The most common path to becoming a manager looks like this: you were the strongest individual contributor on your team, you got promoted, your team grew, and at some point — usually around the third missed deadline, the second hard 1:1, the first time someone on your team needed real coaching — you realized that the skills that got you the title are not the skills the title actually requires.
This is the accidental manager problem, and it's the most common career arc in tech and modern services work. The promotion was a reward for individual performance. The new job is something else entirely. And in the gap between the two — between "great IC" and "capable coach" — most companies invest very little. Some send their new managers to a two-day workshop. Some assign them a mentor who's just as overloaded. Most do neither.
This post is about why manager training rarely closes the gap, what good coaching actually requires, and what's started to shift now that AI can carry part of the load that managers were never going to consistently carry on their own.
The accidental manager problem
The work of an individual contributor is mostly internal: understand the problem, design a solution, ship it. Success is measured in artifacts produced. Feedback comes from the work itself — code compiles or it doesn't, the customer signs or they don't.
The work of a manager is mostly external: understand a person, design a conversation, repeat for every direct report, weekly. Success is measured in how other people perform, develop, and stay. Feedback is delayed by months and shaped by everyone's incentive to be polite.
These are different jobs. The instinct that made someone a great senior engineer — go deep, solve it yourself, ship — is exactly the instinct that makes them a poor manager: take the problem from the report, fix it, hand it back. The behavior that earned the promotion is the behavior that has to be unlearned.
What's striking is that nobody warns the new manager about this. The conversation that should happen on the day of the promotion — "the skills that got you here are not the skills you'll need; the next twelve months are about learning a new job, not doing the old one with more reports" — almost never happens. The new manager assumes their job is to keep doing what they were doing, plus also handle some 1:1s.
By month six, they're underperforming the IC who got the promotion alongside them.
Why leadership training usually fails
There's no shortage of leadership training programs. Most organizations of any size have one. Most managers attend at least one workshop in their first year. Most don't apply what they learned. The failure modes are predictable.
Training is too far from the work. A two-day offsite three months after someone became a manager teaches in the abstract. By the time the manager is actually facing a difficult 1:1, the workshop content is gone. Behavior change requires near-real-time application; classroom content has near-zero retention against a real situation.
Training is too generic to apply. "Give effective feedback" is advice every manager has heard. None of them are paid to give effective feedback in the abstract. They're paid to give effective feedback to Marcus, who reports to them, who missed Tuesday's standup again. Generic advice doesn't translate into specific action without intermediate work the manager has to do alone.
Training models don't fit the team. The frameworks taught at most management programs come from large enterprise contexts — formal goal-setting, annual review cycles, calibration committees, structured 360s. Most readers manage teams of 5–20 at companies that don't have any of that infrastructure. The frameworks don't port.
Training doesn't produce reps. Coaching is a craft. Crafts require practice. Reading about coaching produces no coaching reps. A two-day workshop produces, at best, two days of role-play. Most managers come back to their teams without having actually delivered any of the practice they were taught.
The result is that most managers learn to coach the way most parents learn to parent: by doing it, badly, for the first few years, with nobody watching closely enough to correct them.
The conversation managers avoid
Talk to any new manager about what they find hardest about the job, and the answer is almost always the same: the conversation about underperformance.
Not the strategy meeting. Not the calibration. Not the public-speaking-to-the-team part. The 1:1 where you sit down with someone you like and tell them their work isn't meeting the bar — and what specifically has to change. That conversation.
The reason is simple. Most new managers were trained, implicitly, to be liked. The skills that earned them the promotion — collaboration, going the extra mile, helping teammates — were skills oriented toward making other people's lives easier. Telling someone their work isn't good enough is the inverse of that orientation. It violates the social contract the manager spent years building.
So they avoid it. The standard pattern is predictable:
Instance one: ignored. Probably a one-off, the manager thinks.
Instance two: noted internally. If this happens again, I'll say something.
Instance three: rehearsed, never delivered. I'll bring it up at our next 1:1.
The 1:1 itself: soft-pedaled or skipped. They had a rough week. Next time.
Three months later, the inflection point. The issue has compounded into something that has to be addressed. The conversation that was hard at instance one is now ten times harder.
This is the same compounding pattern from our previous post on delayed feedback, except here the delay isn't about politeness or oversight. It's about training. The manager doesn't avoid the conversation because they're a coward. They avoid it because nobody ever showed them how to have it.
How AI is starting to fill the coaching gap
The historical solution to the manager-training gap was to throw more training at it. The new solution — emerging in the last two years — is to give the manager a coach in the moment.
The case for this is mechanical. Most coaching mistakes new managers make come from a small set of well-documented failure modes: they fix the report's problem instead of asking questions, they soften feedback until the message is gone, they avoid the difficult conversation entirely, they over-index on recent performance, they confuse activity for impact. Each of these is recognizable from the outside. A coach watching a 1:1 could call them out in real time. The problem is that real-time human coaching doesn't scale below the executive tier — it costs $300+ per hour, it requires scheduling, and it's not available the moment a manager actually needs it.
What's changed is that AI can do the recognition work cheaply. A model trained on thousands of management interactions can spot the same patterns a human coach would spot — and offer a prompt back to the manager in real time, on the same channel they're already using, before the conversation goes off the rails.
