I've written code for 15 years. now i build saas, games and apps with AI, in public, without leaning on it. four are live:
- openstage: your github commits become a public progress page (mine is in the first reply, it updates itself)
- trained on: which AI tools train on what you type, with the contract clause quoted
- wheels off: a 3d skate game that runs in the browser
- the goal atlas: world cup history, 1930 to 2026
follow for what I ship each weekday, what breaks and how I fix it, and the real numbers, zeros included.
what's the last build-in-public post that made you try the product?
I moved a production Rails app from 7.1 to 8.1 in one day. One thing broke.
I didn't want to wait. 8.0's security support ended on Nov 7, and 18 of the 21 advisories in the Nov 4 release cover 7.1 with no 7.1 patch coming.
How I did it:
1. One version at a time: 7.2, then 8.0, then 8.1, each in its own commit.
2. I ran the full suite of 1,694 tests at every step and never moved on with a failure or a deprecation warning.
3. I took every new default except YJIT, which stayed off because my host gave each service 512 MB.
4. Before deploying, I built the production image and booted it against a throwaway database.
5. After deploying, I checked the health endpoint. It answers JSON on 8.1 and never did on 7.1, so I knew the new code was live. Then I compared a share image with the one from before, and it was the same 74,919 bytes.
The break came on 7.2. A total of 26 tests failed with "svgload_buffer: operation is blocked". Active Storage was loaded but I never used it, and since 7.2.4 it turns off libvips' untrusted loaders for the whole process, SVG included.
My share images are SVG turned into PNG, so they would have failed in production. I stopped loading Active Storage and allowed the SVG loader back, since every SVG is my own.
Have you upgraded your Rails app recently? What would you have done differently?
59
Thanks Ayush 🚀🚀🚀
This open stage thing is a damn cool project that I came across. I just claimed mine. You guys can check it out as well. Do tell me if my open stage looks good or not.
openstage.dev/YourAKShaw
1
2
52
You can now see how your work would look on Openstage without signing up. Type your GitHub username on the landing page and it builds your stage from public GitHub, with your projects and a year of commits on one page.
If your public year is quiet, the preview lists up to 6 of your public repos instead, and the count of private contributions when your GitHub profile shows it.
The latest builders to sign up did it after I sent them their own preview in a DM.
Reply with your GitHub username and I'll reply with a screenshot of your stage.
4
2
73
And yours is just waiting to be claimed
claimed my stage on openstage
it builds the page straight from public github: 11 projects, aer-1 up top, the standard's name on every description
openstage.dev/zambodotdev
Openstage now shows a project's MRR, verified with Stripe.
Adding MRR to Openstage meant deciding what counts as MRR, so I ran a $7 subscription through Stripe's sandbox six ways and read each one the way Stripe does:
1. Cancelled, still paid up until the period ends: $0
2. One-time 50% coupon: $7
3. 50% coupon forever: $3.50
4. 23% tax inside the price: $5.69
5. Invoice emailed, unpaid, still a draft: $7
6. Trial over, card failing, past due: $7
So a customer who cancelled but has paid to the end of the month counts for nothing, and one whose card just failed counts in full. A one-off refund changes nothing either, since MRR counts what a subscription was charging at the time.
Did your own dashboard count any of these differently?
2
1
144
A stranger signed up to Openstage with 393 GitHub projects. Their page gave every project a card and came out 64,388 px tall. Nothing like real users to show you scenarios you really weren't expecting.
Since this morning a builder pins up to 10 for cards and the chart. The other 383 go in a list to search and sort, and the same page is 6,016 px tall, about 10x shorter.
If you've started that many projects, how many would you want up front?
2
74
I checked 336 of my replies on X against the post each one answered.
31 of them sat under posts past 5,000 views, and those 31 brought 54% of all my reply views.
X's ranking code is public, so I read why. A reply from an account you don't follow gets dropped from your "For You" feed. Strangers see my reply only if they open the post above it, so the reply reaches what that post reaches.
Volume stopped helping too. From October 24 to November 4, I sent 153 replies for 11 new followers, one per 14. Before that, it was one per ten.
I have 53 followers, so this is a small sample. If you reply a lot, what share of your reply views comes from your ten best replies?
2
3
121



