You can read more and follow David Bethoney’s newsletter at The Revive.
The dopamine hit you didn't know you needed until AI took it.
Hi friends, I have a special article this week.
A couple of months ago, I mentioned I wanted to start introducing other voices to The Revive in the form of guest posts. Throughout my career, I’ve met a lot of smart, thoughtful operators who are brilliant at their craft. I’ve started to reach out to many of them in an attempt to bring you a guest post a month.
First up is Jason Uechi. Jason and I met at the Muse. He joined a few years after me as CTO of the Muse. He’s one of those rare engineer leaders who’s both deeply technical and genuinely fun to be around. He’s the kind of person who pushes everyone around him to think bigger and longer-term. Since The Muse, Jason has gone on to co-found Sunrise.AI, where he serves as CTO. I’ve been looking forward to bringing his voice to The Revive for a while now.
Hope you enjoy this and let’s show Jason some love by dropping a like and sharing this piece for him!
The 100% Screen
Developers will tell you they love solving problems. What we rarely admit — and what I suspect we don’t always fully know — is that we love solving problems the way a completionist loves a video game: only if there’s a 100% screen waiting at the end. The rest of it, we merely endure.
You know the 100% screen: the moment at the end of the level, Mario hitting the flagpole, the little jingle, the number ticking up to a perfect 100. That’s the hit we’re really after.
I got hooked on the 100% screen the usual way: staring at a small piece of code for too long, changing one thing, then another, making it worse in stupid, unhinged ways, until — suddenly — it works. The build compiles. The data flows. I won this mission. I lean back and think, with a satisfied grin wildly disproportionate to the tiny change I just made: F**k yeah, I did that! (My inner voice has a potty mouth).
The work didn’t just get done; it confirmed that I did it.
There’s something deeper in problem-solving than merely finishing. It’s the feeling that effort turned into something tangible. That the time and the mumble cursing at the screen wasn’t futile. You don’t get that feeling from outcomes alone. You get it from watching yourself get there.
Working in software development is a similar experience to most knowledge work. Take Finance: The spreadsheet reconciles, the document gets sent, the contract is signed. When we complete a task, we feel a certain dignity. Closure isn’t a perk of good work, it’s part of what makes digital work feel like it happened at all.
And then the asteroid hits.
AI arrives carrying confident language and the subtle menace of a forklift in a pottery store, driven by drunken capital looking for efficiencies to automate. And the cruel irony is that the work easiest to automate, the work we sometimes get the most joy from, is the work with the clearest structure. Clean inputs, clean outputs, a sharp notion of done is AI’s strong suit, robbing task-oriented humans of this simple joy.
So here we are having a ton of fun with AI, wildly productive in dimensions you didn’t expect, look what I can do! But it feels off in a way that’s hard to name. It’s not just that tasks you spent years getting good at are getting automated. It’s that one of the main mechanisms by which we experience progress — start, solve, finish — is getting interrupted.
The good news is that your workday was never just one thing. It’s a pile of tasks of which many of them correlated and complicated, dependent on decisions and human context that can be, sometimes or maybe all the time, a genuine mess.
And messy work has always existed: tasks with no test suite and no answer key. You’re managing a team, or spending the day interpreting vague signals with three equally plausible explanations. Multiple right answers available for every decision; Work where you can do genuinely good things and still not be entirely sure what outcome, exactly, you own.
We used to label that job change from deterministic tasks to managerial ambiguity career growth. The deal was you pay your dues in bounded, checkable work, build real competence, figure out what the organization actually values; then, you graduate up and into the murk. Some people wanted that. Understandably, a lot of people didn’t — it’s a rough trade. You give up the tight feedback loops and the frequent confirmations, and in return, you get a fancier title and a fuzzier set of responsibilities, with no checkmark at the end. That was a legitimate path to opt out of, right up until AI made opting out no longer an option.
AI isn’t inventing ambiguity. It’s expanding it by shrinking concrete, task-oriented roles. It’s eating into the territory where someone could build an entire career in work without ever having to live all day in the murk. The 100% screen is harder to find. Mario’s flagpole extends up into the clouds, and we can no longer see the top
And this process also exposes something the world of work has been getting away with forever.
A remarkable amount of what makes someone effective in ambiguous work is never written down, or even implicitly agreed upon. How decisions actually get made. What “good enough” means, specifically. Where to take risks and where the organization quietly — or not so quietly — pushes back. We call these unwritten rules, culture. In many organizations, reading and understanding a work culture happens through osmosis. When people spend years doing tasks side by side, they invariably learn the unwritten rules of the culture through observing their colleagues, conversations over coffee breaks, and passive-aggressive Slacks.
AI interferes with the time during which humans normally acculturate in a new workplace. You speedrun through task competency and arrive at a starting line where your value is embracing ambiguity as the rule, not the exception, all before you’ve had time to absorb the tacit knowledge that makes ambiguity survivable.
Which means it has to be made legible. What does good judgment look like here? How do we make decisions when the signal is soft? What does “done” mean when there’s no checkmark? Spell it out. Early. Repeatedly. The same clarity that makes ambiguous work possible to do is the same clarity that makes it possible to delegate — to a new hire, or to a system. If it only works as tribal knowledge, it doesn’t really work. It just seems to, until you need to transfer it.
I still like solving problems. I like finishing them. Passing tests, the merge is clean, the moment the system quietly agrees: yes, this is correct. That hasn’t changed.
What’s changed is how much of my work actually behaves that way. More of it now lives in the space where progress is real but hard to point to. The sense of completion is deferred and harder to identify by the time I shut my laptop at night. My 100% screen is still out there. It’s just further up the flagpole than it used to be.