Performance Blocks builds this layer into the product through Henry, our coaching agent. Embedded in Slack, Teams, the web app, and email, it walks managers through the structured tasks they'd otherwise skip — starting a 1:1, kicking off a 360 cycle, writing a quarterly summary — and turns a quick description of what just happened into a structured observation that lives in the file. The agent doesn't replace the conversation; it raises the floor on what managers do alone.
The model isn't a replacement for the senior coach a CEO might have. It is, for the first time at scale, a replacement for the absent coach the average mid-level manager has always lacked.
What good coaching prompts actually look like
The default mode for most new managers is to react. A report says they're stuck on a project, the manager solves the project. A report says they're frustrated with a colleague, the manager mediates the conflict. A report says they're thinking of leaving, the manager makes a counter-offer. Each move is well-intentioned. None of them coach.
A coaching prompt does the opposite work. Instead of solving the problem the manager hears, it surfaces the question the report hasn't asked themselves yet. A few examples, ordered roughly from easiest to hardest to use:
"What have you tried so far?" The simplest coaching move. It costs the manager nothing, it gives the report space to think out loud, and it produces information the manager couldn't have gotten by guessing. Most new managers find that asking this once or twice in a 1:1 turns the conversation around without any other change.
"What's the smallest version of this you could test this week?" Used when the report is frozen on something big. It reframes a paralyzing project as a tractable next step. Borrowed almost verbatim from Michael Bungay Stanier's The Coaching Habit, which is the single most useful book a new manager could read on this topic.
"What would have to be true for that to work?" Used when the report has a plan you're skeptical of, and the cheap thing would be to argue with them. The question forces them to surface the assumptions inside the plan, which is what you wanted them to think about anyway, without you being the one to point them out.
"What's the thing you don't want to tell me?" The hardest of these to use, and the one that produces the most signal. It works because most direct reports have one thing they're avoiding telling their manager — a missed deadline that hasn't surfaced yet, a peer they can't work with, an interview they took. Naming the dynamic invites it forward without asking the report to volunteer the specific.
A senior coach watching a new manager's first three months would remind them to use these — gently, every time the manager defaulted to fixing instead of asking. AI can sit in for the missing coach.
Frequently asked questions
Why aren't managers trained to coach?
Most companies promote individual contributors into manager roles based on IC performance, with little or no targeted preparation for the new job. Even where formal training exists, it's typically a one-time workshop that's far from the actual work, generic across team contexts, and produces no real practice reps. The combination of mismatched promotion criteria and ineffective training is why most managers learn to coach by trial and error during their first few years on the job.
What's the difference between managing and coaching?
Managing is about ensuring work gets done — assigning, tracking, escalating, and removing blockers. Coaching is about helping the person doing the work develop the skills and judgment to take on harder work next time. Managers default to fixing problems for their reports because that's what their IC instincts tell them to do; coaching requires the harder skill of asking questions instead of giving answers, and it has to be learned.
How can AI help managers coach better?
AI sits in the manager's flow and prompts the moves a human coach would prompt — asking questions instead of giving answers, structuring difficult 1:1s before they happen, drafting observations in the moment, generating balanced summaries at the end of a cycle. It doesn't replace the conversation, but it raises the floor on what managers do alone, particularly for new managers who don't have a senior coach available to them.
What is the accidental manager problem?
The accidental manager problem describes the most common path into management: an individual contributor is promoted because they were good at their previous role, given a team, and expected to coach without targeted preparation. The skills that earned the promotion (deep individual focus, problem-solving, shipping) are different from — and sometimes opposed to — the skills the new role requires (asking questions, developing others, delivering difficult feedback). Most managers learn the new job by trial and error.
How do you have a difficult conversation with an underperforming employee?
Anchor it to specific, recent observations rather than aggregated patterns. Lead with what you saw, the impact it had, and what should change next time — not with a label about the person. Have the conversation as soon as the issue is recognizable rather than waiting for a "better time"; the difficulty compounds with delay. And follow up within a week to see whether the change has stuck.
What we know — and what we're refining
If you manage people and you've never been formally trained to coach, the move this week is small: pick one of the four coaching prompts above ("What have you tried so far?" is the easiest), use it three times this week instead of solving the problem yourself, and notice what happens. You'll get more information out of three minutes of asking than you've been getting out of forty minutes of fixing.
Most managers don't need a leadership program. They need three things on every difficult conversation: a structure, a prompt, and a way to capture what they observed. We've built Performance Blocks around this exact need — Henry walks managers through the structured tasks that get skipped without it (the 1:1 they're avoiding, the 360 cycle they don't know how to run, the summary they'd otherwise write at 11pm Sunday), drafts an observation from a quick description of what happened, and turns the file into a summary when it's time. The training that didn't stick at the workshop happens, instead, ten reps at a time over the manager's first year. It's the closest thing to a senior coach in the room that we've been able to make work at scale.
The detail we're still refining is which prompts produce the most movement at which stage of a manager's development. The patterns we see in our pilots are suggestive — newer managers respond strongest to "What have you tried so far?", more experienced managers to "What's the thing you don't want to tell me?" — but we don't yet have enough data to call it. If you've worked with new managers and have a sense of what unlocks them, we'd genuinely like to compare notes.
The accidental manager problem isn't a training problem. It's a coaching problem — managers were never given coaches, and most companies couldn't afford to give them one. That's the part that's changed.