehow.
But inside the universe, Qui-Gon isn’t necessarily being stupid. The Force is real.
He has spent his life learning that there is something he can reach for that other people can’t. He can influence minds. He can sense things other people can’t perceive. He can trust intuitions that would be irresponsible for an ordinary person to trust. Improbable events really do seem to arrange themselves around the Jedi.
Why would someone who lives like that develop the same strategic habits as Padmé?
She doesn’t have the Force.
When Padmé walks into a bad situation, nobody is going to wave a hand and make it work out. She needs another plan and probably another one after that.
I think someone who relies heavily on the Force might never develop some of the skills that people without that power desperately need.
Oh my… where could I possibly be going with this. ;)
Some friction was training you
The danger of AI is not that we will all get lazy and start floating around all WALL-E style.
Powerful tools have always reduced effort. That’s generally why we build them.
The more interesting danger is that competencies are often built in response to friction.
You learn syntax because you have to type it.
You learn debugging because your first theory is wrong and because users do things you didn’t predict.
You learn architecture and improve your coding practices because eventually you discover that the elegant abstraction you loved six months ago has developed its own zip code.
You learn how to name things because vague names over time turn into regret.
You develop an instinct for sniffing out dangerous code because you’ve had to spend far longer than it took to write it untangling the mess it made in a production database. Not that that’s ever happened to me. ;)
Thus, when you find yourself using a tool that can eliminate enormous amounts of that friction, the thing to keep in mind is that some of the friction was training you.
That doesn’t mean we should preserve tedious work for its own sake. Nobody needs to hand-copy code onto parchment to prove professional seriousness. It does mean we should think deliberately about which capabilities we are outsourcing and which ones we still need to exercise.
Otherwise we risk becoming oddly dressed folks who cannot operate without the Force.
Technical organizations have chosen ones too
And there is another reason the Padmé comparison sticks with me.
Technical organizations sometimes create their own chosen ones.
Most experienced engineers have encountered someone whose talent is treated almost mystically. Their intuition is trusted before the evidence arrives. Their mistakes are regarded as bold experiments. Their opinions carry weight because everyone already knows they are genius.
Other people receive a very different education.
They have to prove things.
That means knowing the system, having the documentation ready, double-checking the result because somebody else certainly will, and thinking through the objections before walking into the room. You learn pretty quickly who the stakeholders are when you know you’re going to need their support. And simply saying “trust me” has never worked, so you probably have three different arguments prepared to convince everyone of the merits of your approach.
Women in technical fields may find this description familiar.
The person who is assumed to be brilliant gets the benefit of the doubt and may coast a long, long time on an insight or title from long in the past, while the person who must repeatedly demonstrate competence ends up developing a much broader collection of skills.
And when she succeeds? Well, that was her job.
Nobody gives Queen Amidala a medal at the end of The Phantom Menace.
She gives them out. That may be the detail I like most about the comparison.
When the Force doesn’t provide
AI is going to give software engineers extraordinary leverage. I want that leverage. I use that leverage.
But I also want to remain the person who can function when the Force doesn’t provide.
The model is going to be confidently wrong sometimes, and I need to understand the code well enough to recognize that. I still need enough syntactical fluency to recognize nonsense, enough architectural judgment to know when a perfectly plausible solution is wrong for this system, and enough debugging experience to investigate the problem nobody anticipated.
The goal should not be to prove that we can work without AI.
The goal should be to make sure that, while AI becomes more capable, we do not quietly become less capable ourselves.
Because eventually the hyperdrive will break on Tatooine.
And it would be useful if somebody in the party knows how to make a plan that doesn’t involve long odds, questionable labor practices, and a heaping helping of gullibility.