The Vibe Coder

Finally its here...get the complete vibecoding starter kit with templates and prompts! enjoy - repo in the comments

@Polymarket And yet their Gemini models are so bad at coding!

JUST IN: Google DeepMind says its AI solved 9 open Erdős problems — considered among the hardest unsolved questions in mathematics.

@Nicotama222 Make the best use of time you have!

お早う御座います。 2026年5月25日になりました。 今年は残り221日です。 今年は残り221日です。 今年は残り221日です。 今年は残り221日です。 今年は残り221日です。 2026年はあと5304時間しかありません。

@jjackyliang @DamiDina If iOS is bad, what is good?

i wish apple software was as good as its hardware ios in particular is so bad

@scottmelker Drinking with your best pals is a lot of fun!

I stopped drinking when I was 48. I feel better, want to live longer with my kids. Great. But I’m glad I did a lot of drinking when I was younger. It was a lot of fun.

@KobeissiLetter What ever he writes becomes obsolete the next day!

BREAKING: President Trump releases a statement on the Iran deal.

everyone wants agents to be autonomous nobody wants to define the permissions model fascinating

I blew up @kirodotdev ‘s $20 plan in just 4 hours! I just can’t thank Codex enough for their limits!

For AI-native solo apps, cost is a system design concern. Track tokens, retries, provider errors, and per-user spend from the start. Billing surprises are architecture feedback. AI cost is architecture because every retry has a bill.

The point of managing technical debt is not purity. It is sustained speed. The cleanest system is useless if it arrives after the learning window closes. The goal is not clean code theater. It is sustained product speed.

Andrej Karpathy joining @AnthropicAI

A browser task is not complete when the click happens. It is complete when the agent verifies the expected state: URL changed, row exists, toast appeared, file downloaded, or record updated. Completion requires evidence, not a click event.

I do not measure agent maturity by how much it can do alone. I measure it by how cleanly it stops when it should.

Turn repeated mistakes into checklists. Before migration: backup, lock risk, rollback, defaults, old app compatibility. Before auth change: caller, resource, action, tenant. Notes become valuable when they change behavior. Turn pain into checks.

Every Gemini model starts with massive hype and the hype dies in a couple of days! Why does it happen to a company like Google that has greatest of talent, owns infrastructure, has all the data from internet? Is it skill issue or leadership issue?

When building in public, explain your taste. Why did you remove a setting? Why make the default opinionated? Why refuse a common feature? Technical founders earn trust by showing judgment, not just velocity. Taste is earned when you explain what you refused to add.

"we added ai" ok but did the workflow get shorter or did you just add a text box with anxiety

The best AI coding loop is not prompt, paste, hope. It is prompt, run typecheck, paste the exact error, run tests, paste the failing assertion, inspect the diff. Let deterministic tools narrow the model's imagination. Feed exact errors.

Keep raw webhook payloads for a reasonable window. They are your replay system, audit trail, and debugging record. Parsed fields are useful. The original event is evidence. Raw events are the black box recorder for integration bugs.

@KKRiders Doesn’t matter now !

Crucial game: Bat first or bowl first? 🥶

@iruletheworldmo Every model gets hyped until we hit that production bug after deploying the app…

the next bunch of upcoming models are a huge step change in intelligence. anthropic are not the only company with a ‘mythos’.