If you’re building solo, your two silent killers are usually:
1) runaway AI token spend
2) death-by-feed scrolling between coding blocks
I learned this the expensive way.
A few weeks ago, I had nights where I shipped almost nothing, but somehow burned through a painful amount of AI usage. At the same time, "quick breaks" on social feeds were turning into 40-minute black holes.
Here are the 9 habits that fixed both.
1) Start each coding block with a spend cap
Before opening Claude/Cursor/Codex, I decide the max I’m willing to burn for that block.
Not forever. Just this block.
That single constraint kills the "one more prompt" spiral.
2) Split requests before they hit the model
My worst spend nights happened when I dumped giant context blobs into one mega-prompt.
Now I break work into small tasks:
- understand file
- propose plan
- patch one part
- verify
Smaller prompts = less waste + cleaner output.
3) Use different tools for different depths
I stopped treating tools like a religion.
- fast implementation bursts: Cursor/Codex
- deeper reasoning + architecture: Claude
When I match the tool to the job, quality goes up and retries go down.
4) Put token cost in your face
I built TokenBar because I needed an always-visible reminder while coding.
When cost is hidden, you overuse.
When cost is visible, you naturally tighten prompts.
Real-time awareness changed my behavior more than any spreadsheet ever did.
5) Add a hard stop condition to every long run
Late-night agents can drift forever if you don’t define exit criteria.
Now every long task has:
- done definition
- max retries
- stop condition
No more waking up to surprise spend.
6) Remove feeds during deep work windows
I built Monk Mode after realizing I wasn’t distracted by notifications — I was distracted by feeds themselves.
If a feed exists, my brain negotiates with itself.
If the feed is blocked, there’s nothing to negotiate.
That removed an absurd amount of context switching.
7) Replace "quick checks" with intentional breaks
I still take breaks. I just don’t open feed apps during them.
Instead:
- walk
- stretch
- water
- 3-minute reset timer
You return fresher without burning 25 minutes in algorithmic sludge.
8) End the day with a 3-line log
I write:
- what shipped
- what spend looked like
- what distracted me
Patterns appear fast. Usually the same 1-2 mistakes repeat.
9) Treat attention like budget
Most devs budget money but not attention.
For solo builders, attention is the more fragile resource.
The best combo I found:
- visible cost tracking (TokenBar)
- feed-level blocking during work (Monk Mode)
One protects money. The other protects focus.
Both protect momentum.
If you’re feeling stuck in "busy but not shipping," start with just two changes tonight:
1) set a per-block AI spend cap
2) block feeds for one 90-minute work sprint
You’ll feel the difference immediately.
Top comments (0)