Optimizing my sleep around Claude usage limits
mattwie.se
mattwie.se
If I can't vibe code while sleep deprived, I sure as heck won't be able to react to an AIS alert and change course in the middle of the night!
Anyhow, vibe coding is pretty low stakes compared to the joys and terrors you'll find out at sea.
Bon voyage!
As a landlubber, the terrors are quite easy to imagine.
Marine mammals
Sunsets with no land in sight. Sunsets framing the land. Sunrises with no land in sight. Sunrises framing the land. Thunder and lightning rolling up one side of an island in the distance, putting on a show.
Dark skies and the stars.
The peace and tranquility of quiet places with just nature and you. Until the sod over the hill turns on their generator.
Fresh fish.
The feel of sea spray, wind in the ears, the connection to your boat, knowing if things are right just by the feel and sound.
And that's just from doing non-ocean-crossing sailing.
I had a max x20 account for the past three months and hit limits just about every period that fell within working hours.
I finally cancelled it two days ago due to overspending-guilt/token-grinding guilt, and they shorted me my last day due to an error on their side regarding time zones.
It's really so dependent on your workloads. Conversations around token expenditure are wildly different from individual to individual,and workload to workload.
Dealing with codebases that require contextual reading due to a lack of training corpus (R/Go/Common Lisp, among others..) EAT context and tokens for breakfast.
A while back, I had a big paper deadline a week away and knew I didn’t have enough time to finish without sacrificing sleep.
Rather than cutting my sleep short, I decided to stick with 7–8 hours of rest and instead lengthen my wake window. I worked out a schedule that gave me six nights of sleep across seven days. It meant waking up at stranger and stranger times as the week went on, and getting some odd looks from my roommates when I emerged from my room. But in the end, it was totally worth it. I was waking up well-rested and ready to tackle those extra-long days.
The effort paid off 100%. Not only did I make the deadline, but my paper was accepted as well. A year later, that same paper helped me get into my PhD program of choice.
It’s funny how these short bursts of intense effort can sometimes have such a big impact.
Best of luck with your side hustle!
Personally in that situation I would (and do) get plenty of sleep every night and then skip the final night. I find the fatigue from a lot of lost sleep normally doesn't hit me in full until the second day after, and the final-day panic is enough to counteract the lack of sleep.
So I tried a complete “cycle” of 28 hour days until I realigned with the normal day. Which happens to be the same thing GP did. LCM(24, 28) = 168 which is 7 days with 6 cycles. Roughly 19 hours waking and 9 hours sleeping. I did let myself sleep longer instead of holding to 7-8 hours like GP.
It definitely felt weird because my wife wasn’t on the schedule, but I didn’t feel super sleep deprived. I’m sure with multiple complete cycles you’d see more adverse effects, so it’s probably best to only do this very sparingly. Maybe napping during the wake period could alleviate issues.
https://en.wikipedia.org/wiki/Non-24-hour_sleep%E2%80%93wake...
I'm depressed by their determination.
> My velocity has increased 10x and I'm shipping features like a cracked ninja now, which is great because my B2B SaaS is still in stealth mode.
Does someone have a good book I can read on stealth mode startups?
My B2B SaaS is only small (about $1.25m ARR) but I can't imagine shipping features without someone to use them. We could definitely write a lot more features if our users didn't point out the ways that what we were doing wasn't quite right or didn't fit with the problem we thought we were solving.
I can imagine doing the business part, regulatory part or hard tech part of a startup in stealth mode - but what would does it mean to ship features in stealth mode?
That said, I think the startup book, blog, Tweet, and LinkedIn think piece world went a little overboard with the "ship your MVP early" concept. For a while I was seeing little startups proudly ship things that just didn't work or lacked key features. The only real-world feedback they were getting was that the product was incomplete.
If you burn your early customers badly, you lose precious advocates. In my opinion you're doing the right thing by shipping features for your customers.
It is possible to over-focus on a few vocal customers, though. A common trap for small companies is to get into a relationship with 1 or 2 very vocal customers who will act like they're giving you insights into what the industry wants, but in reality they're just telling you what you want to hear in order to specialize your software for them. It's important to go out and validate the new feature requests with the broader market.
The worst products never move past that stage and just become rentiers. I think you probably know a couple buildings those companies have built.
Why? If you don’t mind me asking.
(Also, I'm assuming this post is a joke, but some of the comments here seem to ve taking it seriously)
I gave it up because I found it wasn't very fault tolerant. If I missed a bedtime even by just a few hours, or ate before trying to sleep, I was in a bad state for a day or two until I could get back on track.
That's what terrifies me about polyphasic sleep at sea. I had a few "cheat" days while attempting this, but of course the safety of myself and others didn't depend on if I hit snooze 5 times in an hour. Claude just got lonely for a bit.
It's really an interesting technique and I hope to find and talk with sailors who've done it. Thanks for sharing your experience.
It was usually social pressure that did me in. People want to meet for coffee or drinks or food at a time when it wasn't well aligned with my nap schedule, and I started making compromises...
