I was running out of things to teach them.
This should have felt like a management success. It mostly made me wonder what I was still there to do.
Over roughly a decade managing at Apple, I promoted about twenty engineers and five managers. Several high performers moved from mid-level engineer to senior IC or senior manager over those years. The fastest made those two jumps in about eight.
Eight years…for the superstars.
Career progression gets drawn as a ladder because HR software needs boxes and people think linearly. In practice, it resembles Chutes and Ladders and is anything but a straight line up.
People want the formula. What do I need to do for the next level? My peer got promoted. What were their ratings? Why didn’t I?
I vividly remember a conversation I had with one of my directs. He’d said “I deserve to get promoted,” and proceeded to show me a laundry list of accomplishments. Impressive in isolation, but I asked him to provide me a list of three or four external managers that knew his work; senior folks who were aware of his listed deeds. And you know what? That particular list came back empty.
The lesson here is that at a certain point, you need both work output and the ability to showcase it. You can't do the work in isolation without anyone else knowing about the impact you're making. And that’s a tough pill to swallow, especially since it’s not something taught in school.
Early in a career, capability is fairly visible. Can you do the work? After a while, the question changes: do people trust you to decide what should happen next? Do they know to reach out to you as the person who Gets Work DoneTM?
I traveled to China a lot in those years. We would spend most of the day sitting on the factory lines, and I split my time between meetings and technical work. One trip I sent an engineer down to debug a few stations instead of going myself, which I would have done a year earlier. That afternoon someone in my meeting mentioned how well those stations were running. Not as a status update to me. Just as a fact about the line. That was when I knew I could stop being heavy-handed, and that a promotion was coming.
Seniority wasn’t pedigree or another training course. It was agency, judgment, and the ability to move without waiting for permission. Once someone understood the team’s culture, values, and strategy, I expected them to act on it.
A manager should create opportunities and clear barriers. Ask tactical questions, sure. But if I still had to identify the problem and prescribe the answer, you weren’t operating at that next level yet.
Judgment comes from reps. People shipped plans that got rejected and debugged systems they had misunderstood. Then they tried again, usually with better instincts.
Now companies are staring at the bottom three rungs and saying, “AI can do that work.”
To be fair, a lot of that work probably should disappear. Tedium does not become educational merely because we made the new hire do it.
But once entry-level work goes away, where do entry-level employees get the reps?
Anthropic recently ran a randomized controlled trial with 52 mostly junior software engineers who knew Python but hadn’t used the Trio library. Half used an AI assistant while learning it. Everyone knew a quiz was coming.
The AI group finished two minutes faster. The difference wasn’t statistically significant. Then they took the quiz.
The AI group averaged 50%. The hand-coding group averaged 67%. The largest gap showed up in debugging; the exact skill people need when generated code looks plausible and happens to be wrong.
The hand-coding group made more mistakes during the exercise. They also had to resolve them. Annoying in the moment. Useful down the road.
There was an escape hatch in the results. AI users who asked conceptual questions, requested explanations, or checked their understanding performed much better. The people who delegated the coding and debugging learned the least.
Same tool and yet very different behavior.
Another study of 5,179 customer support agents found that AI assistance improved productivity by 14% on average and 34% for novice and lower-skilled workers, with little benefit for experienced workers. The system appears to have passed techniques from top performers to newer employees.
AI can help junior employees contribute sooner. It can also make them look more senior before they can operate independently.
A productivity dashboard records tickets resolved and code shipped. It won’t show the judgment that never formed.
And the people who need those reps may be getting fewer chances to earn them. A new Stanford analysis found that employment for 22-to-25-year-olds in the two most AI-exposed occupation groups fell about 11% from November 2022 to June 2026. Employment for the same age group in the three least-exposed groups grew about 10%.
The researchers are careful: this is descriptive, and it doesn’t prove AI caused the decline. Remote work, pandemic disruptions, and other differences may explain some of it. Still, the group that most needs work experience is losing ground fastest in the jobs most exposed to AI.
The easy corporate answer is “mentorship.” Correct, and suspiciously free in the slide deck.
Mentorship costs output today. A senior employee has to explain the context, watch a junior take longer, allow a survivable mistake, and help clean it up afterward.
That’s apprenticeship debt: current productivity spent so someone else can exercise judgment later.
Every company wants a deep bench of senior talent. Fewer budget senior time to create it. AI makes avoidance especially tempting because the agent doesn’t need a 1:1 and won’t ask why another agent got promoted.
The debt still comes due. Usually during an outage, when the code works according to the agent and nobody in the room can explain why it was built that way.
So I would change how we give junior employees work.
Give AI the tedium. Keep the decisions.
Automate the work that teaches nothing. Let AI handle boilerplate and rote transformations.
When a task is new to someone, use AI to explain, question, and critique before it generates the finished answer. Make the junior employee teach the reasoning back. If they can’t explain the choice, the work isn’t done.
Give them whole problems with stakes small enough to survive a wrong answer. Let them sit in the meetings where tradeoffs get decided, then ask what they would do. Shadowing without responsibility becomes corporate tourism.
And measure independence. Throughput tells you what shipped this week. Seniority shows up when the manager can step away and the right work still happens.
Ten years may shrink. That’s a good thing.
The timeline can shrink but the experience has to come from somewhere.
What work taught you judgment that you’d be tempted to automate today?
Nobody actually knows how AI gets used.
AI tool sprawl happened fast, and visibility never caught up. Employees paste contracts into ChatGPT, run code through Copilot, and build workflows in apps security never approved.
Harmonic Security classifies every AI interaction by task, tool, and team, so you can see what’s actually happening across approved and unapproved apps alike.
