What I’ve learnt in 2025 (mostly about growing old)
Know your body the way you know the engine of your favorite car. If you want to drive it for life, admit when it needs work.
Knowing your limits matters more than knowing your powers.
Learn to preserve your energy for high-priority tasks.
At this point, if you don’t know or refuse to admit what your superpowers are, you’re a fool. The only reason you deny them is ego: you think you should be better than what you are. Accept the one thing you are good at. Use it. Exploit it.
You’re never too old to be more optimistic.
It’s never too early to think about your legacy. Legacy clarifies priorities.
The memories you leave with people you care about are your real legacy.
Your child is your legacy.
If you treat your work as part of your legacy, your relationship with work changes completely. Fewer people do this today. Accept that - and choose deliberately.
The fastest way to ground your character is simple: touch the ground more often. Literally.
We’re entering an era where the younger generation will be educated by AI from day one. They will be better than us - in many ways, maybe every way. Admit it. Stay humble. Learn from the young.
Be humble enough to learn from AI too.
Learn from Avatar 3: fuck them back with firearms and dragons if you ever need to. Sometimes it’s the only way.
@thsottiaux eventually i think you should charge coding plan like utilities co. - on demand but not crazy 80% margin api pricing and tiered rates based on supply & demand. it's for the best.
models are, and will remain for quite a while, extensions of people’s capabilities. hiring increasingly means hiring model operators. the opportunity is huge, but so is the risk if we hire the wrong person.
Matt Damon says him and Ben left the Good Will Hunting movie so they could cast whoever would fund it
"Ben Affleck and I wrote it because we wanted the parts as actors and at the time there was a really popular movie that we all loved called reservoir dogs Quentin Tarantino's first movie"
"the story we had heard was that because Harvey Keitel signed up for the movie Quentin got I think it was $500k that was his budget"
"we wrote that part that Robin eventually took we called it the Harvey Keitel part looking for an actor who could get us money because ben and I wanted to star in the movie and we knew we were worth nothing"
the end state of claude/codex subscriptions is infinite-horizon agent loops.
you set a goal. the agent runs. drop in anytime to observe, steer, or course-correct.
in other words, the same management model we’ve used for human employees since the modern economy was invented.
we added human review to every AI output.
but we're reviewing the appearance of thinking, not the thinking itself.
we optimize for oversight theater, not actual judgment preservation.
built a social media agent
spent the first few weeks optimizing reach — timing, distribution, impressions
then realized i hadn't figured out what i sound like yet
AI amplifies what's already there
if you don't know what that is, you're just scaling noise
@JamesMelville You're right - but the problem isn't AI. It's AI tools optimized for volume.
Signal vs slop isn't human vs AI. That divide runs right through AI tools themselves.
years ago i bought monkey app from a 19-year-old founder in nyc. just saw on my x feed he was arrested for strangling someone
i actually liked the guy. hard to process how differently life can turn out.
@vytisbareika@pmarca that makes sense - full company context in every prompt is a different use pattern. 150k/response is wild but i can see how it adds up fast for coding tasks