R.Urena

Shipping SaaS products faster than most people plan them · Full Stack Engineer · Building in public from 🇩🇴

[R.Urena @Esbals]

OpenAI's computer use and seamless voice interaction just work wonders.

First time in a year that I'm thinking of switching from Claude x20 to OpenAI x20, or at least switching from only Claude x20 to both at x5. And the biggest mover isn't even the Astra model.

[R.Urena @Esbals]

@Roadstockz @Stefan_3D_AI I re-watched the video and the mismatch is that he's using Windows while I did my test on macOS. I think the whole difference is just Windows being a trash OS lately...

[R.Urena @Esbals]

Been running Claude Code agents overnight for months now.

Some mornings I wake up to clean PRs ready to merge. Other mornings it's pure chaos.🫠

Either way, it's better than staring at a terminal at 2 a.m.

Building something around this workflow. More soon.

[R.Urena @Esbals]

@trq212 Current mood: afraid that @trq212 will announce a feature that kills one of my SaaS products.

[R.Urena @Esbals]

Obviously this only works if your PRDs are solid and well thought out. And yeah, the elephant in the room, no human can realistically review this much code quickly or safely. This was a controlled test on an internal tool to validate the idea, not something running in production.

[R.Urena @Esbals]

I just shipped 163,000 lines of code and 6,400+ tests in a single night while I slept.

No team. No contractors. Just me and an autonomous workflow I built over a weekend.

Here's how 👇

[R.Urena @Esbals]

I was building a GTM tool for my other projects. What started as 40 internal features snowballed into 144 tasks across services, APIs, UI pages, and cron jobs.

Doing this manually? That's 2-3 months of full-time solo dev work.

I didn't have 2-3 months!! (Nobody got time fo'that!)

So I built an autonomous coding workflow.

It picks a pending task, reads the PRD, runs a pre-check agent, implements + writes tests, validates against acceptance criteria, and retries on failure. (And other custom steps for self-healing and stuff)

Then it moves to the next task. All night. No human in the loop!!

I hit play at 3:15 AM and went to sleep.

When I woke up 14 hours later:

→ 72 tasks completed
→ 163,643 lines of code
→ 6,400+ tests passing
→ 85% first-attempt success rate
→ 0 tasks failed
→ 458 source files
→ 84 test files

And the workflow was still running. 🔥🔥

[R.Urena @Esbals]

tl;dr I built an autonomous pipeline that shipped 163K lines of code overnight while I slept.

Every dev needs this. 🔥🔥

I'm cleaning it up, adding a GUI, and shipping it for free.

Stay tuned 🌙

[R.Urena @Esbals]

@JespernissenSEO Nice, saved!

[R.Urena @Esbals]

@ZenovayDX Exactly 🔥

[R.Urena @Esbals]

Spent today spamming Reddit for my proyects.

Results:

0 new users
15 visits
$0 revenue🥲
12 karma earned
3 posts deleted by mods🫣

Here's what actually worked 👇

[R.Urena @Esbals]

@LaunchStackHQ Totally agree, nowadays distribution is king.

[R.Urena @Esbals]

Code is a commodity now. Your grandma can ship a SaaS with a prompt.🤶

So what actually matters? Distribution.

Starting tomorrow I'm testing a new distribution strategy every day across my MVPs and sharing the raw numbers views, signups, conversions, revenue.

No fluff. No theory. Just data, the good, the bad, and the embarrassing.

Day 1 drops tomorrow. Let's gooo! 🔥

[R.Urena @Esbals]

Agree with the playbook, but don't you think that now that anyone can build and ship, competition will be brutal for those without their own distribution channel? More success stories, sure but failures will be 100x more too. We might just be watching survivorship bias play out in real time. 🤔

[R.Urena @Esbals]

@jonahseguin As I said before, this is pay-to-win. 😅

[R.Urena @Esbals]

Not only that, pay per PR review is just more proof this "game" is pay to win.

Better tools, better distribution, better reach… all behind a paywall. The gap between funded and bootstrapped keeps getting wider.

[R.Urena @Esbals]

@levelsio Distribution wins now more than ever. The dev who ships and has an audience beats the "better" dev every time.
Those of us without one? We're cooked😅 unless we start building it yesterday.

[R.Urena @Esbals]

What people think developing with AI looks like:

"build me a SaaS, no errors, ship fast"

How it actually works:

1. Understand what you're solving and how, decompose the problem, find the solution

2. Group tasks by relation and dependency

3. Write a `yourapp-manifesto.md` with all of the technical and non-technical details.

4. Have Claude break that into tasks, then validate each task against the manifesto with fresh Claude Code instances

5. Review the tasks yourself, each gets its own .md, idempotent, opinionated, and full DOD (Definition of done are your railguards do not forget them)

6. Then you let the agent code

Otherwise you're just vibecoding and praying for good output.