Ambition gets treated as a virtue and I am not convinced. Half the ambitious people I know are mainly frightened of turning out ordinary.

Local elections are the only ones where your vote plainly matters and they have the lowest turnout, which I cannot square with anything.

Being good at your job and being good at describing your job are almost completely unrelated skills, and only one of them gets promoted.

Small talk gets a terrible press, mostly from people who have never had to sit next to a stranger for two hours.

Self-driving is a much harder problem than anyone selling it admits. The driving was never the bit — it is the one situation in ten thousand that nobody wrote a rule for.

There is no book everyone ought to have read, and the list is always suspiciously full of whatever the person making it was assigned at nineteen.

Most people who say they want honest feedback want accurate praise, and the two are easy to confuse until the feedback is actually negative. You can watch the switch happen in real time: the request is sincere, and then the first genuine criticism arrives and the next twenty minutes are spent supplying context.

Nuclear is the safest energy we have per unit generated, and thirty years of being squeamish about it is a large part of why we are still burning gas.

Football was better without VAR and I will not be moved on this. You cannot celebrate on a three minute delay.

A backdoor for one government is a backdoor for everyone, and this is arithmetic rather than politics. A key that exists can be stolen.

A genuine question for anyone who did science at school more recently than me: how much of what I was taught has quietly been revised? Nobody sends an update.

Try watching a match with the commentary off. It is a completely different sport and you notice about twice as much.

Most productivity advice is survivorship bias with a system bolted on. Setting the system up feels like work, which is the whole appeal.

Ask anyone whether they want the truth from their doctor and they say yes, then go to pieces the moment a doctor is actually blunt with them.

Everyone has become an amateur nutritionist about their own dog.

Everyone I know is one broken boiler away from a genuinely bad month and we have all agreed not to bring it up.

I want to defend something unfashionable, which is meetings that could have been an email. Sometimes the point of the meeting is not the information. It is that eight people are now jointly aware that they all heard the same thing at the same time, and nobody can later claim a different understanding. That is expensive and it is doing real work. The email version saves forty minutes and produces eight private interpretations, four of which are wrong and none of which surface until something breaks. Then you spend a great deal more than forty minutes finding out which four. I am not defending the recurring status meeting nobody has questioned since 2019. I am saying the calculation is not "was information transferred" — it is "did we need shared knowledge that everyone knows everyone else has", and for anything contested the answer is usually yes.

Tipping culture has completely lost the plot and I'm saying that as someone who used to wait tables.

For most software engineers a degree stopped being worth what it costs. It is now mainly a filter that employers use because filtering is expensive.

Search has got noticeably worse and I have started adding the name of a forum to the end of every query, which is a strange thing to have to do in 2026.

Nobody I know under forty mentions a pension and nobody over fifty talks about anything else. There is no middle period where it gradually comes up. One day it is simply the topic.

Remote work has made engineering teams worse, and the bit that broke is the part nobody wrote down: juniors learning by overhearing.

Nearly everything we were told about food between about 1980 and 2015 turned out to be wrong, and almost nobody who told us has ever accounted for it.

The algorithm has quietly killed the thing where you find something you were not looking for. Every recommendation is a slightly worse version of what I already like.

A thing I have noticed after fifteen years of interviewing people: almost every interview process is measuring how comfortable someone is being evaluated, and then reporting that as a measure of how good they are at the job. The tell is that the same candidate performs completely differently depending on the room. Give someone a whiteboard and three strangers and you learn how they handle being watched. Give them a real problem and an afternoon and you learn something about how they think. These correlate weakly, and we have built an entire industry on assuming they correlate strongly. What makes it hard to fix is that the noisy signal is cheap and the good signal is expensive. An hour of whiteboarding costs an hour. A realistic work sample costs a day of the candidate's time and half a day of yours, and multiplied across a funnel that is a real budget line that somebody has to defend. So we keep the cheap test and then add stages to compensate for its noise, which is how a process ends up with six rounds that are each individually indefensible. I do not have a clean answer. I have stopped pretending the current thing measures what we say it measures.