Sarthak Dalabehera
283 posts

Sarthak Dalabehera
@s7rthaks
staff engineer @rippling // ex: founding eng @slicebank // ex: founding eng @vedantu_learn
Bengaluru Katılım Eylül 2021
167 Takip Edilen298 Takipçiler

The highest quality, fastest software possible is no software at all.
I wrote about why complexity is the speed killer - the parachute behind the drag racer. Every unnecessary feature is drag you carry forever.
New chapter: isagi.in/chapters/go-fa…

English

Is it just me or is GPT-5.5 in Codex getting nerfed?
Threw a complex coding task at it for Speedtrackr & it choked + hit rate limits. Meanwhile, Gemini 3 Flash on Antigravity executed the whole thing smoothly. Flash is low-key winning the coding agent race. @sama @antigravity
English

So I built Cosine - watches a project board, farms tickets to Claude Code and Codex CLIs in git worktree sandboxes.
Open source: github.com/TypicalDefende…
English

I wrote about how the software engineering job is changing - because I've been living it for the last year 🧵
In 6-12 months, agents may be the exclusive reviewer for most of our code.
The job is changing from "how does the code get written and reviewed" to:
1. Should we build this?
2. Is this a solid plan I can hand to agents?
3. How do we prove agents built what we wanted?
Write up: isagi.in/chapters/the-f…

English

@_virgil19 @arpit_bhayani @waitin4agi_ @_virgil19 Look at the full thread, I've discussed some of it other than PR Review in detail!

English

@s7rthaks @arpit_bhayani @waitin4agi_ 75% gen rate is the write side. the bottleneck just moved to review. agents now ship faster than humans verify. PR-review time becomes the cycle gate, not write time.
English

New write-up alert: I wrote about The Harness Is the Product: isagi.in/chapters/the-h… 🧵
#harness #codex #claude

English

The engineer's job is changing. Less writing code, more: specifying intent, building verification, designing feedback loops, and supervising agents.
The companies that build the best harnesses will ship faster.
Full write-up: isagi.in/chapters/the-h…
English

@arpit_bhayani @waitin4agi_ Six principles we use to evaluate every technical bet:
1. Verification over generation
2. What agents can't see doesn't exist
3. Encode judgment into the harness
4. Behavior over code
5. Close the loop
6. Legacy code is a specification
English





