hey, i'm hashir.
i build software, design things, and occasionally turn random ideas into actual projects.
currently building a helpdesk while figuring out everything from product to distribution.
going to share what i'm building, and learning along the way
let's connect!
a useful code comment explains the weird decision.
"keep this field for older app versions" tells the next person why it can't disappear yet.
"this is a string" probably doesn't help much.
which app would you miss most if it disappeared tomorrow?
i'm curious what earns that spot: a daily habit, a painful job it handles, or something else.
an empty dashboard is a tough first impression.
try giving people:
- a sample to explore
- one clear first step
- a way to undo it
let them understand the product before asking them to configure it.
"would you use this?" is an easy question to say yes to.
"how did you deal with this last time?" gives you something to build around.
the spreadsheet, workaround, or person they already rely on is worth understanding.
before adding a retry button, ask what happens if the first attempt actually worked. a timeout can mean 'the reply got lost', not 'nothing happened'. sending the request twice shouldn't send the customer two emails.
a useful support tag: 'what was the customer trying to do?' 'billing' tells you which team owns it. 'needed an invoice before their finance deadline' tells you what the product needs to make easier.
a loading spinner answers 'is something happening?' it doesn't answer 'can i leave this page?' for a long import, 'you can close this tab. we'll email you when it's done' might be the most useful line on the screen.
who pays when an ai agent retries a failed task?
if i'm charged per attempt, your reliability problem becomes my bill.
feels like pricing is part of the product design here.
would you rather pay per action or per completed task?
a useful test for ai-generated code:
ask it to change one requirement.
does it make a small edit, or rebuild half the app?
the first version gets the screenshot. the next change tells you what you've actually built.
"we'll let the user decide"
sounds reasonable until every screen has 12 toggles.
sometimes a default is the actual design decision you were hired to make.
which app makes you do the most homework?
remember the study where ai made developers 19% slower?
metr's 2026 follow-up hit a problem: some devs didn't want to participate if it meant working without ai.
that complicates measurement. it doesn't prove a speedup.
would you take a week off ai to measure yours?
ai helped newer support agents more than the veterans.
a study of 5,172 agents found 15% more issues resolved per hour on average, with bigger gains for less experienced workers.
could onboarding be the more interesting ai use case?
https://t.co/kGIMsfmQEL