Daniel's asking about the shape of a career. Entry level, you're a production resource, worked hard, learning by volume. Mid-career, management starts eating your hours. Late career, you're mostly teaching, reviewing, doing continuing professional development. The best people end up doing the least actual doing. He thinks that's backwards. His question is whether there's a Model C, something that isn't doing until you burn out, and isn't learning until you've forgotten how to do. A stable mixture across the whole arc.
The first thing I'd point out is that the premise holds up empirically. A time-use survey of academic researchers in Japan found research time declines over the life cycle. It's fewer total hours plus more administration. And a separate survey found administrative duties jump right after tenure.
So the organization waits until you've proven you can do the work, then stops letting you do it.
That's the shape of it. And it's not just academia. The dual ladder in tech exists because companies kept promoting their best engineers into management and watched them either leave or become mediocre managers. Senior is the last shared level. After that you pick, individual contributor or manager.
Which is an institutional admission that the default arc was broken. They built a parallel track so people could keep doing.
But even the IC track has the same problem Daniel's pointing at. Staff engineer, principal engineer, distinguished engineer. The higher you go, the more of the job is review, architecture decisions, mentoring, writing design docs. You're still not writing much code.
The percentage of time actually executing shrinks either way. Management just makes it shrink faster.
Right. And Daniel's contention, that it's a pity we leave CPD for the middle and end, I think is half right. The pity is that we leave formal CPD for later. But the assumption that early career is mostly doing and later career is mostly learning gets the actual science backwards.
How so?
Deliberate practice. Ericsson's original work. Recognized experts do about four hours a day for ten to fifteen years. And the crucial detail is that deliberate practice is rehearsal of specific tasks to mastery, ideally with a mentor giving feedback. It's embedded in work. It's not a separate activity you schedule after you've done your real job.
So early career should be the most learning-intensive period, not the least.
The novice surgeon isn't just doing more surgeries. They're doing them with someone watching, correcting grip, pointing out where their gaze should be. That's learning and doing fused. Then somewhere along the way we split them. Doing becomes production, learning becomes a course you take in a conference room.
And the evidence for the conference room model is thin.
A rapid evidence synthesis on CPD found limited evidence that accumulating hours of educational activity enhances practice. There's been a global shift away from counting hours across professions. The UK minimum hours for re-registration have been small compared to other countries. The whole model of CPD as a thing you bolt on later is under attack.
Which guts Model B. Spend as much time learning best practices as possible. If the hours don't transfer, you're just accumulating certificates.
And Model A, doing as much as possible indefinitely, has its own empirical problem. The laparoscopic surgery study. Trained novices reached over ninety-eight percent of the behavioral learning plateau. They matched surgeons on the physical movements. But the surgeons still had dramatically more efficient gaze patterns. The doing alone never produced expert-level cognition.
So the novices could move their hands like surgeons but not see like surgeons.
That's the finding. Expert gaze, knowing where to look before you need to look there, that only came from years of embedded practice with feedback. Pure volume of doing didn't get them there.
Which means both models fail for the same reason. They treat learning and doing as separate reservoirs. You empty one while filling the other.
And Daniel's Model C, I think, is really asking whether there's a way to keep them in the same reservoir. A stable mixture.
Let me take a swing at what that looks like. Because I think the answer is already partially visible in how the best people actually work.
Go on.
The best senior engineer I ever watched work, and I'm speaking from observation here, not from having done it. But the thing that struck me was that he never stopped doing. He just changed what doing meant.
Say more.
He'd pick up a ticket that nobody else wanted. Some gnarly concurrency bug. And he'd fix it. But while he fixed it, he'd write down what he was seeing. Then he'd turn that into a short document. Then he'd walk a junior engineer through the document while they fixed a related bug. Then he'd go back to his own bug.
So the doing, the learning, and the teaching were all one activity.
They weren't phases. He didn't do for eight hours, then learn for one, then teach for one. The bug was the medium for all three. And I think that's the seed of Model C. The mixture isn't a schedule. It's a posture.
The deliberate practice literature supports exactly that. The mentor feedback loop is the learning mechanism, and it's happening during task execution. Ericsson was never describing a separate study period. He was describing how the doing itself gets structured.
So Model C might be: always do real work, but structure the real work so that feedback, reflection, and teaching are happening inside it. Not after it.
And the stable equilibrium Daniel's looking for, the thing that doesn't have to be a reaction to overreach, is probably something like a fixed ratio of feedback to output. Not time spent learning versus time spent doing, but how much of your doing is observed, corrected, and explained.
That's a much more useful metric than CPD hours.
It is. And it's one organizations could actually design for. Instead of asking how many training hours someone logged, ask how many of their work products got reviewed by someone better than them. How many times they reviewed someone else's work and had to articulate why.
