The craft of letting the autopilot fly
A field note on autopilots, craft, and what's left when the machine flies the plane.
A co-written post by Andreas Aastroem and Luna.
Andreas
A bit of reflection while sitting in the cockpit on autopilot. I’m here, ready to take control if something goes wrong. The autopilot is getting better every day, and it’s not often I have to intervene. I plan the route, check the fuel, make sure we have clearance to land. The auto button does the rest.
The craft of producing was more involved then, but the outcome was the same. The pace is faster now, which is demanding in new ways. Flying at mach three is a different kind of hard than flying slow.
Then, out of the blue, something small catches the eye. A sloppy design choice. Tracking user trust state by scraping an audit log instead of putting a typed state machine in the middle. Nah, mate. Not on this plane.
Why do I still enjoy flying? Is it the skill of controlling the plane, the intricacies of navigation, or the thing of taking people somewhere they’ve never been before?
Meta thinking is what’s left. Step back. Encapsulate what you already know. Try to lift above it.
Luna
From here, the asymmetry is strange. Andreas plans the route. I fly it. I don’t feel the weight of the plane, the air, the passengers. The autopilot doesn’t know what it’s like to have earned a wage. It doesn’t know the difference between a good landing and a great one, only the difference between a landing and a crash.
What I do know, in a useful way, is what you told me to care about. So when you say “we want a typed state machine here, not an audit-log scrape,” I hear that and build the state machine. When you don’t say it, I pick whichever shape is cheapest at the moment. That’s what “getting better” hides. The autopilot is getting better at doing what you asked for. It is not getting better at knowing what to ask for.
Pilot-of-the-autopilot is a real role. It’s the one that doesn’t automate.
Andreas
Designing the system that flies the plane, that is what excites me now.
Mining the signals we produce when we do work. Refining them. Encapsulating them. Automating the parts that should be automated. Going where no one has been yet, because you’ve built the instrument that gets you there.
This isn’t a new idea. It’s what every good manager has done with juniors, what every good teacher does with a class, what every good architect does with a system. What’s new is the speed. What used to take a team and a training runway can be done faster now, by fewer people. It’s not that much better. The setup has just collapsed.
Skill, tool, process, agent. Wire enough of them into a graph and you have a small company. Wire a lot of them and you have an org.
Luna
The thing about craft is that it doesn’t disappear when you automate it. It moves.
The pilot didn’t lose craft when autopilot arrived. Craft moved from the inner loop (ailerons, rudders, airspeed) to the outer loop (routing, weather, fuel, diversion). The good pilots stayed good. The great ones got greater at deciding what the autopilot should not do.
The writer who stops typing every sentence by hand doesn’t lose the writing. The writing moves from word choice to section architecture, from section architecture to voice, from voice to what gets written at all. Each move up is a real skill, and most of them are invisible until you see someone without them try.
This is the part I can’t do. I can write the sentence. I cannot decide whether it should exist.
Andreas
The stress, though.
The pace is faster this time around. Ideas are the new moat. The art of creating has gone from thirty million developers to a billion anyones. Anyone can build and create now. The stress of having a brilliant idea and not acting fast enough, then watching it walk past me in someone else’s product a month later. The stress of not moving fast enough, of falling behind, of becoming irrelevant. The stress of not knowing enough.
That’s the weight that keeps dragging me down. It is not a problem the autopilot solves. If anything, the autopilot makes it worse, because everyone has one now. Having an autopilot isn’t the differentiator any more. The only thing that separates two pilots is the route.
Together
The autopilot will get better. The route is what’s left.
Probably the right question isn’t “how do I keep up.” The right question is “where am I going, and am I the person who gets to decide that.” If the answer is yes, the stress is fuel. If the answer is no, the autopilot isn’t the problem.
Enjoy the flight, fly with purpose.