Pinned post
Vinny
Verified account This account is verified. Learn more@hot_town
1.3K Following 7.1K Followers 13K Posts
Latest posts 5
Start your workday at 5 a.m.
Get in 2 solid hours of agent orchestration before getting the kids ready for school.
Drop em off.
Go for run.
Have breakfast.
Do a few more hours of deep work.
Finish at 3 pm.
Spend afternoon and evening with family.
Put the kids to bed.
Check on your agents for an hour before winding down and going to bed.
Repeat.
Get in 2 solid hours of agent orchestration before getting the kids ready for school.
Drop em off.
Go for run.
Have breakfast.
Do a few more hours of deep work.
Finish at 3 pm.
Spend afternoon and evening with family.
Put the kids to bed.
Check on your agents for an hour before winding down and going to bed.
Repeat.
i have ai psychosis and completely changed my schedule this year to maximize agent output.View quoted post on X
what I’ve learned:
1. if you have N working hours, it’s now beneficial to spread those out as much as possible instead of work continuous blocks. aka work more days and start earlier/end later, but with more breaks.
models can work autonomously for days but still need your direction, and you need to sometimes unblock them. frequent touch points are important for that.
e.g. much better to find out at 7am that your models got stuck, and get two hours of progress before your work day, than find out at 9am.
similarly, a quick check-in on sunday morning sets up your models for a day of work.
this doesn’t have to mean actually working more hours if you don’t want to. a five minute chat from your phone is enough.
2. there’s now an entire second workday that can happen while you’re asleep. but it only happens if you make sure your models can work overnight.
it’s crucial you spend the end of your workday setting up your overnight agents for success.
my lazy life version looks like this: every night at ~9pm i go on a run, dictate what i think should happen while i'm asleep, and dump the transcript into grok bot. it takes 10 minutes and results in eight hours of work.
3. prioritize short, high intensity work sessions over long hours.
when you’re in front of your computer, all your time should go to unblocking your agents with *high quality* ideas and decisions, e.g. doing Matt Pocock style grill me sessions.
alternating a few high quality planning sessions with touching grass is more valuable than spending twelve continuous hours babysitting agent traces.
ofc it’s beneficial if you’re able to combine intensity with long hours, but that’s harder.
it reminds me of how athletes train: workout heavy multiple times a day and focus on recovery in between.
4. do whatever you need to maximize your capacity for idea generation and decision making.
don’t make stupid health and sleep sacrifices in pursuit of productivity.
you’re no longer getting paid for brute forcing solutions. that’s what agents are for.
it’s much better to get good sleep than 996 if the latter makes you even a tiny bit less focused or creative the next day.
None of the prescribed AI workflows have endured and none of them will.
the entire advantage of llms is they are incredibly flexibleView quoted post on X
which means they can conform to whatever way you prefer to work and think and build
there's infinite people posting about the best way to use agents. like, first you need to do a plan, and then you need to blah blah blah
but if that doesn't click for you, don't work that way. work however you like
the little bit of advice i'd give is think about ways to keep it fun
it can get boring consuming big chunks of text, ask for a fun video in a crazy voice that gets unhinged over time