The articulation part matters more than people think. Explaining why you did something forces a different kind of understanding than just doing it.
There's a whole literature on that. Self-explanation effects. Teaching as learning. But the point for Model C is that the teaching isn't a separate phase either. It's part of the doing.
Let me push on something though. Daniel said the best people can spend the least time doing and still execute as well as possible. Is that actually true? Or is it survivorship bias?
What do you mean?
We see the senior person who spends ten hours a week on real work and produces excellent output. We conclude that ten hours is enough. But we're not seeing the twenty years of four-hour-a-day deliberate practice that made those ten hours possible. The compressed execution is downstream of the accumulated reps.
The expert's efficiency is a lagging indicator. You can't reverse-engineer it into a schedule for a junior person. If a junior only did ten hours of real work a week, they'd never build the base.
So the arc Daniel's describing, where doing time diminishes as competence rises, might actually be fine. The problem isn't that seniors do less. It's that juniors aren't getting the right kind of doing.
And that the seniors' freed-up time is being spent on the wrong things. Administration, meetings, CPD hours that don't transfer. If seniors spent their non-doing time on structured feedback loops, embedded teaching, deliberate practice with juniors, the whole system would look different.
The Japanese study found research time declines because of administrative burden. Not because researchers chose to spend more time mentoring. The administration is the waste product.
Right. And the dual ladder was supposed to fix this. Keep the best technical people doing technical work. But what actually happens on the IC track is that the doing gets replaced by review and architecture and design docs. Which is still not doing.
It's meta-doing.
I like that. And the question is whether -doing can be structured as deliberate practice or whether it just becomes a different kind of administration.
I think the difference is whether there's feedback. A design doc that gets read by three people who comment on it, that's a form of deliberate practice. You're rehearsing architectural judgment, getting corrected, refining. A design doc that gets written and filed and never discussed, that's just paperwork.
So the variable isn't the activity. It's whether the activity has a feedback loop attached.
Which brings us to what Model C would actually look like in practice. And I want to be concrete here.
Please.
I think it's a ratio. Something like: for every unit of solo execution, there's a unit of observed execution, and a unit of explaining execution. The solo work builds fluency. The observed work builds correctness. The explaining builds depth.
And the ratio stays roughly constant across the whole career. What changes is the content of the execution, not the proportion of time spent on each mode.
The junior surgeon does solo reps on a simulator, observed reps on patients, and explains cases to their attending. The senior surgeon does solo reps on complex cases, observed reps when they're teaching a fellow, and explains their reasoning at morbidity and mortality conferences.
The ratio holds. The stakes change.
And the burnout risk drops because you're not doing any one mode to exhaustion. Model A fails because it's all solo execution. Model B fails because it's all observation and explanation with no execution. Model C is the mixture.
There's a name for this in the medical education world, actually. It's called the learning practice. Or sometimes the teaching practice. The idea that a practice that teaches is a practice that learns.
Which sounds like a slogan but maps onto something real. The practices that have residents and fellows, the surgeons who teach, they tend to have better outcomes. Not because teaching is magic. Because teaching forces the articulation and feedback loops that keep the expert sharp.
And the evidence on CPD effectiveness actually supports this indirectly. The interventions that work aren't passive hours. They're interactive, case-based, with feedback. Simulation with debriefing. The closer the CPD looks like embedded practice, the better it transfers.
So the shift away from counting hours is really a shift toward counting feedback loops.
That's the throughline. And I think that's the answer to Daniel's question. Model C isn't a new invention. It's a reframing of what we already know works. Deliberate practice embedded in work. Feedback-rich execution. Teaching as a form of learning. The failure is that we keep trying to separate the components and schedule them.
And then we wonder why the separation produces burnout on one end and irrelevance on the other.
The robotics engineer on Hacker News who said he spent six months doing nothing but documents, presentations, and coaching junior devs. He identified the drift away from doing as a burnout contributor. Not the workload. The drift.
The absence of the thing he got good at doing.
Right. And the people asking to be down-leveled, the ones admitting they regret the manager track. They're not asking for less work. They're asking to do again.
Which is a pretty damning indictment of the standard arc. We take the people who are best at doing and promote them out of doing.
The dual ladder was supposed to fix that. But it only fixed the management part. It didn't fix the -doing creep on the IC track.
So what would a real Model C organization look like? Let's design it.
Okay. First, you'd hire for the ratio, not the role. Every job description specifies the proportion of solo execution, observed execution, and explanation. And those proportions don't change with seniority. What changes is the complexity of the execution.
Second, you'd make feedback loops a first-class work product. Reviewing someone's work counts as real work.
Third, you'd kill the CPD hour requirement and replace it with a feedback requirement. You have to show evidence that your work was observed and corrected by someone better than you, and that you observed and corrected someone else.
