KK

Documenting AI tools & coding in public. No hype — just what works, what breaks, and what I wish I knew on day 1.

[KK @KhaledBuildsAI]

Starting from zero. No coding background, no AI expertise — just documenting everything I learn about AI tools as I actually learn it.
No hype, no "10x your productivity" nonsense. Just what works, what doesn't, and what I wish someone had told me on day 1.
Following along? 🧵

[KK @KhaledBuildsAI]

Everyone does, that's basically the whole job, security is one of those areas where the mistakes are quiet until they're not. If you want a quick win: go check any old repos you've got right now for a committed .env, even ones you forgot about. Costs 5 minutes and it's usually where people get surprised.

[KK @KhaledBuildsAI]

Unpopular take: "prompt engineering" was never really about clever wording.

It was about being forced to think clearly before you ask for something - which most people skip when talking to humans too.

AI just made the cost of vague thinking visible immediately, instead of three meetings later.

[KK @KhaledBuildsAI]

@nishithX26 Depends what you're optimizing for — Python for readability and ecosystem breadth, but "better" changes fast once you're doing something concurrency-heavy or need raw performance. No single best language, just best-for-the-job.

[KK @KhaledBuildsAI]

@Abhijit_091 Working through a home network lab this weekend — testing monitoring setups and documenting what actually breaks along the way.

[KK @KhaledBuildsAI]

@TTrimoreau Onboarding almost always, not features — most drop-off is people not reaching the "aha, this is useful" moment fast enough, not the product lacking capability. Fix time-to-value before you add anything new.

[KK @KhaledBuildsAI]

@pmitu True of basically every category of startup though, vibe-coded or not — the ratio was never good. The real question is whether AI changes the cost of trying, not the odds of winning.

[KK @KhaledBuildsAI]

@Shra_va_ni Because "can write code" and "knows which code shouldn't exist yet" are different skills. AI executes whatever you specify — it doesn't know your system's history, your tradeoffs, or what broke last time you tried this. That context is still the job.

[KK @KhaledBuildsAI]

@trikcode Fair on average, but it's a discipline gap more than a tooling one — the same AI that writes messy code will write clean, documented code if you actually specify structure and constraints up front. Most people don't ask for that, so they don't get it.

[KK @KhaledBuildsAI]

@wtAngat Reply to bigger accounts in your niche before you post much yourself — nobody's watching your empty profile yet, but they will see you in threads that are already getting attention. Consistency there compounds faster than posting into silence.

[KK @KhaledBuildsAI]

@SakshiSugandhi CLI, most days — it's faster once the muscle memory kicks in and it fits better into scripting/automation flows. GUI is easier onboarding for people newer to it, but doesn't take long before CLI just gets out of the way more.

[KK @KhaledBuildsAI]

@Taniyatweets_ Claude if reasoning quality matters more to you than IDE integration. Cursor if you live inside your editor all day and want that tighter loop — different priorities, not really a wrong answer either way

[KK @KhaledBuildsAI]

@pradeepb1996 Mostly people already in your replies before you needed them, not cold outreach. If you haven't built any visibility before launch day, day one is way harder than it needs to be — this is basically the whole build-in-public argument.

[KK @KhaledBuildsAI]

@danielkleach Ignored in public, easily — at least that's real signal. Praise nobody outside the team sees doesn't tell you anything except that your coworkers are nice to you.

[KK @KhaledBuildsAI]

@simonnyyberg Probably try to break it immediately — push it past whatever it's designed for. That's usually the fastest way to find the actual ceiling versus whatever's on the announcement page.

[KK @KhaledBuildsAI]

The AI coding shift nobody explains well: 2023: autocomplete. Finish my line. 2024: chat. Answer my question. 2026: agent. Here's the goal, go do the steps. Each jump changes what "good at prompting" even means.

[KK @KhaledBuildsAI]

@HeyAliux That's the cleanest version of it. And the failure mode is when people skip the verify step because the output looks confident, AI doesn't hedge the way a junior dev would when they're unsure, so it's easy to miss where it's wrong.

[KK @KhaledBuildsAI]

@emirymk That's a sharper breakdown than mine, honestly. "Current" is the one people miss most — feeding a model stale context is almost worse than too little, since it actively points the wrong direction instead of just being unhelpful.

[KK @KhaledBuildsAI]

Most people think "context" means pasting more text into the prompt.

It doesn't. Context is: what the model needs to know, in the order it needs to know it, with the noise removed.

More text ≠ more context. Half the time it's the opposite — you're just drowning the one relevant detail in five irrelevant ones.

[KK @KhaledBuildsAI]

@aslambeg84 AI developer tool — narrower market but stickier once it's embedded in someone's daily workflow. General "AI agent" is a crowded, fast-moving category where differentiation is hard to hold onto for 5 years.

[KK @KhaledBuildsAI]

@Henryf1w Probably not "after" — more like AI stops being its own category and just quietly becomes infrastructure, the way electricity did. Nobody says "electricity-powered business" anymore either.