I suppose something that engages you for hours could appear unscheduled while sailing, but it seems like most sources of such things could be mitigated with adequate planning, and they're unlikely to involve coffee or beer or birthday cake.
I would say, generally speaking, that comprises the bulk of the time. Most likely you will spend more time in unscheduled multi-hour long tasks than anything else over the course of an extended trip.
You could just work 24x7 and Vibe code yourself to a trillion dollars.
In all seriousness, you need to rest properly, you will be more productive and make less mistakes and have less rework.
Finally, if you are using Claude to get your "React component to hydrate correctly", you are not being very efficient in using AI as a coding agent.
I wrote a post about using Full.CX MCP that will build complete features for you with test etc. https://dalehurley.com/posts/fullcxmcp
Or writing prompts that get fired off by a script once the usage resets when sleeping so that you at least get some free tokies?
I'm sympathetic to wanting to squeeze out what you can to control costs, but this is something that might only seem sustainable because you're too exhausted to fully appreciate the potential deleterious long-term health effects.
(not affiliated, I was just very surprised when they tried to upsell me last time I renewed my domain :))
though I personally just make a single request right when I wake up, before making coffee.
https://www.cbsnews.com/baltimore/news/elderly-man-reunites-...
He's doing great, still sailing.
That was the third boat he sunk. This was the most famous sinking:
https://www.latitude38.com/lectronic/an-inspirational-sailor...
Three shipwrecks in ten years might be a record?
:(
> He's doing great, still sailing.
:D
> That was the third boat he sunk. This was the most famous sinking:
:o
Your comment was a bit of an emotional rollercoaster, but I'm glad your dad is doing good!
Reminds me of the guy who claimed that eating butter made him smarter, ate half a stick a day, then died of a heart attack.
*Optimising My Claude Usage Around Their Usage Limits
Your sleep regime here is in no way optimal
I'm genuinely unsure if this is a parody or not.
All this AI hype is pushing me even more to think that the value of a job is inversely proportional to the likelihood of it being done by a computer.
Optimizing one’s sleep around rate limits is perhaps worse than “optimizing sleep around my boss’ personal life style”.
Besides, are we in a low wedge industry that $180 ($6 per day) is that difficult?
Inspired by this, a buddy of mine tried "DaVinci Sleep" at our residential high school, and lasted a week before he crashed for 20 hours and went back to a normal schedule.
Apropos of nothing, he's now a very well regarded academic - in an unrelated field.
To be fair, computer science is famous for people rearranging their sleep schedule around when the compute time was available.
Reorganizing your sleep around an activity you refuse to stop is one of the most obvious warning signs of a dangerous addiction.
Please don’t destroy your mental and physical health because you’ve convinced yourself that you achieved impossible and fantastic sounding gains in productivity.
Invalid API key · Please run /login
I know there are a lot of github issues about this but most of them dont seem to directly apply. The weird thing is: when I run the exact same command manually in my terminal, it works fine. When run via cron, it fails. I suspect it has something to do with how Claude CLI stores the session token or environment variables, since cron doesn’t run as a full login shell. I already tried wrapping the command in bash -l -c "..." which helps in some cases, but not consistently. Has anyone here figured out a clean, reliable way to run Claude CLI from cron without having to manually relogin every 5 hours or so? Or is it only restricted for paid anthropic API Key?
I find that each account gives me 1-2 hours of coding in Claude Code usually.
Wait…erm, is this for real?? :face_with_spiral_eyes:
But when I first started tinkering with AI, I found it so addictive, it reminded me of how I used to play videogames in my 20s obsessively. I think it's the randomness of the rewards that makes AI possibly addictive. It sometimes felt like a slot machine, the way that tweaking a setting or two would get me closer to the goal.
Every time you do coder.Health-- for bank.Money++, you have the problem that you are never able to do coder.Health++ for bank.Money-- afterwards.
Never sacrifice health for money. Never. Every idea that needs to be worked on more than 50 hours a week is an idea not worth working on.
I know how it is, I've been there myself. You'll be reluctant to listen now. But maybe in a year you'll come back and remember this comment.
It isn’t always about money, and it isn’t always a choice.
It is a personal decision to build or destroy one’s body, and while your advice is maybe sound in general, we should avoid generalizing for other people.
A little bit of sleep deprivation isn’t life threatening (such as being significantly overweight, or smoking, or consistently eating unhealthy foods). We should avoid over-moralizing to others about the engineering tradeoffs they make in their own lives.
Many a family has been enriched by mothers and fathers overworking themselves to build a better life for their children, for example.
> Never sacrifice health for money. Never. Every idea that needs to be worked on more than 50 hours a week is an idea not worth working on.
If I had taken this advice verbatim in my 20s, I wouldn’t be able to frequently be working 20 hour weeks in my 40s. I would argue that speaking in absolutes like this is actually bad advice.
It is frequently a good thing to work yourself to burnout for a year or three if it means you can work at 20% for the following 20 years.
Burnout is never a good thing. Go slower. Go well. Thank yourself later.