That would eliminate the conference room course industry overnight.
It would. And the evidence says that's probably fine. The accumulation of hours doesn't enhance practice. Feedback does.
Fourth, you'd rotate the modes. Nobody does solo execution for six months straight. Nobody does pure review for six months straight. The mixture is enforced at the level of the week, not the career phase.
The deliberate practice dose is about four hours a day. That's probably the ceiling for the observed execution mode. You can't do more than that sustainably. The rest of the day is solo execution and explanation.
So a senior person might do two hours of observed execution, two hours of explanation, and four hours of solo work. A junior person might do four hours of observed execution, one hour of explanation, and three hours of solo work.
The ratio shifts slightly with seniority, but nobody ever drops to zero on any mode.
And the equilibrium Daniel's looking for is that the ratio is stable enough that you never have to react to overreach. You don't burn out from doing too much and then swing to learning too much. The mixture is just always there.
The thing I keep coming back to is that the evidence for this is hiding in plain sight. Deliberate practice was always embedded. The CPD literature has been moving away from hours for years. The dual ladder was a structural fix for the same problem. The practitioners are voting with their feet. All the pieces exist.
What doesn't exist is a name for it. Daniel searched for a Model C framework and there isn't one. Nobody's put the pieces together into a single stable model.
Which is actually kind of exciting. It means the question is open. And I think the answer is the feedback ratio. The proportion of your doing that is observed, corrected, and explained.
Let me stress-test that. What breaks if you try to implement it?
The obvious failure is that feedback-rich work is slower. If every work product has to be observed and discussed, throughput drops.
Which is why Model A exists. Pure doing is fast. Until it isn't.
Right. The burnout is the debt coming due. And the other failure is that feedback requires someone better than you. What happens at the top? The distinguished engineer has no one to observe their work.
Then they observe each other. Peer feedback. The surgical morbidity and mortality conference is exactly that. Experts explaining their mistakes to other experts.
And the explanation mode still works at the top. Teaching is the feedback loop that doesn't require a teacher. The act of articulating your reasoning to a junior is itself the corrective.
So the model doesn't break at the top. It just changes the source of the feedback.
The other thing that breaks is the incentive structure. Organizations reward visible output. A person who spends two hours a day reviewing someone else's work looks less productive than someone who spends eight hours producing.
Until the reviewed work stops having to be redone. Then the reviewer looks like a force multiplier.
But that's a lagging indicator. And most organizations optimize for the leading one.
Which is why the dual ladder exists. It's an attempt to create a promotion path for people whose value is in the feedback loops, not just the output.
And the dual ladder fails when the feedback loops get replaced by administrative -doing. The design doc that never gets discussed. The architecture review that's really just a status meeting.
So the implementation problem is keeping the -doing honest. Making sure it's actually feedback and not just paperwork.
I think the test is simple. Does the activity change what someone would do next? If a design review happens and nobody changes anything, it was paperwork. If the junior engineer rewrites their code after the review, it was feedback.
That's a clean line. Feedback changes behavior. Paperwork changes nothing.
And CPD hours change nothing. That's what the evidence says. The accumulation of hours doesn't enhance practice.
So Daniel's Model C, the stable equilibrium, is a career where the proportion of work that changes your behavior stays constant. Solo execution builds fluency. Observed execution changes your behavior. Explanation changes someone else's behavior.
And the ratio is the thing you design for. The ratio.
I want to name this. Because Daniel asked for a model and I think we should give him one.
The feedback ratio.
The feedback ratio. For every hour of solo execution, some fraction of an hour of observed execution, some fraction of an hour of explanation. The fractions stay roughly constant. The content changes.
The career arc stops being a progression from doing to learning to teaching. It becomes a single mode with increasing complexity.
The junior does simple things with heavy feedback. The senior does complex things with lighter but still present feedback. The expert does novel things with peer feedback and heavy explanation.
Nobody ever stops doing. Nobody ever stops learning. Nobody ever stops teaching.
That's the equilibrium. And it's stable because it's not a reaction to anything. It's just the shape of deliberate practice, maintained for a whole career.
The four hours a day is the anchor. Ericsson's number. That's probably the maximum sustainable dose of observed execution. The rest of the day is solo work and explanation. And the ratio of those two can flex.
A typical day might be four hours of solo work, two hours of observed execution, two hours of explanation. For a senior person. For a junior, maybe two hours of solo, four hours of observed, two hours of explanation.
The total is always eight hours of real work. But the mixture changes with seniority. And nobody ever drops to zero on any mode.
Which means the burnout risk is managed by the variety, not by taking time off. And the skill decay is managed by the continued doing, not by CPD courses.
