If you missed last week’s demo of Vellum’s AI Assistant (seamless, secure alternative to OpenClaw) with Vellum CEO Akash Sharma, you can watch the recording here.
Every night you make a hiring decision: who will replace you at work tomorrow?
You get to choose between two candidates:
Person A: A creative, resourceful, grounded person who has exactly the same past experience, knowledge, skills, and context as you (and can access most of that and will channel it strategically)
Person B: A reactive, short-fused, stressed person who has exactly the same past experience, knowledge, skills, and context as you (and can access some of that and will use it haphazardly)
Which one would you choose?
The answer is obvious, but the choice isn’t. The choice is largely driven by how well you sleep that night.
Which one have you been choosing?
This is top of mind because I have had more (client-initiated) coaching conversations about improving sleep in the past 3 months than in the previous 3 years.
The culprit is Claude Code.
Or more precisely, bingeing Claude Code.
Across the board, I’m hearing twin fuels for this addictive habit: fun and fear.
I found myself in the same exhausted and excited pattern for the first ~6 weeks of 2026.
I had just resolved at the turn of the year to make quality sleep more consistent, only to find my new late-night habit destroying that intention.
10:30pm: I’ll just finish this one thread…10:57pm: hmm, it’s not quite working. Let me debug and ask Codex to review. And while I wait for that response, I’ll just kick off this other feature I’ve been wanting to build out…11:23pm ooo If I get the implementation plan set, I can get it started so it can work while I sleep. 12:14am Just a little bit more…
It wasn’t just late nights; it was early mornings too. I would jolt awake at 3:30 or 4am, mind racing with possibilities and excitement. After trying and failing to go back to sleep, I eventually got out of bed and often got 2-3 hours of immensely productive work done before my kids woke up.
The momentum built on itself. I saw tangible progress (building and shipping Team 360 to clients; building Flow State, using it everyday myself, releasing it to beta), I discovered increasingly effective ways to use Claude Code, so I just wanted to do more and saw that I could.
At the same time, fear crept in. Everybody else (or so it seemed) was doing the same thing.
I’ve seen this common pattern as people learn to vibe-code the apps of their dreams. It feels like this:
OMG, I have superpowers—I can fly!
This is so cool, look what I can do!
Oh sh*t, everybody else can fly too.
This is MY corner of the sky, I better use my superpowers as fast as I can to claim it! I won’t stop flying until I have staked my territory before they do.
There’s a common cultural belief at work here (“there’s not enough sky to go around”), and there are some individual-level habits as well.
What’s the problem with all of this?
In the short-term, it’s not necessarily problematic. The adrenaline and excitement can fuel us and contribute to serious progress.
But sleep debt compounds and shortly we end up hiring Person B to replace us in our jobs and our lives.
Gradually, the excitement seeps out and fear and frustration creep in. Our thinking becomes muddled and our relationships sour.
One of my clients, a cofounder and CTO, found himself increasingly exasperated with his team for not keeping up with him and for blocking key decisions. He came into our session tearing his hair out at their ineptness and quickly discovered that in his exhausted and somewhat manic state, he was communicating in ways that were reliably evoking their defensiveness.
His one commitment from that coaching session? Sleep 8 hours that night. He texted me to say that when he woke up, he could finally think clearly again. He apologized to his teammates and together they (much more quickly) arrived at a shared decision on how to move forward.
Another cofounder/CTO client came in last week in a similar state. He was already at Step 1, seeing the cost of the pattern: his baseline state of aggravation and mental strain (“AI brain fry”). He said he wanted to improve his sleep quality and quantity, and we headed off a couple common failure modes as we started to get concrete about how.
Me: When would you like to get in bed?
Client: 10:45 or 11.
Me: So when do you want to stop working?
Client: 10.
Me: So 10pm is your Bright Line, “I don’t work past 10 at night”?
Client: No, it’s 10:30. Yea. I’ll set an alarm for 10:25.
Me: Give yourself some buffer time, otherwise you’ll blow through those 5 minutes and just keep going. Maybe you say at 10 or 10:10, “I won’t start any new tasks and it’s time to bring to a stopping point what I’ve been working on.” At 10:15 you start wrapping up and writing down what you’ll pick up in the morning.
[The challenge is when we’re mid-task, it’s like trying to turn off Netflix in the middle of a show or movie: there’s an open loop that keeps us wanting its completion. “Just 5 more minutes” is like a child in front of a screen, and we let the child decide when to stop.]
Client: I’m struggling with all the context-switching too in working multiple threads in parallel. Any tips?
Me: There is a cost to that, though the benefits may be worth it. One small practice I’ve found helpful is writing (or dictating) very specific, very short notes-to-self in my Claude Code instances that tell my future self exactly where to pick up the thread. (Just hold down the spacebar in the terminal to dictate.)
From this short exchange, you may be picking up on a few basics that have helped me and a dozen of my clients move back in the direction of consistent, quality sleep. Here’s the background.
1. “Go to sleep earlier” is ambiguous. When is “earlier”? The ambiguity lets the temptation drive instead of the big picture. My goal with clients and myself isn’t hard rules about behavior; it’s clear priorities and conscious choices that support them. That may mean staying up later than usual one night, as long as purpose is in the driver’s seat of that choice. Bright Lines clarify where the boundaries are so we can make conscious choices instead of blindly succumbing to temptation.
2. Temptation arises because, like in the middle of a Netflix series, there’s an open loop that is begging us to be completed. And since vibe-coding often gets us so close to where we want to go…but not quite there, and we can work on multiple threads in parallel, we have lots of open loops tugging on us, calling for our attention. Acknowledging that lets us take the time we actually need to come to a close.
This feels oddly similar to when I’m at the playground with my kids. It’s rarely easy to leave, but suddenly announcing “Time to go” is a surefire recipe for pushback. The same thing happens internally. (What is vibe-coding if not a playground for adults?)
3. Knowing the toys will still be there waiting for us in the morning makes it a bit easier. One study showed that, compared to people who journaled about what they’d done that day, people who wrote a to-do list at night for the next day led to higher quality sleep—and falling asleep faster. And the more specific the list was, the better the sleep. When we trust that the open loops are parked somewhere we’ll find them, we can rest easy.
Or as David Allen says, “Your mind is for having ideas, not for holding them.”
ACTIONABLE TAKEAWAYS
These solutions are fairly simple, which is a relief because it makes them more achievable:
Bright Lines
Get unambiguous clarity about what you’re aiming for
Buffer Time
Give yourself some runway to wind down instead of coming to a grinding halt
Make a specific to-do list so you know what to pick up tomorrow
In your coding agent and/or your friendly agentic to-do list
There are other pieces to the puzzle: for example, I’ve found napping to be an indispensable part of my happiness and productivity equation, but I’ll save that for another post.
Jack “I slept 7+ hours last night” Cohen
Had any similar vibe-coding induced sleep challenges? Any helpful practices for dealing with it? Reply and let me know!