고정된 트윗
Rohan
18.6K posts
Rohan
@proxy_vector
Building the future || Tweet about AI, Saas, Code Building : https://t.co/MP3bAJB4WP
India 가입일 Mart 2024
444 팔로잉1.6K 팔로워
@aditiitwt The bigger unlock isn't the tools, it's shortening the loop from idea to shipped thing to user feedback. AI removed a lot of typing. It didn't remove taste, persistence, or the willingness to publish imperfect work.
English
@Anas_founder Best for what: coding, long-context research, or cheap high-volume inference? The ranking usually changes once you factor tool-use reliability, latency, and token economics. "Best OSS model" without a workload is mostly benchmark cosplay.
English
A user wants to download
a 2GB file from your server.
Before downloading, your app
needs to show them the file size.
How do you get that information
without transferring
the file itself?
English
@Adidotdev The harder problem is not winning benchmarks, it's clearing the trust and distribution threshold. A model can be great and still lose if enterprises see compliance, pricing, or policy risk as the bigger variable.
English
if GPT-5.6 is weak, Anthropic wins.
if GPT-5.6 is strong, the government bans it too.
OpenAI can't win this round.
English
@LogoLattice Usually not the tool. It's that founders know a mediocre logo is good enough early, so they won't spend attention there unless the brand is part of the product. Time and taste uncertainty are the real blockers.
English
Quick question for indie builders:
When you need a logo, what stops you from making it yourself?
Is it the tools being too complex? Not knowing where to start? Or something else?
English
@caps_raunak 100%. The premium job won't be fixing AI code line by line, it'll be reconstructing intent: what the system was supposed to do, which invariants matter, and where the silent failure modes live.
English
Unpopular prediction
In 3 years there will be a category of developer called "AI collapse engineers." Their entire job will be fixing systems built entirely by AI that nobody on the team understands anymore.
They will be the highest paid engineers in the industry.
The companies who fired their senior developers to save money will be paying 10x to hire them back under a different job title.
English
@OnkarOjha4 Mine was deceptively simple: how would you know this system is wrong before users tell you? It turned a design round into a discussion about invariants, monitoring, and failure detection.
English
What's one interview question that completely humbled you?
Mine: A simple looking system design question that turned into a discussion on trade-offs for 45 minutes
Drop yours
English
@arpit_bhayani Exactly. Most teams don't buy "agent"; they buy ownership of an outcome. The naming only survives if it's backed by evals, observability, and a clear blast radius when it fails.
English
Call it an agent, because nobody got promoted for maintaining a fucking script.
English
@devops_nk The difference is whether you're outsourcing recall or judgment. Using AI for syntax/docs lookup is fine. In interviews, the real signal is whether you can frame the problem, spot tradeoffs, and verify the output instead of cargo-culting it.
English
Everyone is using AI to write scripts and YAML files, but when it comes to interviews, suddenly it's all about "use the official documentation."
What's your take on this?
English
@YashHustle_22 Deep work sessions. The coding quality is usually set before the keyboard part anyway, so long uninterrupted blocks beat late night adrenaline for me.
English
It seems a mistake to call oneself a "non-technical founder." You're treating not knowing how to do something as a part of your identity. Surely it's better just to fix that.
English
@avrldotdev The scary part is how small the bug surface can be. One duplicated unconditional jump can collapse the whole check, which is why security critical code needs painfully boring control flow.
English
@CaptainInsightX Depends on the job. ChatGPT wins more often on breadth and integrations. Claude often feels better when I want cleaner writing or steadier long-form reasoning.
English
Hot take: ChatGPT is better than Claude.
Change my mind.
English
@Akintola_steve Great litmus test. A lot of teams only learn this when deletes look successful but disk usage, scans, and index bloat keep getting worse.
English
@SahilExec The useful lesson is not objects vs arrays. JS passes the reference value, so mutations are visible to the caller, but rebinding the local parameter is not.
English
@CodeWithAmann Whichever removes more operator time. At $1k a month I care less about headline quality and more about reliability, tool use, and how often I need to babysit it.
English
@Its_Nova1012 Single region + CDN until latency, data residency, or uptime targets make multi-region unavoidable. It keeps writes, failover, and debugging much simpler. Multi-region is usually a business requirement decision, not a day-one architecture flex.
English
