Pinned Tweet
Senior Full-stack Engineer ➡️ Indie Hacker.
Goal: $10k MRR SaaS and document the journey.
Focus: Personality Ascend + SaaS portfolio.
Sharing insights on: Next.js, React, TS, Node.js & DB Architecture.
Follow for the code, the fails & the $0 → $10k grind. 🚀
2
12
924
Is anyone else feeling a bit dissapointed by the Fable 5 model? I tried it for couple of different things, one of those was generating unique variations of the landing page. It generated almost the same variations that Opus 4.8 would generate.
27
Meta laying off 10% of their workforce while sitting on all-time high profits is wild. AI just provided the perfect cover for them to fire the people they overhired over the last few years. Layoffs should be an indicator that a business is struggling, not something the market blindly cheers on.
67
We finally have tools that let us build and ship features in hours instead of weeks. The problem is, developers are using AI as an excuse to avoid talking to customers. It’s easier to spend three days vibe-coding a massive new dashboard than to spend 20 minutes finding your first customer.
2
52
Codex offers much more for the same price as Claude, especially if you are using 20$ tier. I am hitting Claude limits so quickly while at the similar timeframe I only spent 30% of Codex limit.
2
59
Google Cloud incorrectly auto-suspending Railway's production account and taking down all their customers is exactly why so many startups refuse to use GCP.
It looks terrible for both companies. Railway’s entire control plane was nuked by an automated ban, and now a lot of their users are migrating away because they lost access to their own clients' sites. If your infrastructure provider can just turn off your business by accident, they are a massive risk.
1
37
There is a massive difference between moving fast and just accumulating tech debt at lightspeed. Shipping a feature in a day feels great until it silently breaks a core user workflow because you skipped basic validation to save an hour.
23
We are entering an era where writing code is the easiest part of the job, and maintaining it is the hardest. The developers who stand out won't be the ones with the fastest prompt engineering skills. They will be the ones who actually enforce architecture and refuse to merge unreviewed, vibe-coded PRs into production.
1
25
After a couple of years on Cursor, I just swapped my subscription for Codex. The current state of AI coding agents for heavy usage:
Cursor: The IDE and tab completion are still great, but the limits on top models are too low. You can burn a month's quota in one day, and their Composer model struggles with complex work.
Claude Code: Excellent output quality, but the rate limits are incredibly constraining.
Codex: Currently the best bang for the buck. The experience is solid, and the limits actually allow for sustained, complex engineering without constant interruptions.
1
3
119
There is a fine line between a "cracked" engineer and a developer who creates a maintenance nightmare. Putting all your business logic in DB triggers and custom-coding every basic utility might feel like high-level engineering, but it makes the system nearly impossible for anyone else to manage. Technical brilliance shouldn't come at the cost of team scalability.
43
LLMs don't magically write good or bad code on their own. They just amplify the discipline of the developer using them.
If you put in the energy to actually guide the agent and review the output, you can write the best code of your career. If you just blindly accept the first response, you will write the worst.
19
Thinking about trying Codex as the Claude is often lacking. Besides Claude I used Cursor for the last couple of years but I am really close to canceling that subscription. The limits on latest models in Cursor are really harsh and the only alternative is their Composer model which is not that great at coding.
1
67
Total tokens spent is the worst metric people have come up with so far. It doesnt say anything about your productivity as its so easy to burn through millions of tokens especially if you dont know what you are doing. Its like we are rewarding incompetence.
28
The people obsessed with shipping velocity right now are usually the ones who have never had to maintain a complex system at scale. Anyone can use AI to flood a repo with features in record time. The real skill is shipping something that doesn't immediately collapse under the weight of its own tech debt.
1
29
The bar for standing out as a Senior Engineer is actually getting lower. While the crowd is busy flooding repositories with unverified AI code and vibe coding their way into outages, the developer who actually understands the architecture and takes accountability for every line merged becomes the most valuable person in the room.
23
A massive wave of future engineering work is going to come from non-technical founders who vibe-coded an MVP, actually got paying customers, and then hit a wall where every new feature breaks the app.
They won't need a full rewrite, just a real engineer to untangle the mess and restructure the core so it can scale.
AI isn't going to replace software engineers. It's going to create endless demand for us to fix what it generated.
1
30
Everyone is chasing more lines of code with AI, but nobody is talking about how impossible it is to properly review all of that output. Because developers are getting overwhelmed, terrible code is slipping into production daily. If you generate it, you need to be accountable for it.
2
2
32
The biggest issue with AI coding tools right now is the complete lack of accountability.
Developers are opening PRs with generated code they haven't reviewed or tested on their end, expecting the reviewer to figure out if it actually works. If you aren't going to validate your own output, don't ask someone else to do it.
1
2
28
AI didn't replace software engineering. It just turned us all into senior reviewers managing a junior developer who works infinitely fast, but absolutely refuses to check if a component already exists before building a redundant one.
1
32
It is incredibly easy to be amazed by an AI agent if you don't know what a scalable codebase is supposed to look like.
The benchmarks always promise a revolution, but in reality, unguided AI just creates fragile systems. The premium on actual engineers who can review, guide, and architect a maintainable system has never been higher.
23
