Quick experiment this week. Bite-size newsletter about one feature. I want to see if you like it more than the usual “long” read, so please let me know.
Last week I tried a new* (tbh, don’t think few weeks counts as new in today’s world) Claude Code feature. It’s called /goal. I love it.
What /goal actually does
/goal is a dead-simple way to force your AI to be persistent in one direction.
Normally you ask, it works, it stops, it waits for you. With a goal, you tell AI the desired output and it keeps going until that line is actually crossed.
After each round a small fast model checks “are we there yet?” If no, it sends your AI back to work. If yes, the goal clears itself.
/goal = /loop towards the result
And Claude isn’t the only one doing this. OpenAI shipped the same idea in Codex back in May.
The example all docs mention is /goal pass all tests - your agent runs the tests, fixes what breaks, reruns, repeats until they’re green.
We’re vibecoders though. Tests are for the weak. So here are two use cases that showed how cool /goal is to me.
The two problems /goal solved for me
1/ A tricky gesture puzzle where I was stuck.
I have one screen where a user can tap, tap-and-hold, tap-hold-and-drag, and scroll.
Four gestures. And they kept clashing: when I tried to scroll, it dragged; or I could drag perfectly but the scroll was not working.
Every time the AI fixed one conflict, something else broke. I was going in circles and it was kind of depressing.
So I set a goal: make sure all four gestures work properly. It cooked for 40 minutes (and only 180k tokens) - and it actually started working.
2/ A long, boring, obvious job.
I had a checklist of factors that a combination of keywords had to pass, and 50 different combinations to grind through.
Long, boring, deliberate work with a crystal-clear definition of “done.”
Textbook example for /goal. I just left it for the day and it solved it perfectly.
When to use /goal?
The scope is narrow. One problem, not “make the app better”
“Done” can be measured. There’s a list of criteria to check against.
No babysitting for hours. This is set-it-and-go, not a quick chat.
An Idea For This Week
Find one job that ticks all three boxes above and let /goal cook.
Some examples I think of could be:
Go through a document and back every claim with a source…
Take a spreadsheet and verify every number against the PDF…
Write 50 LinkedIn post ideas, each one teasing the next…
Pick something long and boring with an obvious finish line.
Set the finish line and get out of the way. Learn to trust your co-pilot*
Shipyard updates
The iOS app journey continues, and it’s a completely different world - everything is new to me. Funny enough, I’m enjoying even the boring admin parts.
But it also makes me think about the “new” trap. The feeling of “new” is seductive. Exploring is exciting. But growing inside one subject is boring.
Switching to a new thing/tool/app/idea is more attractive than doing the same thing over and over until you know all ins and outs.
But as /goal has demonstrated – sometimes it’s great just to have one defined direction and push to it until you get there. Then, you can always start a new session :)
Did the shorter format land better, or do you miss the long version? Reply on this e-mail, I read everything.