Your permission system hasn't kept up with you
You're not overloaded because you've run out of skill. You've run out of filters.
It's 4:42 pm on a Thursday. Your calendar is clear for the first time today. You open the laptop with the small, hopeful intention of finishing something. A Slack message pops up. Someone 2 layers below you has a tricky client situation, one you've seen before, and they want 15 minutes. You can see exactly what's gone wrong from the first paragraph. You could solve it in 7 minutes flat. So you say yes.
By 5:10 pm, the call has spilled into a related issue the team has been quietly avoiding. You spot a second thing that needs your eye. You promise to write up a short note tonight so the rest of the team doesn't lose the thread. The original quiet hour is gone. You close the laptop slightly more brittle than you opened it. You don't feel like you did good work. No, you feel like you got pulled.
Nothing about that moment was unreasonable. The request was real. The team did need you. Your help made things better. The problem is that this exact sequence has now happened roughly 800 times this year, and the cumulative shape of it is what your week looks like.
This is the thing nobody says out loud about senior burnout. It isn't usually one impossible task. It's a thousand things you're competent enough to handle, none of which are obviously wrong to take on, all of which add up to a working life that’s already full. You haven't run out of skill; you've run out of filters.
Anthropic published an engineering post last week called "How we contain Claude across products." It walks through how they build permissions around their most capable models. The idea is clean and a bit confronting: as the model's capability grows, the access controls around it must scale with it. In other words, the more the system can do, the more disciplined the rules about what it's actually allowed to do. Having the capability without matching permission isn't power. It's risk.
That last sentence is also a half-decent description of a senior career.
Over 15 years, you've quietly become someone who can do almost everything that crosses your desk. The client briefings, the awkward conversations, the half-broken process, the strategy memo, the team mediation, the technical fire, the messaging cleanup, the third-round interview, the panicked CEO question on a Sunday night. You have the skills for all of it. None of it scares you the way it used to. Your capability has scaled up enormously.
What hasn't scaled is the permission system around that capability. You're still operating under the rules you absorbed as a mid-career professional trying to prove you were ready for the next step. Say yes. Be useful. Catch the thing nobody else caught. Hold the standard. Make yourself the safe pair of hands. Those rules made sense earlier in your career, when your capabilities were modest. The same rules, applied to a much more capable version of you, produce a calendar that nobody could survive.
The cost of it all isn't obvious on any single day. It becomes obvious over the course of the year. The resentment you can't fully justify. The reluctance to open the laptop on a Sunday. The way you've started feeling slightly tired before the first meeting of the day has even started. The brittleness when one extra request lands on a day that was already too full. These aren't signs of weak capability. They're signs of a missing permission system.
The Anthropic post is useful because it treats access control as an engineering problem, not a motivation problem. They don't ask the model to "be careful" or "use good judgement". They build the constraint into the system. The model literally cannot do things outside its sandbox. The discipline lives in the architecture, not in the model's intentions.
The same logic applies to how you should be thinking about your own week.
Most advice for overloaded senior people is some version of "be more disciplined". Say ‘no’ more. Protect your time. Don't get pulled into things below your level. The advice is right, and it almost never works, because it puts the entire load on in-the-moment judgement. At 4:42 pm on a Thursday, when there’s a request from a person you like, in a tone that's reasonable, about a problem you can clearly help with, your judgement is the worst possible tool for filtering. It's tired, it's social, it's optimised for the next 10 minutes, and it's already lost the thread of what you were going to do with the hour.
A permission system replaces in-the-moment judgement with prior decisions. It's the part of your operating model where you decide - in advance - what kinds of requests you engage with at all. Not what you say yes to today. What you've already decided, before today, you don't even consider.
A partner at a professional services firm I know set a hard rule about 18 months ago: I will not join any recurring meeting scheduled by someone more than 2 levels below me. Not "I'll try not to". Not "I'll review case by case". A flat, public rule that the assistant could enforce on her behalf. The first 2 weeks felt impossible. People were genuinely upset. She missed things. By week 3, the calendar had restructured itself around the rule, and she was reclaiming about 4 hours a week. By month 3, the rule had quietly raised the standard of what got escalated to her, because the team learned to filter before the calendar invite went out.
The interesting part of that story isn't the 4 hours. It's that her capability didn't change at all. The same person who could do everything before could still do everything. What changed was the permission system around the capability. The constraint did the work she was trying to do with willpower.
A team at Varick published a small piece last week on a 3-tier model for agentic work. They split tasks into automate, human-in-loop, and human-only. The idea is meant for agent design, but it maps cleanly onto a senior workday, and most people I talk to are running everything at human-only by default.
Automate is the work where the standard is well-defined enough that a system or a delegate can produce it without you, and the cost of imperfection is low. Status updates. Standard contract redlines you've now seen 400 times. The 80% draft of a regular client report. If you're still personally generating the first draft of anything in this category, your permission system is misconfigured.
Human-in-loop is the work where you don't need to do it, but you need to see it before it goes out. The agent or the delegate produces it. You review, adjust, and approve. This is where most of the senior pay-off lives, and it's also where the trap hides: many people convince themselves they're delegating when, in fact, they're rewriting. The rule of thumb is that if you change more than 20% of what comes back, you haven't delegated, you've outsourced your typing.
Human-only is the small set of things that genuinely should not happen without you. The conversation with that one client. The decision on this hire. The strategic framing of the next quarter. The hard talk with the underperforming senior. These are the things you should say yes to without flinching, because they're the very reason you exist in the role.
The diagnostic question is uncomfortable, so here goes: What percentage of last week was actually in the human-only tier? For most people I ask this, their number is somewhere between 15 and 25%. The rest was capability deployed against work that didn't need it. That's the gap a permission system closes.
You don't need a framework. You need to make 3 small decisions before Monday morning, while you still have a little distance from your inbox. I run them as a short weekly reset. Give it a name so it actually happens. Call it the Monday Permission Reset (MPR, because we love acronyms). Three lines, written down before the week starts:
1. One Default No. Name 1 category of request you'll stop saying yes to by default. Not forever. For 4 weeks. Pick something specific enough that the rule is unambiguous in the moment: "I will not take ad-hoc calls about issues that have an owner who isn't me." Or "I will not join the first meeting on any new initiative I'm not directly accountable for." The narrower the rule, the easier it survives contact with reality.
2. One Demotion. Name 1 category of work you'll move from human-only to human-in-loop. Pick something you can delegate. Decide who picks it up, what the review rhythm looks like, and what good enough looks like. Then, actually let them produce the first draft. The discomfort of seeing a B+ version of something you'd have made A- is the cost of getting the hour back. Pay it, and buy back your time.
3. One Protection. Name 1 thing that should be in human-only and isn't getting enough of you. The hard conversation. The strategic decision. The relationship you've been letting drift. Your permission system isn't only about saying no. It's about freeing up the capability for the work that deserves it.
One no, one demotion, one protection. That's the whole reset. Run it every Sunday night, and your permission system gets tuned a little tighter each week instead of drifting back to "yes by default" the moment your calendar fills up.
None of these fixes burnout. Burnout is more than a calendar problem, and I don't want to pretend otherwise. But for a particular kind of senior burnout, the kind that comes from quietly running a 2026 capability on 2014 rules, this is where the work starts. You’re not becoming less capable; you’re building the part of the system that determines your capability.
You're not overloaded because you can't do it. You're overloaded because nothing in your week is telling you which version of "I could do this" should turn into yes.
If your week is always human-in-the-loop, always at capacity, always crazy, start by pinpointing why it’s like that. The Diagnostic is free, takes less than 5 minutes to complete, and will show you where your week is leaking.