I couldn’t work for two years after it and it was still worth it.
> I couldn’t work for two years after it and it was still worth it.
That sums up kind of the problem I have with that type of survivor's bias.
Question to you:
Was it worth because of the burnout or because of other variables in that specific part of your life?
If the other variables were not the same, would you still recommend it, just for the sake of "recommending the experience of a burnout"?
I disagree, I believe what we mean when we say "productivity" these days was invented maybe during industrialization, maybe 1800s, and a couple etymology dictionaries I checked seem to agree, that the word being used in an economic sense to mean "production per unit" only started occurring in the 1890s. Also, I believe that the modern sense, meaning, "whether a human's time is spent being productive for the economy," is a mid to late 20th century invention of neoliberalism.
I don't really like hyper-generalizations like "all people have been doing this thing for all of human history," because it's just a silly thing to say on the face of it - the English were doing very different things and had very difference concerns in the year 800, 1100, 1700, 1900, and 2025! But also, the English in 1300 were doing very different things than the indigenous Americans in 1300! That said, one generalization I'm comfortable with is that throughout all of human history, until maybe the 1940s, people have been seeking comfort, leisure, and peace, and only recently have we developed a global society, and at that one that is obsessed with finding economic justifications for everything, including how humans spend their time!
You mention, "it isn't always a choice," and I agree, that is the failure of capitalism - there are people out there destroying their lives, minds, and bodies to scrape out a living. Our global economic system has failed these people - in fact it's sacrificed them on the alter of consumerism.
Many a child had stunted development from mothers and fathers subscribing to the cult of capitalism and overworking themselves and never being at home, with the self-serving justification of "I'm making a better life for my child," when in fact they're not.
Can you expound on this for me? This rule is not at all obvious to me. I'm curious what perspective this hails from :)
For example, most of my career, I will take 6+ months off between particularly intense work crunches for contracts/startups/jobs. I find the time off restorative to the point where I get restless for the next crunch.
That is a sign of addiction, not a sign of balance.
The issues I have with this "crunching it" mentality now (post-burnouts) is that even with some time off afterwards you'll pay the price with physical health.
Just the heart issues alone that you'll get because of the absurd and constant stress levels are now for me an indicator that it's not worth it.
A company doesn't give a damn about you. They are not your family. The first sign of risk they'll ditch you. Devs need to see work as what it is: it's a contract with mutual expectations.
And my recommendation is to self-reflect more on the health part, because we (including me) tend to rationalize that it's worth working more for the sake of building something or for the interesting research parts, or for learning experience or whatever we make up to justify it.
You can do that still with basic income. We just can't because society is fucked up, and research and development isn't paid enough to make a living and a healthy life. I also think that huge parts of the open source community that I identify myself with on a moral level are pretty hypocritical, considering that only the top notch famous "leaders" make enough to have a good balanced life. The 99%+ majority doesn't make enough to even rent a flat, and that's the absurd part of our society. I still can't fathom how the richest companies have money laying around on their bank accounts, and were built on the shoulders of unpaid open source contributors that got nothing in return.
That is something I really don't understand because it's honestly really messed up if you think about it.
I'm addicted to coding, not work. I know this because I've tried other jobs, even in late career, and they sucked in comparison and brought me no joy.
"The company" is mine, so yeah, probably it doesn't care about me, but it's definitely not dropping me without consent. :)
48, soon 49, heart still going. Not even sure what that's referring to. I don't feel stressed in these crunches, I feel excited! I build cool shit! They pay me to build cool shit! They pay me way too much to build cool shit!
I guess I don't know what I would balance that excitement with. I have cool hobbies too, and those have their place, but... I just don't resonate with your take and view on the industry.
IF someone hated coding, creating, or the tech industry itself -- then I could squint and get behind your balance suggestions. For them. :)
Thanks for sharing the perspective though. If nothing else, you're fanning my gratitude flame.
I also am building up my own company for the last couple years. But I want it to be a sustainable company that promotes a healthy lifestyle and that doesn't overwork its employees, and one that doesn't aim for 2 years turnover/rehiring of staff...because I think these are the typical effects of a toxic work environment, and reflects the values I don't agree with, both on a personal and a professional level.
In the end we both have a different leading style, I guess!?
Even if they told me themselves I wouldn't believe them.
Even if they were beaming with joy as the black circles in their eyes squinted while swearing there is no better joy than that of labour I would do what one of my ex-employers did and turn off machinery at 17.00 sharp.
There is no happiness in working long hours; there must be no happiness. Happiness must be verboten for anyone who thinks that undermining organized labour gains is "loving what they do".
Also, not satire... although written with a healthy dose of token-in-cheek :]
...right?
Add another 2 minutes and you could have the list of keywords to filter as a configurable url parameter, so you can amend it easily when the next technology you want to hate comes along.
/s
seriously tho.. yeah I'm a skeptical bastard in every aspect of my life these days. its exhausting.
The game is to balance the usage bars. If you run out of one before the others, you aren't max-min'ing it right.