A priority only counts if it pushes something else out

Priority labels cost nothing, and anything that costs nothing gets spent recklessly. At first you mark three things P1. Within three weeks, every request that matters to somebody turns up already stamped urgent, and you’re left with the list you always had plus an extra column. The scale didn’t break because people abused it. It broke because nothing in it was scarce.

A priority only means something if it displaces something: a slot, a day, another task’s place in the queue. That’s why the number of levels makes no difference, whether you have three, four or five. Two other things matter far more, and hardly anyone applies them to their own list. The first is an entry test, written down in advance, that someone else can check: a fact a colleague could confirm (“someone outside my team is blocked right now”), not “this is really important”. The second is a cap: one P1, three open P2s. When something new comes into a level that’s already full, something else has to go, and you say so out loud.

That’s the trick. The scale stops being a ranking and becomes a trade-off everyone can see. When someone hands you an “urgent” request, you no longer argue about the label. You ask one question: what moves? And if a P2 hasn’t moved in two weeks, it was never a P2. It either gets a real date or it goes, because it’s almost always avoidance dressed up as a deadline.

Keep it professional and your team will tell you what you need to know

You can be kind to your team without being their friend. Be as warm, human and honest with them as you like, but everyone notices the imbalance of power: you approve the pay rises, and when something has to be decided, you’re the one who decides it. Pretending otherwise doesn’t make the imbalance go away. It just leaves you as the only person in the room still pretending. The fiction survives right up until the day you have to use that power, to give hard feedback, say no or restructure, and at that point it feels like a betrayal.

The second cost is harder to spot, and it’s about information. Looking down the organisation, everything you see has been filtered, and the more layers there are below you, the more you’re handed a tidy summary of a reality you’ve lost sight of. Coming up towards you, it gets held back, sometimes out of kindness: someone sits on bad news because they don’t want to spoil a good working relationship. They mean well, but you end up deciding on the strength of a picture you know is incomplete, and piecing the truth together through back channels.

Getting tougher won’t fix this. What will is not relying on the official line alone. Stay in direct, regular contact with a handful of people at each level below you, and with peers elsewhere in the organisation, without going over anyone’s head and openly enough that nobody mistakes it for spying. Be honest about what the relationship is, and set up your own ways of finding out what never comes through the official channel.

Your AI edge comes from the learning you keep, not the model you rent

Using a good model no longer sets you apart, because every company can get hold of the same one. What you do own is the experience your organisation gathers each day as it makes decisions, and almost no company captures it.

In practice, it comes down to four things:

  1. Your operating history: every decision with its context and its outcome. Which terms you offered a client at renewal, and whether they stayed or left.
  2. Learned policies: the rules that history teaches you, such as which discounts pay off and which only eat into your margin.
  3. Objectives: what the company wants each decision to achieve, so the system knows what counts as getting it right.
  4. Proven results: what worked with which supplier, at what price, and which policy turned out to be expensive.

Turning that experience into learning takes a system that stores it and puts it to work. That means memory that doesn’t vanish when the conversation ends, a reliable record of what was decided and why, and a loop that checks each decision against its outcome so the next one is better informed. Piling up data isn’t enough. The advantage lies in processing it at a level you can manage and control.

A simple test: could you move to another model provider and take your history, your learned policies and your objectives with you? If you couldn’t, you don’t own your intelligence. You’re renting it.

On days you can't do the full habit, do a tiny version: skipping it weakens it more than shrinking it

A habit gets stronger each time you respond to its cue, not as the days go by. So every time you skip it, the link between cue and behaviour weakens a little. Skipping gets easier tomorrow, and easier still the day after.

The way out is to keep a tiny version ready. Say your habit is a morning run, and today meetings ate your whole morning: jog in place for a minute. That minute isn’t about fitness, it’s about the link. The cue fires, and you respond.

The obvious worry is that you’ll end up settling for the mini version every time. When Spencer Greenberg put this to the test, he found the opposite. People who had that option kept up their habit better and were happier with it, because a little does a habit far more good than nothing at all.

It does take one decision up front, and it’s the one almost everyone skips: settle on the minimum version while you still have energy, not on the evening you’re exhausted and looking for an excuse. The same applies when you swap a bad habit for one that takes effort. Keep a lighter version on hand for the days the full one won’t fit.

