anshul

building | ai engineer

[anshul @luhsnaa]

@suni_code a decision review process for devs who ship fast with ai.
https://t.co/y8j8Q0zB8R

[anshul @luhsnaa]

@Shivam25mishra i think the capabilities are no longer the constraint. token limits are. im happy to switch to whichever side offers me more tokens/buck

[anshul @luhsnaa]

@Akasheth_ composer is defo not opus level yet but the fast version is a lot fun when solving small issues

[anshul @luhsnaa]

been using arbityr to build arbityr lol.

a tool that pressure-tests decisions,
pressure-testing its own decisions and future

that's either really cool or i've gone too deep

[anshul @luhsnaa]

i'm building @ArbityrLive to solve this

[anshul @luhsnaa]

most devs don't have a bad coding problem.

they have a bad thinking problem.

they write great code executing a decision they never actually questioned.

and then they document that code beautifully.

but no one sees the audit trail behind the decisions.

[anshul @luhsnaa]

@findingnimo_ @buildwithshyam i love yapping to codex

[anshul @luhsnaa]

@isha_singh06 npm but it is v dangerous now with all the security stuff recently imma defo switch

[anshul @luhsnaa]

@PrajwalTomar_ crazy. love how you integrate so many tools and create actual workflows and also make it public

[anshul @luhsnaa]

@kedar2sexy @hqmank yeah even i tried it during beta. i've heard it has gotten better tho. i kinda liked the spec-driven development it had

[anshul @luhsnaa]

what if we did decision reviews too? opened PRs for decisions/plans, get them properly reviewed and iterated on.

can decision reviews be more important than code reviews now that our ability to understand code is rapidly atrophying?

[anshul @luhsnaa]

if ai is already writing all the code why are we still doing only code reviews?
yes good devs spend time prompting properly but its still very abstract and changes every time.

[anshul @luhsnaa]

@beyondbhavna perplexity deep research is faster and more accurate compared to others imo.

gemini deep research feels sycophant imo and the response is v bloated with ai slop that feels like me trying to lengthen my essay in school

[anshul @luhsnaa]

@Anas_founder codex as long as @thsottiaux keeps the resets coming lol

[anshul @luhsnaa]

@1Umairshaikh composer 2.5 fast feels like having a conversation in real time. almost all other models feel noticeably slower after using it.

it feels illegal to want higher inference speeds now when i'd be happy to wait for hours for good quality output just months ago

[anshul @luhsnaa]

@brettcalhounn Arbityr is the antidote to outsourcing your thinking to AI.

https://t.co/ZGCzlpMe79

[anshul @luhsnaa]

@ionleu hey! im building a decision making system for devs who ship fast with AI.
https://t.co/ZGCzlpMe79

[anshul @luhsnaa]

Building this because I've felt this pain too many times.
We gotta stop waiting for newer better models every week and start owning up to our decisions.

Much more coming very soon!

Check out @ArbityrLive for updates

Oh and thanks @cartesia for the boatload of credits.

[Arbityr @ArbityrLive]

Most bugs aren't coding errors. They're decision errors.

Arbityr is a pre-commit pressure test. Before you write the first line, it finds the assumption your decision is standing on, and pushes on it.

If you can defend it, you build.