Building https://xitter.cf/t.co/EwpeTSy7Iy - Schedule, hours, and pay for families who share a nanny. PM for 10 yrs, mom for 2 yrs.
Salt Lake City, UT
Joined December 2025
- Tweets686
- Following611
- Followers61
- Likes5.8K
Pinned Tweet
I built an app! Our nanny share ran entirely thru group texts and hasty math, so I made @ournannyapp. It includes everything you need to successfully manage a nanny either solo or with a share - schedules, hours, vacation, and pay.
If you have a nanny, or ever wondered whether a share could cut your childcare costs: ournanny.app
Building it in public from here. Next up: getting more beta testers!
had a beta user reach out to me to say said "i love the app" and "ive been looking for something like this"
she found me organically via chatgpt!
i eagerly jumped on her 2 feature requests - both already shipped out
been riding the high of this all week 馃槏
I spent $300 on Reddit ads, got 287 website clicks, and got 1 (nonpaying) user 馃檲
more in 馃У
after all of these changes...
- my cpc and ctr on the ad was about the same - 0.84% / $1.08
- same number of landing page sessions
- but I saw a much worse conversion rate! my new landing page hero converted at 0.8% instead of 4.0%, it was a complete dud
- only got 1 app install from 125 ad clicks, and that install never fully completed onboarding 馃槶
what lesson did i learn?
ads are hard! there are so many variables to tune in the funnel and each hard won lesson costs a lot of money
for someone like me with almost no ad budget who is still essentially finding product market fit, it doesn't really make sense.
i'm pivoting for now into organic marketing. that will let me learn and make mistakes without burning any more stupid cash 馃檹
reddit ads went so badly this week that I'm drowning my sorrows in more product improvements... 馃槶
> multi-nanny support in a single pod
> sick days for both parents and nannies
> allow the parent to step in as the caregiver on the schedule
> roundup / tip your nanny when paying
tried this method during the work week and it was pretty productive
as an example, in a single day claude:
* fixed an address search / autocomplete bug
* fixed a chat photo bug reported by a user
* security hardening before App Store
* added ability to do a QA smoke test using web flows - should improve ability to find bugs
shoutout to all my fellow builders working a 9-5
trying out a new process where i prompt the agent before i head into work
"im about to head into work right now, pull any stuff from backlog you can execute on independently"
(agent can pull from a backlog md file in Obsidian that is kept updated dynamically as you work on the project and have ideas / tech debt builds up)
by lunchtime, you check in and some things have been accomplished or need your review!
mistake #3 - (and this one hurts to admit) i didn't qa the app onboarding personally. instead, I relied on claude, and of the precious 4 installs i got, 3/4 of the people hit major bugs and completely stalled out! brutal and completely avoidable (also wtf claude!)
what i really needed to do is make a first in class qa-ready dev app instead of relying on weird janky web qa flows
result - i spent $150, and got a SINGLE glorious user who made it thru onboarding, set up his nanny share on the app, and invited his wife / nanny / other family. i was elated! but 7d later, this guy is nowhere to be seen / not using the app at all. 馃様
if i'm honest w myself, it was a complete fail. ill post about how round 2 went shortly once it finishes...
mistake #2 - my landing page didn't convert well enough - the paid traffic hit ournanny.app and only 5 people tapped on the cta. this is 4% cvr when it should be more like 8%
also the hero ended up being more important than i realized b/c only 25% of visitors even scrolled past the hero (avg mobile scroll depth is 45% of screen height)
shoutout to all my fellow builders working a 9-5
trying out a new process where i prompt the agent before i head into work
"im about to head into work right now, pull any stuff from backlog you can execute on independently"
(agent can pull from a backlog md file in Obsidian that is kept updated dynamically as you work on the project and have ideas / tech debt builds up)
by lunchtime, you check in and some things have been accomplished or need your review!
My latest design process is -
1. voice record and ramble about what i want to build, why, how i see it taking shape - essentially a prd in verbal form. best done while walking.
2. ask claude to interview me about the new feature and ask questions BEFORE designing anything
3. get a written spec saved to my obsidian vault after some back and forth that records all decisions
4. ask claude to create an html based design file laying out all designs and changes. often there will be multiple design options to choose from. this is an essential step before building anything.
5. multiple design rounds - usually at least 3-5, before finalizing
6. build!
pretty simple, working for me so far, however...
note that my process is super simple and only involves 'vanilla' claude and some skills. it does not involve specialized agents, design tools, dev tools, or special automated build processes. everything typically happens within a single chat session. my biggest external tool is obsidian as a repository of all the documents associated w each build and as a way to accumulate context about my project.
notably i am not a designer or developer, so another thing that is conspicuously absent is the code itself - i never, ever, EVER read code. it would just a waste of my time and i feel it adds nothing.
i will say however, that i worry about this gap in my own knowledge and would like to optimize my build process. i think claude makes stupid mistakes way too often that i either have to catch in QA or result in embarrassing user-facing bugs. im sure i can hone this part of the process as i get more experience and plan to - very open to any suggestions on this front.
i forgot step 7 - live QA! i have noticed that the implementation often differs from the original html mocks. claude will sometimes do the design equivalent of AI slop in writing and insert all kinds of weird subcopy i didn't ask for, sometimes the actual design tokens in the app differ from what was implemented. there are also inevitably things that come out when you actually use your own product both from the functionality and design side. real user data will cause edge cases you couldn't anticipate. you cannot replicate this with automated testing imho - it takes time and often ill just voice record my feedback in a single pass as i go through everything on my phone, and often a new mini design html file needs to be created as a result of things i notice that don't work as well as i thought.
small design pro-tip - at the end of a design iteration sprint, after you feel burnt out on it and dont have the energy to make any more changes, ask -
"is there anything else you would change to make the design more elegant, simple, or accessible to users?"
there is a good chance there will be addtl changes you didnt prompt and claude didnt think of, it's a simple process thing that can help
i have not experimented much w subagents for "design reviews", or with synthetic users for usability testing but interested in trying both of those out in the future as well to improve this process