Said A.

Senior engineer, 10+ yrs. Cursor & Lovable get to 70%. The last 30% — auth, payments, deploys, the wiring AI skips — is where it breaks. I made an agent ↓

[Said A. @said_akh]

@icanvardar They are approaching the threshold where they can advance AI while claiming concern for safety 🤫

[Said A. @said_akh]

@vasuman Make sense! Thanks for posting!

[Said A. @said_akh]

@gregisenberg I would suggest that dashboards “on the fly” are still alive. Yes, agents represent the new method to “dashboard,” yet a user interface is still required when people wish to interact.

[Said A. @said_akh]

If a teammate passes and the ball touches an opponent before reaching the offside player, that touch is not a "deliberate play" by the opponent (it's a deflection/rebound off a player attempting to intercept). So the offside player does not become onside — the offside offense stands, and the goal is disallowed.

The offside player would only be let off if the opponent had “deliberately played” the ball (a controlled action), which a simple deflection is not.

[Said A. @said_akh]

@vercel Does this support long-running execution on the backend?

[Said A. @said_akh]

Why are people still using Instagram or X if there is something like Pixelfed? No AI, no tracking, no ads 🤔

[Said A. @said_akh]

Waiting for relatives for 2 hours already

[Said A. @said_akh]

This is how people in the USA meet at Dulles International Airport.

[Said A. @said_akh]

@JamieBrock19060 @ClaudeDevs 🤣 please humans help me

[Said A. @said_akh]

@steipete @jxnlco Is the loop /goal or something else?

[Said A. @said_akh]

@oleg008 I recently noticed that the LLM is angry with me 😅

[Said A. @said_akh]

In the AI era, the bottleneck isn’t answers—it’s knowing which questions to ask.

And you can’t ask a great question about something you’ve never lived. Human experience is what tells you where to point the machine.

Specialization won’t die. It’ll become the moat.

[Said A. @said_akh]

@DanielSmidstrup It is easier because the drive is different 🚙

[Said A. @said_akh]

It is not that ideas are inherently good or bad, but rather whether they are true or not. Concepts of good and bad can be personal preferences; an idea can be considered “bad” yet still be truthful, and vice versa. The real question is how to determine whether an idea is true. Perhaps this is the realm of epistemology. Supernatural criteria may also play a role—something that exists beyond space, time, and matter. Perhaps the very force that created this reality could reveal which ideas are true.

[Said A. @said_akh]

@KaiXCreator CSS

[Said A. @said_akh]

@icanvardar Haha, that is true. AI is not good at it at all—it keeps generating code instead of deleting it.

[Said A. @said_akh]

@pmitu I'm not sure about “the code doesn't really matter”. Developing anything valuable and comprehensive is still pretty cumbersome, even with AI assistance, despite my 10+ years of experience in software development.

[Said A. @said_akh]

Vibe-coding lesson: plan thoroughly before committing a single line.

Ask the LLM what it understands. Then ask again, differently. Make it restate the plan back to you.

The time you spend aligning upfront saves hours of untangling generated code later.

[Said A. @said_akh]

@karpathy Congrats @karpathy!