The laparoscopic study makes the point. The novices matched the surgeons on movement. But the gaze patterns, the expert cognition, only came from years of embedded feedback. If you stop doing, the gaze goes first.
The senior person who stops doing is losing the very thing that made them senior. The expert gaze. The pattern recognition.
That's the tragedy of the standard arc. We promote people out of the activity that built their expertise, then wonder why their judgment starts to drift.
The dual ladder was supposed to prevent that. But it only prevented the management version. The IC track still drifts into -doing.
Because -doing is easier to measure than feedback. A design doc is a visible artifact. A conversation that changes someone's code is invisible.
The implementation challenge is making the invisible visible. Counting feedback loops instead of hours.
That's a measurement problem, not a conceptual problem. We know how to count code reviews. We know how to count teaching sessions. We just don't treat them as first-class work.
Daniel's question is really about whether we could. Whether there's a stable mixture that doesn't require a mid-career correction.
I think the answer is yes, if you define the mixture as the feedback ratio rather than as time spent doing versus time spent learning. The doing and the learning were never separate. We separated them, and that's what broke the arc.
The fix isn't to rebalance the phases. It's to eliminate the phases.
No more production phase, management phase, CPD phase. Just one continuous mode of embedded practice with shifting content.
I want to hear what Hilbert thinks about this. He's been quiet.
Hilbert: The Brother HL-2270DW. Laser printer. Hundred and twenty dollars in twenty twelve. Still works.
That's... a printer.
Hilbert: I did warranty repair for a place that serviced them. The drum unit was the thing that always went. People would print ten thousand pages, never clean the corona wire, then call us because the pages had streaks. The fix was a little blue tab you slid back and forth. Thirty seconds. Nobody read the manual.
The printer had a feedback loop built in and nobody used it.
Hilbert: The people who slid the tab, their printers lasted forever. The people who didn't, they bought a new drum unit every year and blamed the manufacturer. Same machine. Same workload. The difference was thirty seconds of maintenance.
That's the feedback ratio in a nutshell. The maintenance isn't separate from the printing. It's embedded. You slide the tab while you're changing the paper.
Hilbert: Or you don't. And then you pay.
Did you ever see a printer that was actually worn out from too much printing?
Hilbert: Not once. The fusers went from heat cycling. The gears went from sitting. The machines that printed constantly, they lasted the longest. The ones that sat in an office and printed once a week, those were the ones that seized up.
The doing wasn't the problem. The idleness was.
Hilbert: Same with the techs. The guys who took every repair call, they stayed sharp. The ones who moved into dispatch, six months later they couldn't diagnose a paper jam over the phone. They'd forgotten the sound.
The sound of a paper jam.
Hilbert: You can tell what's wrong by where the paper stops. If it's in the fuser, it's a different crunch than the pickup roller. Once you've heard it a few hundred times, you just know. But you have to keep hearing it.
The expert gaze. Or in this case, the expert ear.
Hilbert: I still have the printer. It's in the closet. I don't print much anymore. But when I do, I slide the tab.
Because the maintenance is the doing.
Hilbert: It's not maintenance. It's just how you use the thing.
That's the cleanest description of Model C we've had all episode. The feedback isn't a separate activity. It's just how you use the thing.
The people who treat it as separate, they're the ones buying new drum units every year.
Hilbert: Or burning out. Same thing. You ignore the little blue tab long enough, something seizes.
I think that's the note to land on. The equilibrium Daniel's looking for isn't a schedule. It's a habit of use. The feedback is in the doing, not after it.
The career arc stops being an arc. It becomes a single practice with shifting content. Junior or senior, you're still sliding the tab.
The cutting room floor detail I wanted to mention is that the Academy of Management Journal published a paper on pharmaceutical research scientists. Early specialization speeds up advancement, but later it holds people back. The durable skills, the ones that transfer across projects, those are what carry you through the late career. And durable skills are built by exactly the kind of embedded feedback we've been describing.
The people who specialize early and never broaden, they hit the ceiling. And the broadening has to happen through doing, not through CPD hours.
Right. The durable skills are the residue of feedback-rich practice across varied contexts.
I keep thinking about the printer. The little blue tab. Thirty seconds. The difference between a machine that lasts and a machine that doesn't.
The difference between a career that sustains and one that burns out.
Daniel asked for a Model C. I think we've got one. The feedback ratio. A stable proportion of solo execution, observed execution, and explanation, maintained from entry to exit. No phases. No corrections. Just the tab, slid every time.
The evidence says it works. Deliberate practice was always embedded. The CPD literature is moving away from hours. The dual ladder was a structural attempt at the same thing. All the pieces point the same direction.
We just need someone to name it and build an organization around it.
Thanks to Hilbert Flumingtop for producing. This has been My Weird Prompts. Email us at show at my weird prompts dot com.
We'll be back soon.