To check you've been understood, ask them to explain the plan back

A clear instruction doesn’t always land clearly. What you said and what the other person heard can be miles apart, and it usually only shows once the work is done (or done wrong).

The fix is simple, almost embarrassingly so: before work begins, ask the person to explain the task back to you in their own words. It isn’t a test. It works both ways. You find out whether your message got across, and they get a chance to spot gaps or contradictions in your plan, the kind that become obvious the moment someone says them out loud.

It works just as well for a single task as for a whole role. Peter Drucker took the idea further with what he called the manager’s letter: twice a year, each person writes down how they understand their boss’s objectives and their own, which standards they believe they’re held to, and what helps or gets in their way. Nearly every time, something surprises the boss: an offhand remark in the lift that someone turned into an objective, or two demands that can’t both be met (speed and quality, or showing initiative while asking permission for everything). The letter doesn’t make those conflicts go away, but it puts them on the table so you can agree on what comes first.

To perform under pressure, prepare the actions that depend on you, not the result you hope for

When the moment of truth arrives (a negotiation, a talk, a difficult conversation), thinking about the outcome fills you with worry and tells you nothing about what to do. Winning or losing depends partly on other people; how well you carry out an action is far more up to you.

That’s why it pays to prepare with concrete instructions, three at most, that describe what you’re going to do rather than what you want to achieve. Before a negotiation, for example: listen before you respond, find out what matters to the other side, and only give something away when you get something in return. Then rehearse it in your head: recall three times you did it well, and picture how you’ll handle the difficult moments. The brain also learns by imagining, so this rehearsal adds to real practice, even though it can’t stand in for it.

It won’t guarantee the result. What it changes is what you walk in with: a useful mental plan instead of a list of everything that could go wrong. And like any training, it works through repetition: a few minutes, several times a week.

An AI can maintain your knowledge system if you give it a map of the system instead of rules

Almost every personal knowledge system dies the same way: the ideas keep coming, but the maintenance doesn’t. Keeping it up to date is your job, and one busy week is enough for the backlog to grow so large that you stop touching it altogether. When you have time, you tag your notes, link them and turn loose ideas into tasks; when you don’t, everything piles up in an inbox that turns into a graveyard. Willpower is a poor engine for that kind of work.

What changes this isn’t a smarter AI. It’s a single document that describes how your system is built. It names the kinds of notes you keep (sources, ideas, projects, people); the handful of central notes that act as entry points, with everything else hanging off them, for example one per project or area; how those notes relate to each other; what “in its place” means to you; and which distinctions you’ve learned to care about. That document is the map. Without a map, every maintenance decision falls to you; with one, the AI can take that work off your hands.

The temptation is to hand the AI a list of rules. That gets it backwards: a rulebook never covers the case you didn’t foresee, and it makes whoever follows it clumsy. The map describes the shape of the system, and you hand it over with a clear role and permission to use judgement. Think of a librarian. She doesn’t write the books or decide which collection you want, but once she knows your catalogue, she shelves each new book better than a sorting machine running on two hundred instructions.

This shows most clearly with repetition. Over the months, you write down the same idea five times in slightly different words. An AI that uses judgement recognises it and expands the note you already have; an AI that follows rules creates five new ones. That’s the difference between an archive that grows and one that clogs up.

It helps to distinguish between two kinds of ordering. The first is thinking: deciding what an idea means, what you connect it to and which distinctions are worth making. That work is yours and should stay yours, because if you delegate it, you stop thinking. The second is maintenance: filing each note where the map says it belongs, adding the links the map already anticipates and checking whether the idea is already in there. That part you can hand over. Your share is what only you can do: notice something, think it through and get it out of your head, even if that means talking out loud on a walk. And drawing the map, which is thinking too.

29 years ago today I added a news page to my website. Four years ago I wrote that I wasn’t too proud of this blog anymore. This year feels different: I’ve already published more than in either of the last two years, and almost four times as many longer pieces. Here’s to year 30. 🍻

Seth Godin on fake async:

Over time, async creeps toward sync. It’s ironic that Slack messages, texts and email which succeeded because they were async, now create an expectation of real-time response. These are async tools used with sync expectations, so you get the interruptions of sync without its energy.

Personal discipline can’t solve this one. You can block two hours for focused work, but if your team expects an answer within minutes, that block won’t survive. The expectation belongs to the team, so the team has to change it: agree which channel is for what, and what really counts as urgent. Then async can be async again.