Vipin Yadav

Engineer building at the intersection of AI × Product × Systems.

[Vipin Yadav @heyvipin_]

I keep finding great articles, tools, and ideas in AI and tech, but there’s just too much to keep up with.

I’m exploring a way to make this less overwhelming.

If you work in tech, I’d love to understand how you deal with this.

A quick survey: https://t.co/mmb7KMAxHy

[Vipin Yadav @heyvipin_]

@pmitu can only confirm the second one

[Vipin Yadav @heyvipin_]

@ThePrimeagen infinite?

[Vipin Yadav @heyvipin_]

@Henryf1w people 💀

[Vipin Yadav @heyvipin_]

@Tristanrhee3 this one hit a little too close to home

[Vipin Yadav @heyvipin_]

@emilkowalski oh i really needed this. thanks for sharing @emilkowalski

[Vipin Yadav @heyvipin_]

@HeyAliux why not

[Vipin Yadav @heyvipin_]

at this point we all know human code review is becoming a bottleneck with coding agents.

reviewing intent and evidence instead of every line makes sense to me. but the trust issue is still there.

especially with large enterprise apps involving sso, permissions, complex workflows and multiple services. how do we reliably verify all of that without a human in the loop?

[Vipin Yadav @heyvipin_]

@slopmath @guiseiz

[Vipin Yadav @heyvipin_]

@TTrimoreau Because $209 is where I draw the line

[Vipin Yadav @heyvipin_]

GitHub’s full write-up:

https://t.co/YOMIiuxypa

[Vipin Yadav @heyvipin_]

GitHub rewrote its Copilot runtime from TypeScript to 830K+ lines of Rust using AI agents

128 PRs. ~3.5 months. ~$120K in tokens

Lessons:
• Migrate in small, shippable slices
• Keep E2E tests independent so agents can’t weaken them
• Invest in repo exploration: agents read/searched 10x more than they edited

[Vipin Yadav @heyvipin_]

Saw something similar while building an agent harness at work. Adding a dedicated code exploration phase actually reduced overall execution time and cost, while improving output quality severalfold.

[Vipin Yadav @heyvipin_]

@thdxr 😂

[Vipin Yadav @heyvipin_]

@sandislonjsak Slop profiling now?💀

[Vipin Yadav @heyvipin_]

@lucas_builds Did you even ask?

[Vipin Yadav @heyvipin_]

@SahajGarg6 @WisprFlow Do you guys have any process in place to ensure/enforce this?

[Vipin Yadav @heyvipin_]

@elonmusk 💀

[Vipin Yadav @heyvipin_]

Interesting read.

With coding models improving so quickly, I wonder how much we should invest in building elaborate harnesses today.

Will most of this become obsolete with better models, or are there parts we’ll always need?

https://t.co/lcQvZLdsrf

[Vipin Yadav @heyvipin_]

Apple just showed Meta’s 816B parameter AI model running locally across four Mac Studios.

That’s pretty wild.

Not exactly an affordable setup, but I love where this is going

https://t.co/6oJjrlBnXL